XLVII INTERNATIONAL SCIENTIFIC CONFERENCE ON INFORMATION,

COMMUNICATION AND ENERGY SYSTEMS AND TECHNOLOGIES

2012
PROCEEDINGS OF PAPERS

VOLUME 1

SOFIA, 2012
ICEST 2012 Proceedings of the XLVII International Scientific Conference on
Information, Communication and Energy Systems and Technologies
organized by the Faculty of Telecommunications, Technical University
of Sofia, June 28-30, 2012, Veliko Tarnovo, Bulgaria

Proceedings of Papers: Volume 1 of 2 volumes

Editor: Prof. Rumen Arnaudov, PhD
Published by: Faculty of Telecommunications
Printed by: Publishing Company, TU-Sofia

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any
means, electronic, or mechanical, including photocopying or any information storage and the
retrieval system now known or to be invented, without written permission from the Publisher.

ISBN: 978-619-167-002-4
 WELCOME TO ICEST 2012

Dear Colleagues,

First of all I would like to thank you for attending our conference, in this way you
contribute to the improvement of the scientific value of this conference and make it one of the
most popular conferences in Balkan region. At the same time this conference turns to be a tool
for spreading our scientific achievements with the help of colleagues from Japan, USA, United
Kingdom, France and many more. For second year in addition to the founders of this
conference: Faculty of Telecommunications, Technical University of Sofia, Technical Faculty,
University “St. Kl. Ohridski”, Bitola and Faculty of Electronics, University of Nish, we enjoy the
participation and hospitability of Faculty of Mathematics and Informatics, “St. Cyril and St.
Methodius” University of Veliko Tarnovo and on the behalf of all of us I would like to express
our deep gratitude for their support.

This year there are 169 contributions with 352 authors from Bulgaria, Macedonia,
Serbia, Canada, Portugal, Romania, Hungary, Finland and Germany to be presented at our
conference. The submissions are evaluated by 74 reviewers from Bosnia and Herzegovina,
Bulgaria, Macedonia, Romania, Serbia and USA.

Dear colleagues and guests, I wish you successful participation in the conference. I
believe that this conference will go forward in raising its standard and influence and it will be
helpful in the scientific development and maturity of our young colleagues. I believe that here
we will witness the birth of new ideas for joint projects and future scientific achievements,
because we all work in the same direction – the development of technical and technological
foundations of future economy.

Good luck to all participants!

Prof. Dr. Rumen Arnaudov

ICEST 2012 Conference General Chairman
 XLVII INTERNATIONAL SCIENTIFIC CONFERENCE ON INFORMATION,
COMMUNICATION AND ENERGY SYSTEMS AND TECHNOLOGIES

2012
Organized by:

Technical University of Sofia

Faculty of Telecommunications
Bulgaria

St. Cyril and St. Methodius University of Veliko Tarnovo

Faculty of Mathematics and Informatics
Bulgaria

University “St. Kliment Ohridski” of Bitola

Faculty of Technical Sciences
Macedonia

University of Niš
Faculty of Electronic Engineering
Serbia
 TECHNICAL PROGRAM COMMITTEE
General Chairman:
Arnaudov R. Technical University of Sofia, Bulgaria

Vice Chairmen:
Milovanović B. University of Niš, Serbia
Mitrovski C. University “St. Kliment Ohridski” Bitola, Macedonia

Members:
Acevski N. University “St. Kliment Ohridski” Bitola, Macedonia
Atanasov I. Technical University of Sofia, Bulgaria
Bekiarski Al. Technical University of Sofia, Bulgaria
Bock J. University of Ottawa, Canada
Boumbarov O. Technical University of Sofia, Bulgaria
Ceselkoska V. University “St. Kliment Ohridski” Bitola, Macedonia
Dimitrov K. Technical University of Sofia, Bulgaria
Dochev I. Technical University of Sofia, Bulgaria
Dončov N. University of Niš, Serbia
Iliev G. Technical University of Sofia, Bulgaria
Iliev I. Technical University of Sofia, Bulgaria
Janković D. University of Niš, Serbia
Janković N. University of Niš, Serbia
Jeftić M. University of Niš, Serbia
Jolevski I. University “St. Kliment Ohridski” Bitola, Macedonia
Jordanova L. Technical University of Sofia, Bulgaria
Kostov M. University “St. Kliment Ohridski” Bitola, Macedonia
Makal J. Technical University of Byalistok, Poland
Marković V. University of Niš, Serbia
Markovski A. University “St. Kliment Ohridski” Bitola, Macedonia
Mitrevski P. University “St. Kliment Ohridski” Bitola, Macedonia
Nakamatsu K. University of Hyogo, Japan
Nedelkovski I. University “St. Kliment Ohridski” Bitola, Macedonia
Nikolov T. Technical University of Sofia, Bulgaria
Nikolova B. Technical University of Sofia, Bulgaria
Nikolova Zl. Technical University of Sofia, Bulgaria
Pencheva E. Technical University of Sofia, Bulgaria
Perić Z. University of Niš, Serbia
Pleshkova Sn. Technical University of Sofia, Bulgaria
Popova A. Technical University of Sofia, Bulgaria
Poulkov Vl. Technical University of Sofia, Bulgaria
Radevska P. University “St. Kliment Ohridski” Bitola, Macedonia
Stanković R. University of Niš, Serbia
Stanković Z. University of Niš, Serbia
Stefanović M. University of Niš, Serbia
Stefanovski M. University “St. Kliment Ohridski” Bitola, Macedonia
Stojčev M. University of Niš, Serbia
Stojmenov L. University of Niš, Serbia
Tasić D. University of Niš, Serbia
Todorov G. St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria
Todorova M. St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria
Trpezanovski Lj. University “St. Kliment Ohridski” Bitola, Macedonia
Tsenov A. Technical University of Sofia, Bulgaria
Valtchev S. NOVA-University, Lisbon, Portugal
Zieleznik L. Brookes University Oxford, UK

ORGANIZING COMMITTEE
Chairman:
Arnaudov R. Technical University of Sofia, Bulgaria

International coordinators:
Todorov G. St. Cyril and St. Methodius University of Veliko Tarnovo
Milovanović B. University of Niš, Serbia
Mitrovski C. University “St. Kl. Ohridski", Bitola, Macedonia

Local coordinators:
Todorova M. St. Cyril and St. Methodius University of Veliko Tarnovo
Poulkov Vl. Technical University of Sofia, Bulgaria

Members of the organizing committee:

Iliev G. Technical University of Sofia, Bulgaria
Dochev I. Technical University of Sofia, Bulgaria
Goleva R. Technical University of Sofia, Bulgaria
Dimitrov K. Technical University of Sofia, Bulgaria
Koleva P. Technical University of Sofia, Bulgaria
Tsankova J. Technical University of Sofia, Bulgaria
Stanković Z. University of Niš, Serbia
Milijić M. University of Niš, Serbia
Dimitrijević T. University of Niš, Serbia
Pargovski J. University “St. Kl. Ohridski", Bitola, Macedonia
Petkovski M. University “St. Kl. Ohridski", Bitola, Macedonia

CONFERENCE SECRETARIAT
Tsankova J. Conference Coordinator
Dochev I. Technical Secretariat
Members:
Nikolova M. Technical University of Sofia, Bulgaria
Stoyanova K. Technical University of Sofia, Bulgaria
Kirilova A. Technical University of Sofia, Bulgaria
Ivanova M. Technical University of Sofia, Bulgaria
Popova K. Technical University of Sofia, Bulgaria
Markova G. St. Cyril and St. Methodius University of Veliko Tarnovo
Kalushkov T. St. Cyril and St. Methodius University of Veliko Tarnovo
Markov A. St. Cyril and St. Methodius University of Veliko Tarnovo
Address:
Technical University of Sofia
Faculty of Telecommunications
Kl. Ohridski Blvd. 8, 1000, Sofia, Bulgaria
Phone/Fax: (+359 2) 965 30 95
E-mail: fktt-dekan@tu-sofia.bg
icest@tu-sofia.bg

CONFERENCE INTERNET SITE

For further information, please visit the Conference Internet Site: http://www.icestconf.org

BRIEF ICEST HISTORY

The ICEST Conference is successor of a series of conferences started in 1963 at Technical
University of Sofia under the name “Day of the Radio”. In 1977 the name of the Conference
was changed into “Communication, Electronic and Computer Systems”. Since 2000 has
become an international conference organized jointly by The Faculty of Telecommunications,
Sofia and the Faculty of Technical Sciences, Bitola.

In 2001 the conference was renamed EIST (Energy and Information Systems and
Technologies). In 2002 the Faculty of Electronic Engineering, Niš, Serbia became a co-
organizer of the conference and the conference changed its name into ICEST (International
Scientific Conference on Information, Communication and Energy Systems and
Technologies).

This year the conference is organized by the Faculty of Telecommunications at the Technical
University of Sofia and Faculty of Mathematics and Informatics of St. Cyril and St. Methodius
University of Veliko Tarnovo.

CONFERENCE TOPICS

[1] Radio Communications, Microwaves, Antennas

[2] Telecommunication Systems and Technology
[3] Signal Processing
[4] Digital Image Processing
[5] Computer Systems and Internet Technologies
[6] Informatics and Computer Science
[7] Electronics
[8] Energy Systems and Efficiency
[9] Control Systems
[10] Measurement Science and Technology
[11] Remote Ecological Monitoring
[12] Engineering Education
 TABLE OF CONTENTS
VOLUME 1

RADIO COMMUNICATIONS, MICROWAVES, ANTENNAS – PART 1

Multiuser IR-UWB System Performance .............................................................................1
Razvan Craciunescu, Simona Halunga, Octavian Fratu
University Politehnica of Bucharest, Romania
Doppler Fading Effects on OFDM Transmissions..............................................................5
Ioana Bucsa, Razvan Craciunescu, Simona Halunga, Octavian Fratu
University Politehnica of Bucharest, Romania
Investigation of the modulation type’s influence on the DVB-T signals quality ..............9
Oleg Panagiev
Technical University of Sofia, Bulgaria
Improving the reception of class DVB-T receivers ..........................................................13
Oleg Panagiev
Technical University of Sofia, Bulgaria
Monolithic Integrated Antennas with High Radiation Efficiency.....................................17
Hristomir Yordanov
Technical University of Sofia, Bulgaria
SCP-RPSC – The New Technology for Microwave Broadband Mobile
Communications ................................................................................................................21
Veselin Demirev
Technical University of Sofia, Bulgaria
Study on Hybrid FSO/RF Systems Availability Depending on the Meteorological
Conditions ..........................................................................................................................25
Tsvetan Mitsev, Maxim Shupak, Boncho Bonev
Technical University of Sofia, Bulgaria
Cylindrical Mesh TLM Model of Probe-Coupled Cavity Loaded with Planparallel
Dielectric Layers.................................................................................................................29
Tijana Dimitrijevic, Jugoslav Jokovic, Bratislav Milovanović
University of Niš, Serbia

RADIO COMMUNICATIONS, MICROWAVES, ANTENNAS – PART 2

Implementation of pseudo random noise generator in FPGA for Free
Space Optics BER testing..................................................................................................33
Nikolay Kolev, Tsvetan Mitsev
Technical University of Sofia, Bulgaria
Experimental Estimation and Correction of the Methods for Radio
Waves Attenuation Prediction in Rain ..............................................................................37
Boncho Bonev, Kliment Angelov, Emil Altimirski
Technical University of Sofia, Bulgaria
Multiresolution Analysis of Multiple Reflections in Transmission Lines .......................39
András Fehér, Ádám Békefi, Szilvia Nagy
Széchenyi István University, Hungary

i
Ad-Hoc Supported, Connection Fault-Tolerant Model for Mobile
Distributed Transaction Processing .................................................................................43
Tome Dimovski, Pece Mitrevski
University "St. Kliment Ohridski" of Bitola, Macedonia

Fast Synthesis of High Order Microwave Filters by Coupling Matrix Optimization.......47

Marin Nedelchev, Ilia Iliev
Technical University of Sofia, Bulgaria
Random High Voltage Impulses Modeling for EMC Testing............................................51
Kliment Angelov, Miroslav Gechev
Technical University of Sofia, Bulgaria
Optimum Divergence of the Transmitter Optical Radiation in FSO Systems.................55
Tsvetan Mitsev, Nikolay Kolev, Hristo Ivanov, Kalin Dimitrov
Technical University of Sofia, Bulgaria
Body Bias Influence on Ring Oscillator Performance for
IR-UWB Pulse Generator in 0.18m CMOS technology ...................................................59
Jelena Radic, Alena Djugova, Laszlo Nagy, Mirjana Videnovic-Misic
University of Novi Sad, Serbia

TELECOMMUNICATION SYSTEMS AND TECHNOLOGY – PART 1

Comparative Performance Studies of Laboratory WPA IEEE 802.11b,g
Point-to-Point Links.....................................................................................................................63
José Pacheco de Carvalho, Cláudia Pacheco, Hugo Veiga, António Reis
University of Beira Interior, Portugal
Customer Satisfaction based Demand Analysis of Mobile Services..............................67
Aleksandar Tsenov
Technical University of Sofia, Bulgaria
Investigate common work of software phone systems in virtual environments
and real switching systems ...............................................................................................71
Borislav Necov, Krasen Bankov, Mario Georgiev
Technical University of Varna, Bulgaria
Analysis of current methods and technologies for encoding, distribution and
consumption of IPTV services...........................................................................................74
Jordan Kanev, Stanimir Sadinov
Technical University of Gabrovo, Bulgaria
Average SIR Comparison for SC Systems Using Different Decision Algorithms
in the Presence of Interference .........................................................................................77
Aleksandra Panajotović, Dragan Drača, Nikola Sekulović*
University of Niš, Serbia
*School of Higher Technical Professional Education, Serbia
Optimization of Traffic Distribution Coefficients in IP Radio-Relay Network
with Path Diversity .............................................................................................................81
Dragana Perić, Miroslav Perić*, Branislav Todorović**, Milan Šunjevarić**, Miroslav Popović**
IMTEL Komunikacije a.d., Serbia
*VLATACOM d.o.o., Serbia
**Institute for Computer Based Systems, Serbia
Optical Receiver Sensitivity Evaluation in Presence of Noise in
Digital Communication System .........................................................................................85
Krasen Angelov, Stanimir Sadinov, Nataliya Varbanova
Technical University of Gabrovo, Bulgaria

ii
 TELECOMMUNICATION SYSTEMS AND TECHNOLOGY – PART 2
New Teletraffic Loss System – Polya/G/n/0......................................................................89
Seferin Mirtchev, Rossitza Goleva, Georgi Balabanov, Velko Alexiev
Technical University of Sofia, Bulgaria
An Evaluation of an UMTS/WLAN Interworking Architecture using IEEE 802.21 ..........93
Alexandru Vulpe, Octavian Fratu
University Politehnica of Bucharest, Romania
Simulation of Rare Events in Teletraffic Systems with Single Queue ............................97
Elena Ivanova, Rostislav Raev, Dimitar Radev
University of Ruse "Angel Kanchev", Bulgaria
VoIP over a Cognitive Network with Limited Availability...............................................101
Yakim Mihov, Boris Tsankov
Technical University of Sofia, Bulgaria
BEP Performance of DE-QPSK and DE-OQPSK over composite fading
channels in the presence of imperfect signal extraction...............................................105
Milica Petković, Bojana Nikolić, Bata Vasić, Goran Đorđević
University of Niš, Serbia
Quality of Service (QoS) – main principles and managing tools ..................................109
Miroslav Slavov, Pencho Penchev
Technical University of Gabrovo, Bulgaria
Pitch perception in complex sound ................................................................................113
Marko Janković, Dejan Ćirić
University of Niš, Serbia

SIGNAL PROCESSING
Digital Bandpass IIR Filers with High Selectivity ...........................................................117
Peter Apostolov
College of Telecommunications and Posts, Bulgaria
Features of time-frequency analysis visualization of large
dynamic range signals.....................................................................................................121
Tihomir Trifonov, Ivan Simeonov*, Rosen Dzhakov**
St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria
*Vasil Levski National Military University, Bulgaria
**Vasil Levski National Military University - Shumen, Bulgaria
Accuracy Improvement of Allpass-based Digital Hilbert Transformers .......................125
Kamelia Nikolova, Georgi Stoyanov
Technical University of Sofia, Bulgaria
Acoustic Standing Waves in Closed Cylindrical Enclosures ........................................129
Ekaterinoslav Sirakov, Hristo Zhivomirov
Technical University of Varna, Bulgaria
Control of Radiation Directivity Applying Independent Element Dodecahedral
Loudspeaker.....................................................................................................................131
Marko Jelenković, Dejan Ćirić, Jelena Zdravković, Stefan Tomić
University of Niš, Serbia
Modulated bandpass Farrow Decimators and Interpolators .........................................135
Djordje Babic, Vesa Lehtinen*
Union University, Serbia
*Tampere University of Technology, Finland
iii
Simulation of Codec for Adaptive Linear Prediction .....................................................139
Rumen Mironov
Technical University of Sofia, Bulgaria

DIGITAL IMAGE PROCESSING

Image Compression with Inverse Pyramid Decomposition over Wavelet
Spectrum...........................................................................................................................143
Teodora Sechkova, Ivo Draganov
Technical University of Sofia, Bulgaria
Efficient Adaptive Local Binarization Algorithm for Text Extraction from
Image with Complex Background ...................................................................................147
Antoaneta Popova
Technical University of Sofia, Bulgaria
Text Skew Detection using Log-polar Transformation ..................................................151
Darko Brodić, Zoran Milivojević*, Dragan Milivojević**
University of Belgrade - Bor, Serbia
*Technical College Niš, Serbia
**Institute for Mining and Mettalurgy, Serbia
Directional Transforms Applicability in Image Coding ..................................................155
Ivo Draganov
Technical University of Sofia, Bulgaria

COMPUTER SYSTEMS AND INTERNET TECHNOLOGIES

An Algorithm and a program module for calculating the border height
of the mass centre of a vessel.........................................................................................159
Emiliya Koleva, Mariya Nikolova, Mariya Eremieva, Viktoriya Sabeva
Naval Academy "Nikola Vaptsarov", Bulgaria
Power consumption analysis of fault tolerant real-time systems .................................163
Sandra Djosic, Milun Jevtic, Milunka Damnjanovic
University of Niš, Serbia
Challenges of Personalization and Collaboration Learning Process by
Using Blogs ......................................................................................................................167
Teodora Bakardjieva, Boyka Gradinarova*
Varna Free University, Bulgaria
*Technical University of Varna, Bulgaria
Implementation of Web 2.0 in the Bitola Museum - Successful Marketing
Tool ...................................................................................................................................171
Pargovski Jove, Irena Ruzin*, Aleksandra Lozanovska**
Cultural Heritage Protection Office, Macedonia
*NI Institute and Museum Bitola, Macedonia
**Gauss Institute, Macedonia
Attacking the cloud ..........................................................................................................175
Vlad-Andrei Poenaru, George Suciu, Cristian-George Cernat, Gyorgy Todoran,
Traian-Lucian Militaru
University Politehnica of Bucharest, Romania
An Implementation of SMS Communication with Patients in
a Medical Information System .........................................................................................178
Ivica Marković, Aleksandar Milenković, Dragan Janković
University of Niš, Serbia
iv
A Comparative Analysis of Mobile AR Software with the Application to the
Archeological Site Medijana ............................................................................................182
Dušan Tatić, Časlav Stefanović, Dragan Stanković*
University of Niš, Serbia
*University of Pristina Kosovska Mitrovica, Serbia
A comparative analysis of dynamic programming languages for application
in multi-agent systems.....................................................................................................186
Ana Stankovic, Dragan Stanković*, Dušan Tatić**
Metropolitan University, Serbia
*University of Pristina Kosovska Mitrovica, Serbia
**University of Niš, Serbia

INFORMATICS AND COMPUTER SCIENCE – PART 1

GPU Accelerated Construction of Characters of Finite Abelian Groups......................190
Dušan Gajić, Radomir Stanković
University of Niš, Serbia
Modern Processor Architectures Overview....................................................................194
Danijela Jakimovska, Aristotel Tentov, Goran Jakimovski, Sashka Gjorgjievska, Maja Malenko
University Ss Cyril & Methodius in Skopje, Macedonia
The mechanism for flexible symbology in mobile GIS ..................................................198
Miloš Roganović, Bratislav Predić, Dragan Stojanović, Marko Kovačević
University of Niš, Serbia
GinisED tools for spatial analysis of electric power supply network ...........................202
Aleksandar Stanimirović, Leonid Stoimenov, Danilo Vulović
University of Niš, Serbia
Computer Methods and New Values for Cut Set Catalan Numbers ..............................206
Iuliana Dochkova-Todorova
St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria
Accelerating Strategies in Evolutionary Algorithms......................................................208
Vassil Guliashki, Leoneed Kirilov
Bulgarian Academy of Science, Bulgaria

INFORMATICS AND COMPUTER SCIENCE – PART 2

2D Weather product visualization using Marching Squares algorithm ........................212
Igor Antolović, Dejan Rančić, Vladan Mihajlović, Dragan Mihić*, Marija Đorđević*
University of Niš, Serbia
*Republic Hydrometeorological Service of Serbia, Serbia
Efficient Implementation of BDD Packages on the GPU Platform ................................216
Miloš Radmanović, Dušan Gajić
University of Niš, Serbia
Architecture of Distributed Multiplatform GIS for Meteorological Data Analysis
and Visualization ..............................................................................................................220
Marko Kovačević, Aleksandar Milosavljević, Vladan Mihajlović, Dejan Rančić
University of Niš, Serbia

ELECTRONICS
Tracking Analogue to Digital Converter Modelling using VHDL-AMS ..........................224

v
Marieta Kovacheva, Ivailo Pandiev
Technical University of Sofia, Bulgaria
Development of Parameterized Verilog-AMS Model of Photovoltaic Cells ..................228
Elissaveta Gadjeva, Georgi Valkov
Technical University of Sofia, Bulgaria
Optical Control through Stencils Cutting in Surface Mount Technology.....................232
Valentin Videkov, Aleksei Stratev, Georgi Furkov
Technical University of Sofia, Bulgaria

ENERGY SYSTEMS AND EFFICIENCY

Power Quality According to EN 50160 ............................................................................235
Nikolce Acevski, Kire Mijoski, Tomce Mijoski
University "St. Kliment Ohridski" of Bitola, Macedonia
Using H synthesis for finding settings of single channel power system
stabilizers of synchronous generators...........................................................................239
Konstantin Gerasimov, Petko Petkov*, Krum Gerasimov
Technical University of Varna, Bulgaria
*Technical University of Sofia, Bulgaria
Technical Conditions for PV Plants Connection on the MV Distribution Grids
in the Republic of Macedonia ..........................................................................................243
Ljupco Trpezanovski, Metodija Atanasovski, Dimitar Dimitrov*
University "St. Kliment Ohridski" of Bitola, Macedonia
*University Ss Cyril & Methodius in Skopje, Macedonia

CONTROL SYSTEMS
Control cards. Control cards and control points as part of the manufacturing
process .............................................................................................................................247
Violina Georgieva, Alexander Hadjidimitrov*
Technical University of Sofia, Bulgaria
*Team VISION Bulgaria Ltd., Bulgaria
Computer Simulation and Analysis of Two-Coordinate Position Electric Drive
Systems.............................................................................................................................251
Mikho Mikhov, Marin Zhilevski
Technical University of Sofia, Bulgaria

MEASUREMENT SCIENCE AND TECHNOLOGY

Automated multichannel broadband spectrum analysis of fiber-optic grating
sensors .............................................................................................................................255
Plamen Balzhiev, Wojtek Bock*, Tinko Eftimov**, Rumen Arnaudov
Technical University of Sofia, Bulgaria
*Université du Québec en Outaouais, Canada
**University of Plovdiv, Bulgaria
Measurement of the Position by Using Hybrid Pseudorandom Encoder.....................259
Dragan Denić, Goran Miljković, Jelena Lukić, Miodrag Arsić, Milan Simić
University of Niš, Serbia
Method for calculating the stability at moderate and big heeling angles of
a vessel .............................................................................................................................263

vi
Mariya Eremieva, Viktoriya Sabeva, Mariya Nikolova, Emiliya Koleva
Naval Academy "Nikola Vaptsarov", Bulgaria
Design of a high – sensitive capacitive sensor for wireless monitoring of bulk
material’s level..................................................................................................................265
Teodora Trifonova, Valentina Markova, Valentin Todorov, Ventseslav Draganov
Technical University of Varna, Bulgaria
Different Implementations of Serial Pseudorandom/Natural Code Converters ...........269
Dragan Denić, Goran Miljković, Jelena Lukić, Miodrag Arsić, Dragan Živanović
University of Niš, Serbia

REMOTE ECOLOGICAL MONITORING

Cloud systems for environmental telemetry – A case study for ecological
monitoring in agriculture .................................................................................................273
George Suciu, Octavian Fratu, Cristian Cernat, Traian Militaru, Gyorgy Todoran, Vlad Poenaru
University Politehnica of Bucharest, Romania

ENGINEERING EDUCATION
High-quality Primary School Education in the Field of Electrotechnics and
nformatics - Beginning of the Development of Successful Engineers.........................277
Sonja Cvetkovic, Zoran Stankovic*
Primary school "Cele Kula", Serbia
*University of Niš, Serbia

VOLUME 2

RADIO COMMUNICATIONS, MICROWAVES, ANTENNAS (POSTER)

Sensor Network Topology as Low-interference Factor .................................................283
Vasil Dimitrov, Rozalina Dimova, Paskal Novakov
Technical University of Varna, Bulgaria
A Proposal for Harmonic Rejection Mixer Avoiding Irrational Weighting Ratios ........287
Ludwig Lubich
Technical University of Sofia, Bulgaria
Efficient Estimation of the Antenna Noise Level Using Neural Networks ....................291
Ivan Milovanovic, Zoran Stankovic*, Marija Milijic*
Singidunum University, Serbia
*University of Niš, Serbia
Software for Automated Measuring Pattern Diagrams of Wide Frequency Bands
Antennas with Integrated Receivers ...............................................................................295
Dragan Obradović, Igor Stančić, Aleksandar Kopta, Zoran Mićić, Predrag Manojlović
IMTEL Komunikacije a.d., Serbia
Numerical Model of Enclosure with Receiving Dipole Antenna for Shielding
Effectiveness Calculation ................................................................................................299
Tatjana Cvetković, Vesna Milutinović, Nebojša Dončov*, Bratislav Milovanović*
Republic Agency for Electronic Communications, Serbia
*University of Niš, Serbia
vii
Reliability of Radio-Relay Systems .................................................................................303
Nataša Bogdanović, Dejan Blagojević, Dragiša Milovanović*
School of Higher Technical Professional Education, Serbia
*University of Niš, Serbia
New Architectural Solutions to Improve the CATV System Performances ..................307
Lidia Jordanova, Dobri Dobrev, Kalin Dimitrov
Technical University of Sofia, Bulgaria
Measurement Site and Procedures for Experimental 2D DOA Estimation ...................311
Marija Agatonovic, Zoran Stankovic, Bratislav Milovanović, Nebojša Dončov
University of Niš, Serbia
Simulation influence of the thermal noise of PIN photodetector on
performance DWDM optical network ..............................................................................315
Petar Spalević, Dejan Milić*, Branimir Jakšić, Mile Petrović, Ilija Temelkovski*
University of Pristina Kosovska Mitrovica, Serbia
*University of Niš, Serbia
Study of ICI in PRS-OFDM systems ................................................................................319
Stanio Kolev, Ilia Iliev, Stoicho Manev
Technical University of Sofia, Bulgaria
BER simulation analysis of PRS-OFDM systems with MLSD........................................321
Ilia Iliev, Stanio Kolev, Stoicho Manev
Technical University of Sofia, Bulgaria

TELECOMMUNICATION SYSTEMS AND TECHNOLOGY (POSTER)

A New Modified Algorithm for Multi Constraint Routing ...............................................323
Yavor Tomov, Georgi Iliev
Technical University of Sofia, Bulgaria
A Schema Based Approach to Access Network Discovery and Selection in EPS ......327
Ivaylo Atanasov
Technical University of Sofia, Bulgaria
System for thermal comfort monitoring in working and living environment ...............331
Uros Pesovic, Dusan Markovic, Zeljko Jovanovic, Sinisa Randjic
University of Kragujevac, Serbia
Some Integral Characteristics of MRC Receiver in Nakagami-m fading
Environment .....................................................................................................................335
Hana Stefanovic, Dejan Milić*, Dimitrije Stefanovic*, Srdjan Milosavljevic**
College of Electrical Engineering and CSAS, Serbia
*University of Niš, Serbia
**University of Pristina Kosovska Mitrovica, Serbia
Adaptive Filtering Algorithms Suitable for Real-Time Systems....................................339
Maria Nenova
Technical University of Sofia, Bulgaria
Presentation of a model to study facsimile coded signals from a fourth group..........343
Todorka Georgieva
Technical University of Varna, Bulgaria
Design and implementation of a device for a cloudiness measurement......................347
Cvetan Kitov, Bulgaria
Classification and comparative analysis of localization approaches for
Wireless Sensor Networks...............................................................................................351
Ivanka Tsvetkova, Plamen Zahariev, Georgi Hristov, Mihail Iliev
viii
University of Ruse "Angel Kanchev", Bulgaria
Comparative analysis of routing approaches for wireless sensor networks...............355
Plamen Zahariev, Georgi Hristov, Ivanka Tsvetkova, Mihail Iliev
University of Ruse "Angel Kanchev", Bulgaria
Upstream Power Control for Digital Subscriber Lines Based on Role Game
Approach ..........................................................................................................................359
Pavlina Koleva, Oleg Asenov*, Vladimir Poulkov
Technical University of Sofia, Bulgaria
*St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria

SIGNAL PROCESSING (POSTER)

A possibility for Edge detection using LabVIEW graphical programming
environment......................................................................................................................363
Liljana Docheva
Technical University of Sofia, Bulgaria
Implementation of Analog Neural Networks with Labview............................................367
Liljana Docheva, Alexander Bekjarski
Technical University of Sofia, Bulgaria
Noise-Resistance Performance Estimation of a Chaos Shift Keying Signals..............371
Galina Cherneva, Georgi Pavlov, Elena Dimkina
Todor Kableshkov University of Transport, Bulgaria
On a combination of amplitude and frequency modulation used for processing
speech signals in cochlear implants...............................................................................375
Svetlin Antonov, Snejana Pleshkova-Bekiarska
Technical University of Sofia, Bulgaria
Safe Operating Area Limitations in Class B Amplifiers .................................................379
Hristo Zhivomirov
Technical University of Varna, Bulgaria
Complex Criterion for Linearity Segment Detection in the Subtraction
Procedure for Removing Power-line Interference from ECG ........................................383
Georgy Mihov
Technical University of Sofia, Bulgaria

DIGITAL IMAGE PROCESSING (POSTER)

School promoting tool Multimedia project – Documentary film ...................................387
Ilche Acevski, Valentina Acevska, Mimoza Jankulovska, Igor Nedelkovski
University "St. Kliment Ohridski" of Bitola, Macedonia
Animation of shadow –- Advantages and disadvantages when rendering
3D project..........................................................................................................................391
Valentina Acevska, Ilche Acevski, Igor Nedelkovski
University "St. Kliment Ohridski" of Bitola, Macedonia
Investigation of Mixture of Gaussians Method for Background Subtraction
in Traffic Surveillance ......................................................................................................395
Boris Nikolov, Nikolay Kostov, Slava Yordanova
Technical University of Varna, Bulgaria
Applied Aspects In Static Images Processing ...............................................................399
Gergana Markova
St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria

ix
 COMPUTER SYSTEMS AND INTERNET TECHNOLOGIES (POSTER)
Management of Software Project using Genetic Algorithm ..........................................403
Milena Karova, Nevena Avramova, Ivaylo Penev, Yulka Petkova
Technical University of Varna, Bulgaria
Railway Infrastructure Maintenance Efficiency Improvement by Using
Tablet PCs.........................................................................................................................407
Slobodan Mitrović, Svetlana Čičević, Slađana Janković, Norbert Pavlović, Slaviša Aćimović,
Snežana Mladenović, Sanjin Milinković
University of Belgrade, Serbia
Intelligent learning system for High education ..............................................................411
Aleksandar Kotevski, Gjorgi Mikarovski
University "St. Kliment Ohridski" of Bitola, Macedonia
Using Cloud Computing in e-learning.............................................................................415
Gjorgi Mikarovski, Aleksandar Kotevski
University "St. Kliment Ohridski" of Bitola, Macedonia
An Approach to Define Interfaces for Mobile Telemetry................................................418
Ivaylo Atanasov, Ventsislav Trifonov, Evelina Pencheva
Technical University of Sofia, Bulgaria
Architecture of Automated System Software for Testing Petrol Engines ....................422
Georgi Krastev
University of Ruse "Angel Kanchev", Bulgaria
Mazes - Classification, Algorithms for Finding an Exit..................................................425
Maya Todorova, Nedyalko Nikolov
Technical University of Varna, Bulgaria
Integration of Biometrics to the E-Health .......................................................................428
Milena Stefanova, Oleg Asenov
St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria
Psychology of the Perpetrators of Computer Criminal Acts and Review
of Legal and Economic Consequences for the Community ..........................................431
Zaklina Spalevic, Jelena Matijasevic, Dejan Rančić*
University Business Academy, Serbia
*University of Niš, Serbia
P2P Wireless Network Based on Open Source Linux Routers .....................................435
Hristofor Ivanov, Miroslav Galabov
St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria
An Approach to Optimization of the Links' Load in the MPLS Domain ........................439
Veneta Aleksieva
Technical University of Varna, Bulgaria
Performance Study of Virtualization Platforms for Virtual Networking Laboratory.....443
Hristo Valchanov
Technical University of Varna, Bulgaria
Integration of Video Conference into eLearning Platform Based on Moodle
for the Vocational School ................................................................................................447
Ilche Acevski, Valentina Acevska, Linda Fahlberg-Stojanovska
University "St. Kliment Ohridski" of Bitola, Macedonia
System for Multi-variant Multi-parametric WEB-based Test Control ............................451
Vladimir Karailiev, Raicho Ilarionov, Hristo Karailiev
Technical University of Gabrovo, Bulgaria

x
Methods for Assessing Information Sites ......................................................................455
Tihomir Stefanov
St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria

INFORMATICS AND COMPUTER SCIENCE (POSTER)

One Approach for Development of Software Modules Adding
New Geometric Primitives in 3D Graphics Applications ...............................................459
Emiliyan Petkov
St. Cyril and St. Methodius University of Veliko Tarnovo, Bulgaria
Analysis of Platform Dependencies in Software Solution for Auction and
Trading in Electric Energy Market...................................................................................463
Milos Gajic, Marko Djukovic, Sasa Devic, Branislav Atlagic, Zvonko Gorecan, Dragan Tomic
Telvent DMS LLC, Serbia
Rapid development of GUI Editor for Power grid CIM models ......................................467
Sasa Devic, Lajos Martinovic, Branislav Atlagic, Zvonko Gorecan, Dragan Tomic
Telvent DMS LLC, Serbia
Building an 8085 Microprocessor Module for the HADES Simulation Framework ......471
Goce Dokoski, Dimitar Bojchev, Aristotel Tentov
University Ss Cyril & Methodius in Skopje, Macedonia
Аlgorithms for scheduling of resource-constrained jobs .............................................475
Ivaylo Penev, Milena Karova
Technical University of Varna, Bulgaria
Hybrid Automatic Repeat Request (HARQ) Overview ...................................................479
Ginka Marinova, Slava Yordanova, Nikolay Kostov
Technical University of Varna, Bulgaria
Digital information transfer systems an overview .........................................................482
Ginka Marinova, Slava Yordanova, Nikolay Kostov, Boris Nikolov
Technical University of Varna, Bulgaria
Towards applicability of agile software development methodologies .........................485
Aleksandar Dimov, Stavros Stavru, Dessislava Petrova-Antonova, Iva Krasteva
University of Sofia "St. Kl. Ohridski", Bulgaria

ELECTRONICS (POSTER)
FPGA (Field Programmable Gate Arrays) – Based Systems-On-a-Programmable-Chip
(SOPC) Development for Educational Purposes............................................................489
Valentina Rankovska
Technical University of Gabrovo, Bulgaria
Electronic Simulator of Sound (Noise) Effects for Electric Vehicles in Urban
Areas .................................................................................................................................493
Georgi Pavlov, Galina Cherneva, Radoslav Katsov, Ivaylo Nenov, Ilko Tаrpov
Todor Kableshkov University of Transport, Bulgaria
Spray Deposition of PVDF Layers with Application in MEMS Pressure Sensors ........495
Georgi Kolev, Mariya Aleksandrova, Krassimir Denishev
Technical University of Sofia, Bulgaria
Different Technological Methods for Offset Compensation in Si Hall Effect
Sensors.............................................................................................................................499
Ivelina Cholakova
Technical University of Sofia, Bulgaria

xi
Multipoint Video Control System Applicable in Assistance of Elderly and
People with Disabilities....................................................................................................502
Ivo Iliev, Serafim Tabakov, Velislava Spasova
Technical University of Sofia, Bulgaria
Computer Modeling of RF MEMS Inductors Using SPICE.............................................505
Elissaveta Gadjeva
Technical University of Sofia, Bulgaria
Deposition of Transparent Electrodes for the Future Generation of Flexible
Displays ............................................................................................................................509
Mariya Aleksandrova, Georgy Dobrikov, Kostadinka Gesheva*, Georgy Bodurov*, Ivelina
Cholakova, Georgy Kolev
Technical University of Sofia, Bulgaria
*Bulgarian Academy of Science, Bulgaria
Investigation of Over Voltage Protection Circuit for Low Power Applications ............513
Tihomir Brusev, Nikola Serafimov, Boyanka Nikolova
Technical University of Sofia, Bulgaria
Realization of Low-frequency Amplitude Modulator and Demodulator
with FPAAs .......................................................................................................................517
Ivailo Pandiev
Technical University of Sofia, Bulgaria
Modification method to determining the output parameters in the audio
power stage with complex load.......................................................................................521
Plamen Angelov
Burgas Free University, Bulgaria
Modified method for design of the low-frequency audio driver ....................................525
Anton Petrov, Plamen Angelov
Burgas Free University, Bulgaria
SPICE Modelling of Magnetoresistive Sensors..............................................................529
Boyanka Nikolova, Georgi Nikolov, Milen Todorov
Technical University of Sofia, Bulgaria
Maximizing Power Transfer to the Remote Terminal of PCM4 System ........................533
Zoran Zivanovic, Vladimir Smiljakovic
IMTEL Komunikacije a.d., Serbia
Design, Analysis and Modifications of a Telecom Converter........................................537
Zoran Zivanovic, Vladimir Smiljakovic
IMTEL Komunikacije a.d., Serbia
Image processing of infrared thermograms for hidden objects ...................................541
Anna Andonova
Technical University of Sofia, Bulgaria

ENERGY SYSTEMS AND EFFICIENCY (POSTER)

Study and analysis of optimization approaches for insulation of an industrial
grade furnace with electrical resistance heaters ...........................................................544
Borislav Dimitrov, Hristofor Tahrilov, Georgi Nikolov
Technical University of Varna, Bulgaria
Improving energy efficiency of industrial grade furnaces with electrical
resistance heaters and comparative model-experiment analysis.................................548
Borislav Dimitrov, Hristofor Tahrilov, Angel Marinov
Technical University of Varna, Bulgaria

xii
Dynamic Braking in Induction Motor Adjustable Speed Drives ....................................552
Nebojša Mitrović, Milutin Petronijević, Vojkan Kostić, Bojan Banković
University of Niš, Serbia
Application of Active Front End Rectifier in Electrical Drives.......................................556
Bojan Banković, Nebojša Mitrović, Vojkan Kostić, Milutin Petronijević
University of Niš, Serbia
Cyclic Current Rating of Single-Core XLPE Cables with Respect to Designed
Life Time ...........................................................................................................................560
Miodrag Stojanović, Dragan Tasić, Aleksa Ristić
University of Niš, Serbia
The Influence of the Geometry of the Inductor on the Depth and Distribution
of the Inductively Hardened Layer ..................................................................................564
Maik Streblau, Bohos Aprahamian, Vladimir Shtarbakov*, Hristofor Tahrilov
Technical University of Varna, Bulgaria
*METAL PLC, Bulgaria
Examination of Frequency Controlled Asynchronous Drives at Variable Load
Torque - Laboratory Simulator ........................................................................................567
Vasil Dimitrov
Todor Kableshkov University of Transport, Bulgaria

CONTROL SYSTEMS (POSTER)

Multi Leveled Hierarchical Approach for Monitoring and Management
Information Systems Construction .................................................................................571
Emiliya Dimitrova, Galina Cherneva
Todor Kableshkov University of Transport, Bulgaria
Improving Control System in the Sulfuric Acid Plant ....................................................573
Viša Tasić, Dragan Milivojević, Vladimir Despotović*, Darko Brodić*, Marijana Pavlov
Institute for Mining and Mettalurgy, Serbia
*University of Belgrade - Bor, Serbia
DLadder – an Integrated Environment for Programming PIC Microcontrollers ...........577
Viša Tasić, Dragan Milivojević, Vladimir Despotović*, Darko Brodić*, Marijana Pavlov,
Vladan Miljković
Institute for Mining and Mettalurgy, Serbia
*University of Belgrade - Bor, Serbia

MEASUREMENT SCIENCE AND TECHNOLOGY (POSTER)

Frequency Measurement Using Compact DAQ Chassis ...............................................581
Georgi Nikolov, Boyanka Nikolova
Technical University of Sofia, Bulgaria

REMOTE ECOLOGICAL MONITORING (POSTER)

Wind Energy and Steps Towards 100 Percent of Renewable Energy
Penetration .......................................................................................................................585
Aleksandar Malecic
University of Niš, Serbia

xiii
 ENGINEERING EDUCATION (POSTER)
Multitool Online Assisted Design of Communication Circuits and Systems ...............589
Galia Marinova
Technical University of Sofia, Bulgaria
E-learning Systems as a Behavioural Analyst ...............................................................593
Valentin Videkov, Rossen Radonov
Technical University of Sofia, Bulgaria
Application of Remote Instrumentation in Learning using LabView ............................595
Ivo Dochev, Liljana Docheva
Technical University of Sofia, Bulgaria
Simulation of third-order dispersion in single optical channel.....................................599
Kalin Dimitrov, Tsvetan Mitsev, Lidia Jordanova
Technical University of Sofia, Bulgaria
Developing of a Video Information System for the Technical University of Sofia .......603
Kalin Dimitrov, Rumen Mironov, Alexander Bekjarski
Technical University of Sofia, Bulgaria

xiv
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Multiuser IR-UWB System Performance

Razvan Craciunescu1, Simona Halunga2, Octavian Fratu3
Abstract – This paper analyzes a multiuser system performance (AWGN), while the UWB signal will be an Impulse Radio
under the Standard Gaussian Approximation - SGA. In particular Ultra Wideband signal (IR-UWB). The IR-UWB transmitting
it focuses on the analysis of the multiuser interference method is the most common way of transmitting UWB
contribution. Simulations have been performed for a pulse position signals. This type of signals are radiated pulses very short in
modulation and a pulse amplitude modulation for 5, 20 and 50
time.
interfering users. The performances have been evaluated in terms
of the error probability over the signal to noise ratio, assuming that Several methods of evaluating the effects of the multi-user
the signal is transmitted through a AWGN channel. interference have been developed over the years [3,4]. This
paper focuses on the Standard Gaussian Approximation
Keywords – Ultra Wideband communication, Standard (SGA) hypothesis, which models the effects of all the
Gaussian Approximation, Probability of Error. interferences as a Gaussian additive noise, with a uniform
power spectrum distribution over all frequencies of interest.
The hypothesis is very accurate for a large number of
I. INTRODUCTION interfering users, and has very optimistically results for low
values of bit rates or for a small number of interfering users.[
Ultra Wideband Technology (UWB) has been described as
one of the most promising technologies during the last decade II. THE PERFORMANCES OF A MULTI-USER IR-UWB
[1-3]. It offers the possibility of achieving higher rates for SYSTEM IN SGA HYPOTHESIS
indoor systems with a reduced range of action due to the
resistance encountered in multi-path environments. Moreover, In this section we will present the general simulation
this technology offers lower implementation cost and reduced scenario and the analytical expressions of the probability of
power requirements then most of other technologies error for the two types of modulation which will undergo our
mentioned in the literature [1-2]. Taking into consideration the analysis, PPM and PAM.
FCC regulations, a UWB is defined as being any signal in In the following work, the following assumptions are
which the 3 dB bandwidth is at least 25% of the central assumed to be valid:
frequency or any signal with a bandwidth larger than 500
1. All sources produce binary vectors b, bk={0,1}, ∀k .
MHz [1].
2. All sources use the same pulse code period frequency, Ts.
The UWB radio channels can use frequencies from 3.1
GHz to 10.6 GHz, using a frequency bandwidth larger then 3. The spreading codes c k ∈ {± 1}, (∀)k are independent and
7GHz, with the restrictions imposed by the spectral equally likely, with the same code period Tc.
frequency-power masks given in standards [1]. Every radio 4. For each transmission/reception path, a different code –
channel may occupy a bandwidth of at least 500 MHz, in known at the receiver - is used.
accordance to its central frequency. Regarding the multiple 5. It is assumed that the base impulse has a limited duration,
access techniques used, the original proposal for UWB was to Tm and a symmetrical frequency shape.
use Time Hopping – TH combined with the Pulse Position 6. Propagation is achieved on a channel with multi-paths.
Modulation – PPM. Later on, various modulation techniques For a given user, n, the channel impulse response is a function
have been used: PAM (pulse amplitude modulation), OOK of the path gain, α(n), and of the path time delay τ(n). Delays
(On-off keying) or other multiple access techniques such as are considered to be independent and uniformly distributed
DS (direct sequence) [2]. within the [0,Ts) interval. The channel impulse response is
In this paper we will present the performances of two types thus given by:
of pulse modulation, in position and in amplitude (TH-PPM
( n) (n)
and TH-PAM). The communication channel will be an h (t ) = α δ (t − τ (n) ) (1)
modeled as affected by Additive White Gaussian Noise
7. The channel is affected only by AWGN, with the spectral
2 power density of N0/2(W/Hz)
Razvan Craciunescu is with the Faculty of Electronics,
Telecommunications and Information Technology at University
8. Coherent single user correlation reception is
POLITEHNICA of Bucharest, 313 Splaiul Independentei, 060042 implemented at the receiver for all users. The signal will be
Bucharest-6, Romania, E-mail: rcraciunescu@radio.pub.ro . thus correlated with the user code and integrated over a bit
3 period Tb=NsTc.
Simona Halunga is with the Faculty of Electronics,
Telecommunications and Information Technology at University
POLITEHNICA of Bucharest, 313 Splaiul Independentei, 060042
Bucharest-6, Romania, E-mail: shalunga@elcom.pub.ro .
A. TH-PPM
4
Octavian Fratu is with the Faculty of Electronics,
Telecommunications and Information Technology at University The binary signal, TH-PPM, transmitted by user n can be
POLITEHNICA of Bucharest, 313 Splaiul Independentei, 060042 written as [3,4]:
Bucharest-6, Romania, E-mail: ofratu@elcom.pub.ro .

1
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

(n) ∞ ( n) ( n) ( n) ,where m(t) is the correlation mask upon reception defined by:
s (t ) = ∑ E p (t − jT − c T − a ε ) (2)
TX TX 0 s j c j
j = −∞ N −1
S (1)
m(t ) = ∑ v(t − jT − c T ) (8)
(n ) is the energy
where p0(t) is the normalized base impulse, ETX s j s
j=0
transmitted by each pulse, c (jn )Tc is the time shift imposed by
v(t ) = p (t ) − p (t − ε ) (9)
0 0
the TH code, c (nj ) is the j-th TH code sequence used by user n,
and Tc is the chip duration. Each TH code is a sequence of Np Combining Eqs. (7) and (9) we obtain [4]:
identically and independently distributed random variables,
each of them with a probability of 1/Nh, and with values Z =Z +Z +Z (10)
u mui n
within [0,Nh-1] interval, where Nh is the cardinality of the TH
code. In order to identify the users, each of them will be Under the SGA hypothesis, Zmui and Zn represents random
assigned a specific TH code in such a way to avoid collision Gaussian processes with a 0 mean and a variance of σ mui 2
and
at the receiver. The term a (nj )ε the time-shift introduced by σ n respectively. The average bit error rate Prb can be written
2

the data; ε is the specific PPM time delay and aj is the binary as [3,4,5]:
value assigned to the jth pulse for user n. The binary vector a
represents the output of a (Ns,1) repetition coder, that receives  
1  E
b 
as input the binary vector b, meaning that Ns pulses carry the Pr = erfc 
b 2 2
 2(σ + σ 2
information of one bit. The binary vector a length is the length ) 
 n mui 
of b time Ns.[1,3,4,5]
Assuming that the channel is modeled by Eq. (1), in the =
1


erfc
((SNR )
n
−1
+ (SIR ) )
−1 −1 
 (11)
2  2 
presence of AWGA noise the received signal can be written as  
N
u ∞ ( n) ( n) ( n)
The bit energy of the received signal, Eb, can be obtained
r (t ) = ∑ ∑ E p (t − jT − c T − a ε − τ ) + n(t ) (3) by calculating the energy of the useful components at the
RX 0 s j c j
n = 1 j = −∞ output of the receiver for all Ns pulses that form a bit.
Therefore
where Nu is the number of users and ERX is the energy of each
transmitted pulse at the receiver. [1,3,4,5]. (1) 2
E =E N (1 − R (ε )) 2 (12)
Referring to user (1) and assuming that the receiver is b RX s 0
perfectly synchronized, such that the time delay is accurately
known at the receiver and can be assumed 0, for simplicity, ,where R0(t) is the autocorrelation function of the base impuse
the received signal can be written as [4]: p0(t) pulse. In the presence of the thermal noise, the signal to
noise ratio (SNR) can be written as:
r (t ) = r (t ) + r (t ) + n(t ) (4)
u mui (1)
N s E RX E (1)
SNR n = (1 − R 0 (ε )) = b (1 − R 0 (ε )) (13)
N0 N0
Next, focusing our analysis on the bit interval Tb and taking
into account the symmetry of the system, the analysis can be
Regarding the signal to interferences ratio (SNI) it can be
performed within [0,Tb] interval. The ru(t) and rmui(t)
contribution can be written for t ∈ [0, Tb ) as[4]:
written as
(1) 2
E N (1 − R (ε )) 2 (1 − R (ε )) 2 γ (14)
N −1 SIR = RX s 0 = 0 R 1
S (1) (1) (1)
r (t ) = ∑ E p (t − jT − c T − a ε ) (5) N σ 2 N ( n )
RX 0 s j c j 1 u ( n) u E RX
j=0 N σ2 ∑ E M
R ∑
T s M RX b
s n=2 n = 2 E (1)
∞ RX
N ( n)
U ∑ E p (t − jT
r (t ) = ∑ j = −∞ RX 0 s By combining Eqs. (13) and (14) and replacing them in Eq.
mui
n =1 (11) we obtain the average bit error rate [3,4,5]:
 − 1 
(n) (n) ( n)   −1
−c T − a ε −τ ) (6)      
j c j    
 −1   
   (1)    (15)
 (1 − R (ε )) 2 γ 
In the decision process performed at the reception, the 1  1   Eb   
Pr = erfc  (1 − R (ε ))  + 0 R   
b 2 0  N E ( n) 
correlation output at reception is thus given by [2,4,5]:  2   N 0 
 2 u RX 
 
    
  σ M Rb ∑  
T   n = 2 E (1)  
b    RX   
Z = ∫ r (t )m(t )dt (7)    
 
0

2
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

C. TH-PAM

The TH-PAM signal can be analyzed by following a similar

procedure as the one presented above. The binary signal
transmitted by user n can be written as [6]:

(n) ∞ ( n ) ( n) ( n)
s (t ) = ∑ E a p (t − jT − c T ) (16)
TX TX j 0 s j c
j = −∞

The signal at the output of the correlator at reception has

the same expression as in Eq. (7) with a correlation mask m(t)
defined as [1,2,6]: Fig. 1 Probability of error for 5 users and PAM and PPM
N −1
S (1)
m(t ) = ∑ p (t − jT − c T ) (17)
0 s j c
j=0

With the same decision criterion as in the above paragraph,

the error probability, Prb obtained in Eq. (11) is still valid.
The expressions for the signal to noise ratio (SNR) in the
presence of the thermal noise, as well as the signal to
interference ratio (SIR) can be written as:
(1) (1)
N E 2E (18)
SNR = s RX = b
n N N
0 0
2 Fig. 2 Probability of error for 20 users and PAM and PPM
(1)
E N T γ 1 (19)
SIR = RX s s = R
N T N ( n)
u ( n) M u E RX
σ2 ∑ E ∫ R 2 (τ )dτ R ∑
M RX 0 b
n=2 −T n = 2 E (1)
M RX

Therefore, the probability of error, Prb, is given by [6]

 − 1 
  −1 
     
   
 −1    (20)
  (1)   
 γ 
 1   2 Eb   
 + 
1 R
Pr = erfc   
b 2  ( n ) 
 2   N 0  
N E
u RX M
T

  Fig. 3 Probability of error for 50 users and PAM and PPM
   2  
 R ∑ ∫ R0 (τ )dτ  
  b n = 2 E (1) − T  
   RX M   
    From these figures we can extract the following
 
conclusions. First, in Fig. 1 we can notice that the multi-user
interference term can be neglected, because the probability of
III. NUMERICAL RESULTS AND SIMULATIONS error decreases as the SNR increases. We can say that, if the
number of users is small enough, only the thermal noise
As far as the numerical results are concerned we will affects the probability of error. The Eb/N0[dB] distance
analyse the performances of a IR-UWB system in the between PAM and PPM is approximately 3dB, for a Prb =10-2,
presence of the multi-user interferences (MUI). First, we will and it increases for the lower values of Prb. From Figs. 2 and
evaluate the probability of error, Prb, in the case of using the 3 we can observe that the probability of error Prb tends to a
TH multiple access technique both for a PPM binary constant value, as the signal to noise ratio increases, showing
modulation and for a PAM one. In both circumstances Prb the fact that, for large Eb/N0, the system performances are
will be estimated in accordance with the theoretical results dominated by the multi-user interference. We can, moreover,
obtained in Sections II.A. and II.B. identify two regions: for low values of Eb/N0, Prb is
We compared the performances for PPM and PAM in three determined mostly by the thermal noise, in which case we can
different scenarios, using 5, 20 and 50 interference signals. improve the performances of the system by increasing the
The results are presented in Figs 1, 2 and 3. The transmitted transmission power; for high Eb/N0 ratios, the systems
signal has a rate of 20 Mbit/s. The used pulse is given and has performances trends asymptotically to a constant value and
the shape of the second Gaussian derivative with a shaping does no longer depend on Eb/N0. In this case the performances
factor of 0.25ns. In the case of PPM the timeshift is ε=0.5 ns. are dominated by the multi-user interference effects.

3
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

IV. CONCLUSION
In this paper we have evaluated the performances of an IR-
UWB system in the presence of multi-user interferences and
in the Standard Gaussian Approximation hypothesis. We have
noticed that the multi-user interferences influence on the
performances of such a system is important only if the number
of users is large enough. We have also noted that the PAM
modulation is far more robust than the PPM one, achieving
error probabilities ten times smaller at the same Eb/N0 ratio.
As the number of users increases, the Prb is asymptotically
Fig. 3 Probability of error for 1: 200 users
limited to a higher value.
As the number of users increases, the Prb is asymptotically Next, investigating the validity of the SGA hypothesis in
limited to a higher value. If the number of users increses up to comparison with simulation results, we observed that, for a
50, the probability of error decrease only to 2 10-3, in the PAM relatively low number of users (up to 10) the results obtained
case, and to 2.5 10-2, in the PPM case. All graphs shows that based on simulation are better with respect to the error
PAM is slightly more robust then PPM with respect to Prb probability then the ones obtained under the SGA
performances. assumptions, showing that the formulas developed represents
As it can be seen in Figs. 2 and 3, Prb increases with the an upper limit for the Prb.
augmentation of the number of users. In order to evaluate the In a future work we will focus on increasing the number of
performances of the system with the number of users users in the simulation, developing simulations under different
increases. we represented graphically Prb as a function of sets of parameters and checking other hypotheses, like the
number of users. Thus, in Fig. 4 we can see that in the multi-user interference model based on package collision or
presence of more than 200 users, the error probability is very the chip-synchronous hypothesis.
high, with a magnitude order of 10-1 , 10-2, and the system is
dominated by M.U.I.
In order to validate the theoretical results, we simulated the ACKNOWLEDGEMENT
UWB receiver in the presence of multi-user interference. We
have been particularly interested in how precise the SGA This research activity was supported by Ministry of
hypothesis used for error probability estimation is fulfilled. Communications and Information Society of Romania under
We simulated a system with 7 and 10 users. Each user the grant no.106/2011 "Evolution, implementation and
generates stream of data with a bit period Tb=18ns, leading to transition methods of DVB radiobroadcasting using efficiently
a bit rate of 55.55Mbit/s. Every bit period is organized in 3 the radio frequencies spectrum".
frames with a duration Ts=6ns, meaning 3 pulses are
transmitted for every bit. Each frame is then divided into 6 REFERENCES
slots with a length of Tc=1ns. All users transmit with the same
format of the signal. The results are shown in Fig. 5.
[1] M. Ghavami, L. B. Michael, R. Kohno, “Ultra Wideband
We can notice that the theoretical model used for the Signals and Systems in Communication Engineering”, John
evaluation Prb underestimates the effect of MUI. The error Wiley & Sons, 2007.
probability obtained based on the simulation is larger than the [2] Kharrat-Kammoun, F. Le Martret, C.J. Ciblat,“Performance
one the SGA hypothesis predicted. In the 7 user case the analysis of IR-UWB in a multi-user environment”, Wireless
difference between the error probability from the theoretical Communications, IEEE Transactions on, Issue 11,Noiembrie
model and the simulated one is larger than in the 10 user case: 2009.
[3] G. Durisi, S. Benedetto, “Performance evaluation and
comparison of different modulation schemes for UWB
multiaccess systems”, Communications, 2003. ICC '03. IEEE
International Conference on
[4] Shing TenqChen,Ying-Haw Shu, Ming-Chang Sun, Wu-Shiung
Feng and Chao-Hao Lee, “Performance comparison of PPM-
TH, PAM-TH, and PAM-DS UWB Rake receivers with channel
estimators via correlation mask”, ICCOM'05 Proceedings of the
9th WSEAS International Conference on Communications
[5] M. Z. Win and R. A. Scholtz. “Ultra-wide bandwidth time-
hopping spread spectrum impulse radio for wireless-access
communication”. IEEE Transactions on Communications,
vol.48, no. 4 pp.:679-691, 2000.
[6] Zhiquan Bai, Kyungsup Kwak, “Performance analysis of TH-
PAM of UWB system and the coded scheme” Wireless
Communications, Networking and Mobile Computing, 2005.
Proceedings. 2005 International Conference on
Fig. 4 Comparasion between theoretical ans simulation results for 7
and 10 users

4
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Doppler Fading Effects on OFDM Transmissions

Ioana Bucsa1, Razvan Craciunescu2,Simona Halunga3, Octavian Fratu4
Abstract – This paper presents the performances of different Interferences), considerably decrease and reduce the
Dopler fading models for a OFDM communication. The complexity of the equalizer at the receiver [1.2].
simulations were carried out over a AWGN channel with a
BPSK modulation, and the performances have been evaluated in
terms of probability of error for each model. II. TYPES OF DOPPLER FADING THAT AFFECT THE
RADIO CHANNEL
Keywords – Fading, Doppler Effect, OFDM, MATLAB.

A. Multipath Channel

I. INTRODUCTION The performances of any wireless communication system is

strongly affected by the multipath phenomenon, very common
Radio propagation for mobile communications, especially in dense urban areas, when the transmission path between the
in dense urban areas or in indoor environment, is transmitter and the receiver is severely obstructed by
characterized by severe multipath phenomenon that causes buildings, trees or other objects.
fading and distortion effects, leading to inter-symbol The variation of the signal in the communication systems
interference (ISI) [1]. can be caused by the short and/or long term fading. Short term
The fading is one of the major factors affecting the fading includes multi-path fading (Rayleigh, Rice fading) and
performance of such systems [1,2]. The large-scale fading is the Doppler fading [2].
given by path loss and shadowing effects of buildings or The multi-path fading determines whether a channel is flat
prominent terrain contours. The small-scale fading is the or selective in frequency, while the Doppler fading divides
common reference to the rapid changes in signal amplitude channels into slow fading channels and fast fading ones.
and phase. In this work, the combined effects of large- and While the multi-path fading causes a scattering of the pulse in
small-scale fading are considered. time, the Doppler fading causes a scattering in frequency
Multi-carrier communication is a way to increase [2,3].
bandwidth without amplifying the noise in the signal on
account of the frequency selectivity fading that affects the
B. The Doppler Fading
channel. This type of communication implies sending several
narrower band signals (called subcarriers) instead of a single
The multi-path fading (Rayleigh, Rice) does not take into
broadband one. The signals are multiplexed in frequency and
consideration the possible movements of the emitter and/or of
they are transmitted together to the same receiver, on the same
the receiver. If the emitter/receiver is mounted on a moving
radio link. By sending M signals in parallel on the same radio
vehicle the Doppler fading occurs. If a signal is emitted with
link we increase the rate of transfer by M times. In the same
the f0 frequency, then the spectrum of the received signal will
time, the impact of frequency selectivity fading depends on
broaden and it will contain spectral components from f0-fd to
the bandwidth of each subcarrier [1].
cos ( )
f0+fd, where fd is the Doppler deviation given by [4]:
= (1)
An extended band involves higher transfer rates, especially
for the descending connection. As the symbol rate for each
subcarrier is much smaller that the initial symbol rate, the
where v is the velocity of the emitter/receiver, and θ is the
effects of delayed scattering, for example ISI (Inter Symbol
angle between the emitter/receiver forward velocity and the
line of sight from the emitter to the receiver.
1
Ioana Bucsa is with the Faculty of Electronics, If the band which is occupied by the useful signal is wider
Telecommunications and Information Technology at University than the Doppler bandwidth, the Doppler scattering will cause
POLITEHNICA of Bucharest, 313 Splaiul Independentei, 060042 no problem either for emission or reception. In this case we
Bucharest-6, Romania. are dealing with a channel affected by a slow fading. On the
2
Razvan Craciunescu is with the Faculty of Electronics, other hand, if the band is smaller than the Doppler bandwidth,
Telecommunications and Information Technology at University the movement produces a fast variation of the channel during
POLITEHNICA of Bucharest, 313 Splaiul Independentei, 060042 the length of the pulse. Thus the fading is considered to be
Bucharest-6, Romania, E-mail: rcraciunescu@radio.pub.ro .
3 fast.
Simona Halunga is with the Faculty of Electronics,
Telecommunications and Information Technology at University In the following paragraphs we will analyze in detail some
POLITEHNICA of Bucharest, 313 Splaiul Independentei, 060042 Doppler power spectrum models from the point of view of
Bucharest-6, Romania, E-mail: shalunga@elcom.pub.ro . their applicability and that of the theoretical expressions of the
4
Octavian Fratu is with the Faculty of Electronics, power spectral density (PSD) of fading processes [2,4,5].
Telecommunications and Information Technology at University The Jakes Doppler power spectrum model is applied to a
POLITEHNICA of Bucharest, 313 Splaiul Independentei, 060042 mobile receiver. It is called the classic model and it is built
Bucharest-6, Romania, E-mail: ofratu@elcom.pub.ro . based on the following hypothesis: propagation of radio

5
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

-?
( )= (6)
waves parallel to the ground; at receiver, the angle of arrival is
uniformly spread in [-π,π]; the receiver antenna is ?
omnidirectional.[3,4]. 1−( ⁄ )
1
The baseband normed Jake spectrum is:
( ) = , | | ≤ (2) where 0 ≤ ,ABC ≤| |≤ .AEF ≤ and the norming factor
1−( ⁄ )
defined as:
⁄2
where fd is the maximum Doppler deviation. -? = (7)
It has been proven that in a 3-D isotropic scattering sin': G ,AEF
H − sin': G ,ABC
H
environment, in which the arrival angles are uniformly
,ABC șK .AEF are the minimum and maximum
distributed in the elevation plan and in that of the azimuth, the where

is: ( , )=
total PSD for the θ angle of elevation and the α azimuth angle
sin
,0 ≤ < ,0 ≤ < 2 , the theoretical
positive frequencies for which the spectrum is non-zero.
, 4
These frequencies can be determined from the PSD of the
spectrum is flat.[2,3,4]. angle of arrival.

1
The baseband normed flat Doppler spectrum is: The restricted Jakes Doppler spectrum was considered to be

! ( ) = , | | ≤ (3)
symmetrical. The asymmetric spectrums occur in the case of
2
directional antennae, of aeronautical channels and of the
The next model corresponds to the multi-path components satellite mobile radio channels. Taking an asymmetrical
with high rates of delay in the UHF communications. It has spectrum in consideration, the pulse response will be
also been put forward in the case of the high frequency (HF) complex.
channels as well as for the aeronautical channels with a VHF The baseband asymmetrical restricted Jakes
Doppler(ajakes) spectrum is expressed analytically [2,4]:
-E
band [3,4].
( )= (8)
The baseband normed Jake Gaussian spectrum is:
1
() E
'
# ( ) = & *+ )
(4) 1−( ⁄ )
$2 %#
where σg is the standard deviation. where , − ≤ ,ABC ≤ ≤ .AEF ≤ and the norming
The Jakes bi-Gaussian model is built from two Gaussian factor defined as:

-E = (9)
spectrums which are shifted in frequency. This is used for
sin': G ,AEF
H − sin': G ,ABC
H
modeling the long echoes which can occur in urban areas and
hilled terrains [2,3,4].

,ABC șK .AEF are the minimum

The baseband normed Jakes bi-Gaussian spectrum is :
) )
where and maximum
2343+0 5 2343+) 5
' '
,# ( ) = -,# .
/+0 /+)
& )6+0 )
+ & )6+) )
8 (5)
positive frequencies for which the spectrum is non-zero.
The round spectral power density is approximated by the
1*+0 ) 1*+) ) measured PSD of a scattering component, while taking into
where %#: and %# are the standard deviations , #: and # are
consideration a wireless channel of 2.5 GHz. In this case, the

the central frequencies, ;#: and ;# are the power gains, and
PSD representation is also influenced by the frequency of the

-,# = is the norming coefficient. If ;#: = 0 or

:
central carrier [1,2,3,4].
/+0 </+)
The baseband rounded normed Doppler spectrum is
;# = 0, we can obtain a Gaussian Doppler spectrum which is
defined as follows:
P
? ( ) = ;? NO + O G H + OP G H Q (10)
shifted in frequency. In case both central frequencies are 0 and
the standard deviations are equal, the result is a Gaussian

where | | ≤
Doppler spectrum.
and the norming factor defined as:
1
As we have mentioned beforehand, the Jakes Doppler

;? = O O (11)
spectrum is build based on the fact that the angle of arrival at

covers the =− , > frequencies. If the angles are not 2 RO + + PS

3 5
the mobile receiver is uniformly distributed, and the spectrum

uniformly distributed, the spectrum does not cover this

We can notice that the rounded Doppler spectrum is a
interval, a fact which occurs in the case of a directional
polynomial in frequency function, of the fourth order, in
antenna. This type of spectrum is called restricted. The
O , O , OP are the coefficients of the polynom. In the IEEE
which only even exponents appear. The real numbers -
spectrum will also be considered symmetrical in order to
802.16 standard the following values are used: O = 1, O =
obtain a real impulse response [4].
−1.72, OP = 0.785.
The baseband normed restricted Jakes Doppler(rjakes)
spectrum is:

6
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA
Curba probabilitatii de eroare de bit pentru OFDM cu BPSK
0
10
doppler jakes
III. ORTHOGONAL FREQUENCY-DIVISION doppler
doppler
flat
rjakes
-1

MULTIPLEXING – BASIC PRINCIPLES 10

doppler ajakes
doppler gaussian
doppler rounded
-2
10
The Orthogonal frequency-division multiplexing(OFDM)

Pe
transmission is a multi-carrier type of transmission.
-3
A complex OFDM signal, x(t), during the mTu<t<(m+1)Tu 10

interval, can be written as follows [5,6,7]:

Z[ ': Z[ ':

T(U) = V TW (U) = V OW & (12)

(A)
-4

1WX(Y
10

W\ W\
-5
10
0 5 10 15 20 25 30 35 40
where xk(t) is the k-th subcarrier modulated with the Eb/No, dB

frequency of fk=k∆f and ak is the modulator symbol for the k-

th subcarrier in the m-th order OFDM time interval and Tu is Fig. 1. Probability of error for a AWGN, Rayleight fading channel,
the useful symbol duration [1,5,7]. and the Doppler frequency of 10Hz
The OFDM transmission is based on blocks, which means
that during the length of every OFDM symbol, Nc modulating simulations is 10Hz, 50Hz and 100Hz. Thus, the following
symbols are transmitted in parallel. The symbols can pe variations of the bit probability of error are obtained.
modulated using one of the following modulations: BPSK, For a Doppler deviation of 10Hz, we can notice in figure 1

10-5 for ], /_ =8-10 dB in the case of rounded, Akajes,

QPSK, 16QAM sau 64QAM. that the probability of error(Pe) reaches its minimum value of
At any rate, for a channel with time dispersion, the

Doppler and a Jakes one, Pe is minimum when ], /_ = 20

orthogonality between the subcarriers will be partially or Gaussian type of Doppler scattering. For a flat type of
definitely lost. Consequently, in the case of a channel with
time dispersion there will be both inter-symbol interferences dB, and then the error increases significantly for a Rjakes type
in the same subcarrier and interferences between different
Curba probabilitatii de eroare de bit pentru OFDM cu BPSK
subcarriers. In order to solve this problem and to make the 0
10

OFDM signal impervious to the time dispersion that occurs doppler

doppler
jakes
flat
within the channel, the cyclical prefix is introduced. The -1
10
doppler rjakes
doppler ajakes
insertion of the cyclical prefix consists of copying the last part doppler gaussian
doppler rounded
of the OFDM symbol and introducing that at the start of the -2
10
symbol. Once this cyclical prefix is inserted, the OFDM signal
Pe

increases from Tu to Tu+Tcp, where Tcp is the length of the -3

10
cyclical prefix, but also there occurs a decrease in the symbol
rate. If at the receiver end, the correlation is performed during -4
10
the Tu=1/∆f interval, then the orthogonality between the
subcarriers will be kept even if there is a channel with a time
-5
dispersion, with the condition that the duration of the 10
0 5 10 15 20 25 30 35 40
Eb/No, dB
dispersion is lesser than the length of the cyclical prefix
[1,2,6].
Due to its features and to the choice of an adequate spacing, Fig. 2. Probability of error for a AWGN, Rayleight fading channel,
∆f, between subcarriers, the OFDM allows for a less complex and the Doppler frequency of 50Hz
implementation from the point of view of the calculus
Curba probabilitatii de eroare de bit pentru OFDM cu BPSK
efficiency by using the Fast Fourier Transform (FFT) [1,2]. 10
0

doppler jakes
doppler flat
-1 doppler rjakes
IV. PERFORMANCE ANALYSIS 10
doppler
doppler
ajakes
gaussian
doppler rounded
-2
Simulations were carried out over an Additive white 10

Gaussian noise (AWGN) multipath channel, with

Pe

Ricean/Rayleigh fading and different types of Doppler fading 10

-3

presented in II.B. For the OFDM communication, the number

of subcarriers is 52, with a BPSK modulation and a frequency 10
-4

spacing of 312,5 kHz . The bandwidth is 20MHz, the useful

symbol duration is 3,2 µs and a cycling prefix of 0.8 µs. The 10
-5

length of the FFT algorithm used is 64. 0 5 10 15 20

Eb/No, dB
25 30 35 40

We are focusing our interest on the variation of the bit

probability of error for the different types of Doppler fading
Fig. 3. Probability of error for a AWGN, Rayleight fading channel,
presented in II.B. First of all a AWGN channel with Rayleigh
and the Doppler frequency of 100Hz
fading is analyzed. The Doppler frequency used in the
of Doppler.

7
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

In figure 2 the most favorable cases from the point of view ACKNOWLEDGEMENT
of the bit error probabilities are obtained for the rounded and
Gaussian type of Doppler, while the most unfavorable This research activity was supported by Ministry of
situation is that of the channel which is affected by the Rjakes Communications and Information Society of Romania under
Doppler. For the Ajakes Doppler, the proportion between the the grant no. 106/2011 "Evolution, implementation and
bit energy and the power spectral density reaches the value of transition methods of DVB radiobroadcasting using efficiently
13,8 dB, increasing until 22,8 dB and 23,7 dB in the case of the radio frequencies spectrum".
the channel which is affected by the Jakes Doppler, flat
respectively. Modifying the Doppler deviations to 100Hz, the
REFERENCES
around the value of the ], /_ proportion that increases by 8
only major difference that occurs in figure 4.25 is noticeable

dB, from 13,8 dB to 21,8 dB, in the case of the channel [1] Andreas F. Molisch, Wireless Communication Second Editions,
affected by the Ajakes Doppler. Wiley & Sons.,2011.
[2] Franz Hlawatsch , Gerald Matz , Wireless Communications
Over Rapidly Time-Varying Channels , Elsevier 2011.
V. CONCLUSION [3] R. H. Clarke, “A statistical theory of mobile-radio reception”,
Bell Sys.Tech. J., vol. 47, no. 6, pp. 957–1000, July-Aug. 1968.
[4] P. M. Shankar, Fading and Shadowing in Wireless Systems,
In this paper we compared the performances of several
Springer 2011.
Doppler fading models, presented in II.B, for an OFDM [5] Haiying Zhu, Bouchard, L., Boucher, L. , Performance of
communication. Thus, for lower Doppler frequencies we OFDM based wireless LAN system under Doppler over
obtained a better probability of error than for higher ones. Rayleigh fading, Communication Technology Proceedings,
Furthermore, in the case of Rayleight fading, for the same International Conference on. Vol.2,pp 1234 - 1237 , 2003.
probability of error, Eb/N0 is at minimum in the case of the [6] K.A.Hamdi, Unified Error-Rate Analysis of OFDM over Time-
rounded Doppler model and at a maximum for the Rjakes Varying Channels, Wireless Communications, IEEE
Dopller model. For Rician fading we observed the minimum Transactions on, Vol.10, No. 8,pp. 2692 – 2702, 2011.
for Eb/N0 is still for the rounded Doppler model but the [7] L. Noor A. , Anpalagan ;S. Kandeepan. SNR and BER
derivation and analysis of downlink OFDM systems with noisy
maximum is, in this case for the Jakes Doppler model.
fading Doppler channels, Signal Processing and Its
Therefore, both for Rician and Rayleight fading the more Applications, 9th International Symposium on, pp 1-4, 2007.
robust Doppler model is the rounded Doppler model. The
worst model, in terms of probability of error is the Rjakes
respectively the Jakes Doppler model.

8
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Investigation of the modulation type’s influence on the

DVB-T signals quality
Oleg Borisov Panagiev1
Abstract – In this paper are made and described studies that S/N [dB] = C/N + kr-f, where (2)
represent the influence of key critical factors, ranging from RF
performance characteristics, such as error vector magnitude kr-f - the factor for the roll-off Nyquist filtering in the demo-
(EVM), modulation error ratio (MER), through to roll-off and dulator/receiver.
FEC for M-ary QAM modulation of subcarriers in the OFDM Then Eq.1 can be written as fallowing:
signal.
Eb/N0 [dB] = S/N - kr-f - kRS - kQAM. (3)
Keywords – DVB-T, EVM, BER, S/N, roll-off. The values and the expressions for their obtaining for every
parameter and factor are given in Table 1, where α is the roll-
I. INTRODUCTION off factor by Nyquist skew, b is the number of bits per symbol
and RRS is Reed-Solomon code rate.
Advanced digital terrestrial systems behave quite
differently when compared to traditional analog TV as the TABLE I
signal is subjected to noise, distortion, and interferences along PARAMETERS AND FACTORS FOR QAM
its path. The secret to maintaining reliable and high-quality
modulation
services over DVB-T transmission systems is to focus on
critical factors that may compromise the terrestrial reception. 16-QAM 64-QAM
Once reception is lost, the path to recovery isn’t always Parameters
obvious. The problem could be caused by MPEG table errors, and Factors
or merely from the RF power dropping below the operational M 16 64
threshold or the cliff point. RF problems can include any of α 0,10 0,30 0,10 0,30
the following: terrestrial RF signal reflections, poor noise RRS 188/204
performance, or channel interference [1-4]. The results are b = log2M 4 6
based on the recent analytical research by used laboratory kr-f = 10lg(1-0,25α) -0,1169 -0,3345 -0,1169 -0,3345
tests and theoretical analysis. kRS = 10lgRRS - 0,3547
kQAM = 10lgb 6,0206 7,7815
II. MATHEMATICAL ANALYSIS
B. Calculation of BER
This mathematical analysis is made according to the For the calculation of the bit-error-rate (BER) for OFDM
international standard requirements [1], [2] for terrestrial radio DVB-T signal with 16-QAM and 64-QAM modulations of the
and television broadcasting and the characteristic parameters subcarriers we used expressions [1], [2]:
and features of the broadcasting in Republic of Bulgaria. The
researches are for the main figures of merit and factors of 2(b -1)
OFDM modulation signal with channel frequency and 16- PB = .PS or (4)
2b - 1
QAM/ 64-QAM modulation of the subcarriers.
1
PB = .PS , (5)
A. Calculation of Eb/N0 b
which give the relationship between Symbol Error Probability
Eb/N0 [dB] = C/N - kRS - kQAM, where (1) PS and the Bit Error Probability Pb of QAM with M constella-
Eb/N0 is the energy per useful information bit Eb referred to tion points, arranged in a rectangular set, for b even.
the normalized noise power N0; The Eq.4 makes no assumption about the constellation
C/N - carrier-to-noise ratio in-channel [dB]; mapping and is based on the probability that any particular bit
kRS - the factor for FEC to Reed-Solomon [dB]; in a symbol of b bits is in error, given that the symbol itself is
kQAM - the factor for the QAM modulation [dB]. in error. The Eq.5 assumes that an erroneous symbol contains
just one bit in error. This assumption is valid as long as a Gray
When we study the dependence of Eb/N0 of S/N can be used
coded mapping is used and the BER is not too high. These
kr-f, which gives the relationship between C/N and S/N:
equations give different results for symbols of two or more
bits. The second approach is generally adopted because DVB
1
Oleg B. Panagiev is with the Technical University of Sofia, systems employ Gray code mapping.
Bulgaria, E-mail: olcomol@yahoo.com.

9
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

As per Eq.5 when the received signal is distraught by 2) Calculation of MER

Additive White Gaussian Noise (AWGN) there is a
MER is the preferred measurement for the following
probability that any particular symbol will be wrongly
reasons [1]:
decoded into one of the adjacent symbols. The PS is given by:
• The sensitivity of the measurement, the typical
1
) . erfc [
3b.( Eb / N 0 ) magnitude of measured values, and the units of
PS ( Eb / N 0 ) = 2 . (1 - ]
M 2.(M - 1) measurement combine to give MER an immediate
familiarity for those who have previous experience of
3b.( E b / N 0 ) ,
×{1 - 21 . (1 - 1
) . erfc [ ]} (6) C/N or S/N measurement;
M 2.( M - 1)
• MER can be regarded as a form of S/N measurement
where erfc(x) is the complimentary error function given by: that will give an accurate indication of a receiver's
∞ 2 ability to demodulate the signal, because it includes,
erfc( x) = ( 2/ π ). ∫e -t dt. (6a)
x
not just Gaussian noise, but all other uncorrectable
For practical purposes Eq.6 can be simplified by omitting impairments of the received constellation as well;
the (generally insignificant) joint probability term to give the • If the only significant impairment present in the signal
approximation: is Gaussian noise then MER and S/N are equivalent
(they are often used interchangeably).
1 3b.( Eb / N 0 ) . The relationship between EVM and MER given by the
PS ( Eb / N 0 ) = 2 . (1 - ) . erfc [ ] (7)
M 2.(M - 1) expression:
This approximation introduces an error which increases EVM V = 1 /( MERV . V ) , (9)
with degrading Eb/N0, but is still less than 0,1 dB for 64-QAM
at Eb/N0 = 10 dB. As already stated, the above equations for where the peak to mean voltage ratio V, is calculated over a
Symbol Error Probability are based on certain simplifying large number of symbols (10 times the number of points in the
assumptions which can be summarized as "the system is constellation is adequate if the modulation is random) and
perfect except for the presence of additive white Gaussian each symbol has the same probability of occurrence then it is
noise", but within this rather generous constraint the equations a constant for a given transmission system. The value tends to
for PS are exact. a limit which can be calculated by considering the peak to
mean of all the constellation points. Table III lists the peak-
C. Calculation of EVM and MER to-mean voltage ratios for the DVB constellation sizes. When
expressed as simple voltage ratios MERV is equal to the
EVM and MER take into account the combined effects of reciprocal of the product of EVMV and the peak-to-mean
C/N; transmitter, upconverter phase noise; impairments such voltage ratio for the constellation.
as second and third order distortions; group delay; in-channel
frequency response problems (amplitude tilt or ripple) and TABLE III
microreflections. EVM and MER measure essentially the same PEAK-TO-MEAN RATIOS FOR THE DVB CONSTELLATION SIZES
quantity and easy conversion is possible between the two
measures if the constellation is known. Peak-to-mean voltage ratio
modulation
V
1) Calculation of EVM
16-QAM 1 341
For calculation of EVM, we use the relationship between 64-QAM 1 527
EVM and S/N [5]. After performing the appropriate mathe-
matical transformations we obtain the expression for EVM:
III. SIMULATION INVESTIGATION
S / N +k
-
EVM = 10 20 . 100% , where (8) In this section the simulated researches, of main parameters
k is the peak-to-average energy ratio (Table II). The Eq. 8 of DVB-T signals with 16-QAM and 64-QAM modulations of
used to calculate the EVM is directly from vector the subcarrier, are made for an existing (real-life) channels in
considerations of signal plus noise in relationship to the ideal the area of the city of Sofia (Figs.1 and 2). The research uses
constellation points of any M-QAM signal. The minus sign in the mathematical dependences from above and its results are
Eq. 8 is necessary because S/N is the ratio of signal to noise given in tables and graphics.
whereas EVM is the ratio of noise to signal. On Fig.1a is shown a picture with the parameters of
received television channel by D/K standard at a frequency of
TABLE II 818 MHz (ch.64), and on Fig1b – the signal’s spectrum. Table
EVM VALUES FOR VARIOUS MODULATIONS IV contains the numeric results from the simulation research
for Eb/N0, SER, BER and EVM at an amendment of C/N from
k 2÷30dB and respectively S/N from 1,88-29,88dB at α=0,1 [6].
modulation
dB Fig.2a shows a picture with the parameters of received tele-
16-QAM 2,5527 vision channel by D/K standard at a frequency of 626 MHz
64-QAM 3,6796 (ch. 40), and Fig2b - the signal’s spectrum. Table V gives the
numeric results of the simulation research [7] for Eb/N0, SER,

10
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

a) Menu for ch.64 (16-QAM) a) Menu for ch.40 (64-QAM)

b) Spectrum ch.64 (DVB-T) b) Spectrum ch.40 (DVB-T), ch. 36 and ch.41 (PAL-K)
Fig.1. ch.64 (DVB-T) Fig.2. ch.40 (DVB-T)

TABLE IV TABLE V
16-QAM VALUES 64-QAM VALUES
C/N S/N Eb/N0 SER BER EVM C/N S/N Eb/N0 SER BER EVM
dB dB dB % dB dB dB %
2 1,8831 -3,6659 0,3743 0,1995 60,01 2 1,8831 -5,4268 0,5629 0,2871 52,71
4 3,8831 -1,6659 0,3153 0,1681 47,67 4 3,8831 -3,4268 0,5298 0,2702 41,87
6 5,8831 0,3341 0,2469 0,1316 37,86 6 5,8831 -1,4268 0,4878 0,2488 33,26
8 7,8831 2,3341 0,1732 0,0923 30,08 8 7,8831 0,5732 0,4351 0,2219 26,42
10 9,8831 4,3341 0,1027 0,0547 23,89 10 9,8831 2,5732 0,3704 0,1889 20,98
12 11,8831 6,3341 0,0472 0,0251 18,98 12 11,8831 4,5732 0,2941 0,1500 16,67
14 13,8831 8,3341 0,0146 7,79E-03 15,07 14 13,8831 6,5732 0,2103 0,1073 13,24
16 15,8831 10,3341 2,46E-03 1,31E-03 11,97 16 15,8831 8,5732 0,1281 0,0654 10,52
18 17,8831 12,3341 1,61E-04 8,60E-05 9,51 18 17,8831 10,5732 0,0611 0,0312 8,35
20 19,8831 14,3341 2,39E-06 1,27E-06 7,56 20 19,8831 12,5732 0,0200 0,0102 6,64
22 21,8831 16,3341 3,37E-09 1,80E-09 6,00 22 21,8831 14,5732 3,68E-03 1,88E-03 5,27
24 23,8831 18,3341 1,16E-13 6,17E-14 4,77 24 23,8831 16,5732 2,76E-04 1,41E-04 4,19
26 25,8831 20,3341 0 0 3,79 26 25,8831 18,5732 5,03E-06 2,56E-06 3,33
28 27,8831 22,3341 0 0 3,01 28 27,8831 20,5732 9,89E-09 5,04E-09 2,64
30 29,8831 24,3341 0 0 2,38 30 29,8831 22,5732 5,74E-13 2,93E-13 2,10

BER and EVM at an amendment of C/N from 2÷30 dB and indication for Quality on Figs.1a and 2a, which reaches 99%.
respectively S/N from 1,88-29,88 dB at α=0,1. For low values of S/N (1,88÷19,88 dB) BER is very high
The numeric results are shown graphically in Figs.3, 4 and and the receiving is either with errors or not possible, and
5, while in both cases (16-QAM/ 64-QAM) after the Viterbi EVM amends from 7,3% to 0,92%. For S/N ≥23 dB EVM
and Reed-Solomon decoding BER is low (10-9 ÷10-11) - for values are almost equal. At an S/N =29,88 dB for 16-QAM is
S/N between 22 and 28 dB. That confirms itself through the EVM=2,38% and for 64-QAM is EVM=2,1%.

11
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The operation mode is in both cases 8k and the modulation

1,E-01 non-hierarchical. There is also a difference between the levels
of received signals (level of ch.64 is less than a one of ch.40).
1,E-03 Near to the ch.40 is transmitted an analog television channel
(ch.41 with picture carrier 631,25 MHz - PAL-К), which in
some DVB-T receivers causes interference that deteriorate the
1,E-05
researched parameters and mostly BER (which increases).
BER

That makes receiving ch.40 impossible. Methods and means

1,E-07 for the elimination of this problem and rising of S/N and
16-QAM reduction of BER are an object of review in the second paper
1,E-09 64-QAM at the conference.

1,E-11 IV. CONCLUSION

1,E-13 It should be mentioned that for QAM systems DVB only

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 employs Gray coding within each quadrant, the quadrant
S/N, dB boundaries are not Gray coded, and the mapping is partially
Fig.3. BER=func(S/N) differentially coded. Further work is required to establish the
exact PS to PB relationship for this combination of mapping
and coding [1], [2], [8], [9].
1,E-01
REFERENCES
1,E-03
[1] DVB Document A014 Rev. 2, “MEASUREMENT
GUIDELINES FOR DVB SYSTEMS”, 2001.
1,E-05 [2] ETSI EN 300 744 V1.4.1 (2001-01), Framing structure, channel
coding and modulation for digital terrestrial television, 2001.
BER

[3] E. Benoit, Digital television, 3rd ed., Focal Press by Elsevier,

1,E-07
Oxford, 2008.
16-QAM
[4] K. Koichev, S. Sadinov, G. Traikovski. Defining the channel
1,E-09 64-QAM parameters and their influence on SER for OFDM transmission.
ICEST 2007, Proc. of Papers, Ohrid, Macedonia, 2007, pp.433-
436.
1,E-11
[5] Technical reference for the RF engineer from the RF experts,
TechNotes, AXCERA LLC, 05.12.1999.
1,E-13 [6] http://www2.rohde-schwarz.com, application note, 2005-2011.
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 [7] http://www.wolframalpha.com
[8] R. Hranac, “Carrier-to-Noise versus Signal-to-Noise”, Cisco
Eb/No, dB
Systems, sisco.com, 2003.
Fig.4. BER=func(Eb/N0) [9] E. Lecomte, “DVB T/H Seminaire”, Agilent technologies, 2008.

60
55
50
45
40 16-QAM

35 64-QAM
EVM, %

30
25
20
15
10
5
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
S/N, dB
Fig.5. EVM=func(S/N)

12
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Improving the reception of class DVB-T receivers

Oleg Borisov Panagiev1
Abstract - This paper presents some early results from defined for outdoor reception using an omnidirectional
experimental measurements in the service area of the DVB-T antenna located at least at 1.5 meters above ground level [5].
(8K) network. The aim of this study has been to determine the The results presented here apply to both cases.
degradation suffered by digital signals in presence of analog
PAL-K adjacent channel broadcasting. The result of this study
is a circuit solution and curve that estimates protection ratios II. EXPERIMENTAL AIMS AND TASKS
between analog and digital received power levels.
The formulation of the aims and the tasks is made on the
Keywords – DVB-T, PAL-K, receiver, adjacent cannel
basis of trials on several (different) DVB-T receivers. Further
interference, BER.
those would be called for shorter set-top-box (STB). Two of
them are for connecting to a TV receiver (television set): one
I. INTRODUCTION (STB1) is connected through an additional Euroscart-
Euroscart or Euroscart-Chinch cable; the other one is
Recently digital terrestrial TV using the DVB-T standard connected directly to Eurocscart of the TV receiver
[1] is gaining more and more interest in Europe and other (television set), whereby it is supplied through an adapter.
parts of the world. This fact leads to a simulcast environment The infrared (IR) receiver is a separate module, and its
during a period of several years, depending on different connection with the DVB-T receiver is through a cable,
countries and estimations, in which analog and digital which allows controlling the DVB-T receiver even though it
services will be sharing the same spectrum frequency bands. is behind the TV receiver. The third STB3 is connected to a
During the transitional period due to the coexistence of both PC through a USB port. The dialog with it is fulfilled via the
technologies, minimizing the digital service quality display (monitor) of the PC, as for the functioning of the
degradation caused from analog transmitters (and vice versa) DVB-T receiver we have to install certain software (sold in a
in the same coverage areas will be one of the main factors to CD together with the DVB-T receiver).
consider when planning digital services. The three DVB-T receivers were connected respectively to
At the same time manufacturers offer DVB-T receivers a TV and a PC, then set and every digital TV channel that is
with different application: domestic, portable, notebook, PC broadcasted in the region of the city of Sofia was chosen. The
(Fig.1), built-in TV. Some are with low adjacent channel signals were received by three different antennas:
selectivity and lower sensitivity. This in many cases leads to 1) Active antenna with G=45dB;
poor or impossible to receive digital signals in the presence 2) Yagi with G=18dB;
of a adjacent (upper) channel, low level of the received signal 3) Whip
and others [2], [3]. Each one of the antennas was successively connected to
STB1, STB2 and STB3 and the levels of the signals with the
Antenna
TV respectively carrier frequency were measured by STB in % as
well as with a level strength meter (LM) in dBµV. The results
of the measurement are given in Table I.
Note: Antennas 1) and 2) are with a direct visibility to the
transmitter centers (Kopitoto and the old TV tower).
After the measurements were done was established that
PC STB1 and STB2 receive all digital channels without any
problems, qua the strength and quality indicators are with
STB3 evidence ≥50% independently of the antenna type. However,
Fig.1. DVB-T reception with personal computer (PC) -TV tuner STB3 had a problem receiving on ch. 40 and ch. 52, although
the conditions for receiving were the same as for STB1 and
In this paper are presented the results of theoretical and STB2.
experimental investigations carried out to study the TABLE I
relationship between BER and protection ratio (S/N
An-
respectively) for portable reception in an urban environment 1) 2) 3)
tenna
inside the coverage of a Single Frequency Network (SFN).
Strength

Strength

Strength

Two types of portable reception have been defined by the

Channel

Quality

Quality

Quality
Level

Level

Level

European Standards Institute (ETSI): Class A and Class B

portable reception [4]. Class B portable reception is defined
as indoor (inside a building) reception whereas Class A is % % dBµV % % dBµV % % dBµV
40 61 99 91,2 63 99 68,1 60 99 58
1
Оleg B. Panagiev is with the Technical University of Sofia, 41 - - 96,8 - - 74 - - 63,8
Bulgaria, E-mail: olcomol@yahoo.com

13
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

On the base of those results were formulated the aims and is the degradation in Power received between digital TV
tasks of this research: signal (PrD) and analog TV signal (PrA) and is given by the
a) To be established the reason for not receiving ch.40 following formula [9]:
and ch.52;
b) To eliminate the problems with receiving ch.40 and PR[dB] = PrD [dBW ] − PrA [dBW ] (1)
ch.52, i.e. to be provided their receiving with STB3 and other Eq.1 could be written as follows, having in mind the
similar to it [2], [3]. connections between power and voltage [10] by impedance
To accomplish those aims were made a number of 75Ω:
measurements according to the standards [1], [4], [6] and the
results were analyzed. As a result of that was made a PR[dB] = U rD [dBµV ] − U rA [dBµV ] . (2)
conclusion, that such a problem could appear at a low
selectivity at adjacent channel and by an interference of The practical determination of PR is made due to Eq.2 for
analog signal (PAL-K), when it is transmitted in higher ch.40 and measured levels of two signals (digital and analog)
adjacent channel (by measurements with a spectrum analyzer, via spectrum analyzer (SA) Promax AE-476 (Fig.3),
was established, that such channels are available: ch.41 and according the requirements of [7].
ch.53 at standards G and K), (Fig.2). Note: During the simulation was established that the solutions of
the placed aims for ch.40 are also valid for ch.52, but at
To increase the selectivity at a adjacent channel (without corresponding values of constructive elements (coils and capacitors)
changing anything in the STB3) was made the decision for and frequency parameters of band pass filter.
creating a band filter with a certain slope and frequency band, The dependence of deterioration of the signal-to-noise ratio
which would be formed in a different module. Its assembly (∆S/N) from the amendment of PR is shown on Fig.4. The
must be outside STB3 and to be applicable to different STBs resultant ratio signal-to-noise (S/NI) at a presence of
with similar parameters. interference from the analog channel in the digital one are
calculated through the following formula:

Fig.3. Measurement setup

Fig.2. Spectrum of ch.40and ch.41
14
The first task was to design a BPF and simulate the
influence of its parameters onto S/N and BER, respectively 12
QoS (Quality of signal).
The second task is to be made an actual BPF to be 10
assembled between the antenna and STB3.
∆S/N, dB

The third task is to be checked the receiving of ch.40 and 8

ch.52 (strength and quality) and to be made an analysis for
the application of the proposed solution. 6

4
III. CIRCUIT SOLUTIONS AND RESULTS
2
For researching the influence of the parameters of Band
Pass Filter (BPF) on improving the quality of receiving for 0
the aforenamed channels, as well as for optimal choosing of -15 -10 -5 0 5 10 15 20 25 30 35 40
suitable scheme solution (Table II) was used an Ansoft PR, dB
Designer VS 2.0 [7]. A large number of simulations were
made to determine the scheme solution and the number of Fig.4. ∆S/N[dB] =func (PR [dB])
links in filter, which give the necessary values of Protection
ratio (PR), [1], [4], [6], [8]. The theoretical and experimental
researches were made with five values for PR, where in dB, S / NI [dB] = S / N [dB] − ∆S / N [dB] , (3)

14
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

TABLE II

BER
PR
№ BPF Experimental Theoretical
dB 1) 2) 3) 1)* 2)* 3)*
1 no -6 0,0297955 0,0153234 0,0787797 0,0150099 0,0076909 0,0402155

2 1 0,0024338 0,0006898 0,0153234 0,0012177 0,000345 0,0076909

3 12 6,967E-05 8,322E-06 0,0015181 3,484E-05 4,161E-06 0,0007593

4 23 3,21E-07 1,009E-08 4,724E-05 1,61E-07 5,046E-09 2,362E-05

5 35 3,095E-08 5,401E-10 1,052E-05 1,547E-08 2,701E-10 5,259E-06

where S/N is signal-to-noise by the absence of interference 1,E+00

from the analog PAL-K signal into the digital DVB-T
signal (cannel).
The theoretical determination of BER is made through: 1,E-02

1,E-04
PB = 0,89.[erfc ( 10( S / N I −7, 43) / 10 / 7 )]
BER

1,E-06 1)
× [1 − 0,44.erfc ( 10( S / N I −7, 43) / 10 / 7 )] . (4) 2)
3)

1,E-08 1)*
The dependence of BER from PR is given in Fig.5 2)*
wherefrom is visible, that with antennas 1) and 2) and BPF 3)*
the receiving of ch.40 is already possible, when PR≥15dB.
1,E-10
At values less than 15dB, receiving with active filter (Fig.3)
-10 -5 0 5 10 15 20 25 30 35 40
and any of the antennas is difficult (presence of sampling)
PR, dB
or impossible (Fig.6).
Fig.5. BER = func (PR [dB])

15
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

IV. CONCLUSION
The offered solution for improving the receiving with
DVB-T receivers (computer TV-tuners) gives the
opportunity to achieve the (placed) aims. By the practical
realization is needed to change the filters for ch.40 and
ch.52 or to remove them, and points 1 and 2 (Fig.3) to be
connected directly with each other in order to receive all
other channels. On Fig.3 is offered a version with shift of
the three regimes. BPF can be realized with varicaps so that
the passing channel is changed electrically. However, this
solution requires an additional supply and so is the practical
realization elaborated. At a subsequent publication will be
offered another version, by which the mentioned problems
would be removed and the practical application of the
offered solution will be facilitated.
a) BER>10-4
REFERENCES
[1] ETSI EN 300 744 v.1.4.1, Digital Video Broadcasting
(DVB); Framing structure, channel coding and modulation
for digital terrestrial television, 2001.
[2] DIGIVOX mini Hybrid TV tuner and video capture unit.
MICRO-STAR INT’L CO., LTD.
[3] USB TV tuner AVerTV Hybrid Volar HD. AVer Media
Technology.
[4] TR 101 190: “Digital Video Broadcasting (DVB);
Implementation Guidelines for DVB Terrestrial Services:
Transmission Aspects”. European Telecommunications
Standards Institute (ETSI), 1997.
[5] P. Angueira, M. Vélez, D. De La Vega, G. Prieto, D. Guerra,
J. M. Matías and J.L. Ordiales, “DTV reception quality field
tests for portable outdoor reception in a single frequency
network”, IEEE Trans., Broadcasting, vol.50, no.1, pp.42-48,
2004.
[6] Rec. ITU-R BT. 1368-2 “Planning Criteria for Digital
b) 10-6<BER<10-4 Terrestrial Television Services in the VHF/UHF Bands”,
2000.
[7] http://www.ansys.com/Products/Simulation+Technology/Ele
ctromagnetics/Ansoft+Designer.
[8] S. O’Leary “Digital/Analogue Co-channel Protection Ratio
Field Measurements” IEEE Trans., Broadcasting, vol. 44, no.
4, pp. 540-546, 1998.
[9] Regional Radiocommunication Conference for planning of
the digital terrestrial broadcasting service (RCC-04), ITU-R
Report, Geneva 2004.
[10] D. Dobrev, L.Yordanova, Receiving on radio- and TV
programs by means satellite and cable, Sofia, Electroninvest,
1996.

c) BER<10-6
Fig.6. Pictures with BPF for different PR and BER

16
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Monolithic Integrated Antennas with

High Radiation Efficiency
Hristomir Yordanov1
Abstract— The wireless chip-to-chip interface is an interesting There are two main problems in integrating the antenna
alternative to wired digital buses. An important element from a within a CMOS chip. First is the area covered by the antenna.
wireless interconnect is the monolithic integrated antenna. On- Normally this area is several square millimetres, which is
chip antennas have to meet several restrictions—they have to
cover very little chip area and they need to have high radiation unacceptable by integrated circuit manufacturers. The second
efficiency in a large enough frequency range to allow for fast problem is the antenna radiation efficiency. The silicon sub-
inter-chip communication. This work describes two methods strate used in CMOS technology is lightly p− doped, which
for designing efficient on-chip antennas, namely using high- decreases its resistivity to several Ohm.cm. The electromag-
impedance substrate with standard thickness, or using very thin netic field within such substrate is subject to heavy dielectric
substrate.
losses due to this low resistivity, which reduces the antenna
Keywords— Embedded antennas, Near-field communication, efficiency.
On-chip integrated systems.
The first of these two problems can be tackled by double us-
ing available metallisation structures on the integrated circuit.
I. I NTRODUCTION For example the chip ground supply planes can simultaneously
The wired inter-chip interconnects within a system are serve as antenna electrodes [5]–[7]. This has been achieved by
becoming a substantial bottle-neck for the chip-to-chip digital cutting the top metallisation layer into patches and feeding the
communications. The wired bus suffers frequency bandwidth RF signal across the gap between the patches.
limitations due to auto-interference effects like cross-talk and The efficiency of the integrated antenna can be optimised
dispersion [1]. These effects can not be reduced by optimising by two methods. One is to use high-resistivity substrate [6]. A
the design of the printed circuit board, because the limita- drawback of this method is that the thermal conductivity of the
tions of that technology have long been met. The standard substrate is also reduced and this changes the standard CMOS
approach for optimising inter-chip communication is based on technology. Another interesting option, presented in detail
using parallel lines and using software techniques like data in the current work, is to use very thin substrates. Current
buffering. The cost of these solutions is increased circuit board technologies like the ChipFilm technology, developed by the
complexity. Parallel lines connecting multiple devices result Institute for Microelectronics in Stuttgart, Germany, offers
in complicated board layout with very long design time—it is the possibility to fabricate integrated circuits on substrates as
possible to have a multi-layer board design time up to several thin as 6 µm [8]–[10]. Such a substrate increases the antenna
months. efficiency by reducing the volume where the electromagnetic
An interesting alternative to wired chip-to-chip interface is field is subject to a loss. A drawback of this solution is the
the wireless communication. An efficient wireless interconnect introduction of several additional steps to the standard CMOS
has the potential to offer data rate higher than its wired technology.
counterpart, while keeping the system design much simpler II. A NTENNA M ODE
and area-efficient [2].
A cross-section view of the integrated antenna using the
The technical requirements for a wireless interconnect are
circuit ground plane as electrodes is presented in Fig 1. The
specified by several conditions. First, the whole transceiver
antenna is fabricated either on high-resistivity (about 1 kΩ.cm)
circuitry including the antennas must be integrated within the
substrate with thickness of 675 µm, or on thin low-resistivity
chip. Second, wireless link must provide sufficient bandwidth
substrate, as discussed in the previous section. The active el-
for fast enough communication. And third, the range of the
ements are fabricated atop the substrate. Several metallisation
link should be very short—a distance of several centimetres
layers follow, where the on-chip interconnects are fabricated.
is to be covered. Therefore a suitable carrier frequency in
The top metallisation layer holds the ground supply plane. This
the millimetre range can be used to design good chip-to-chip
plane is cut into patches and an RF generator is connected
wireless link. The RF circuitry required for the transceivers
across the resulting gaps, exciting the antenna.
can be readily implemented even in 90 nm CMOS technology
Since the patches serve as ground planes also for the
[3], [4], the small wavelength allows for on-chip antenna
power supply for the CMOS circuitry underneath them, a
integration and the high frequency allow for higher absolute
low-frequency galvanic connection must be provided between
bandwidth and therefore higher data rates.
them. The block inductors, providing a DC connection be-
1 Hristomir Yordanov is with the Electromagnetic Compatibility
tween the patches are not shown in the figure.
Lab at the Technical University of Sofia, Faculty of Telecommu- The antenna structure consists of two or more patches, sep-
nications, Kliment Ohridski Blvd. 8, Sofia 1000, Bulgaria, E-mail: arated by narrow gaps. If the length of the gaps is comparable
h yordanov@tu-sofia.bg with the wavelength, they can be treated as slot lines. If these

17
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA
RF generator
Antenna electrode Antenna electrode

2 mm 2 mm
1.1 mm

GND GND 1.5 µm

2 mm 50 µm

Fig. 3. Top view and current distribution of a two patch antenna, operating
t at 66 GHz.
8 µm Antenna input resistance
140 g=20
g=40
120 g=60
g=80
3 µm 100
Transistors Transistors g=100
80

Re{Z in}
Substrate 675 µm 60
or
6 µm 40

20

Fig. 1. Detailed view of the cross-section of the integrated on-chip antenna, 0

using the ground planes as antenna electrodes. The separated areas of the 30 40 50 60 70 80 90 100
ground planes have to be connected to each other using inductive connections. Frequency, GHz
The RF generator is also integrated in the CMOS circuit. Figure is not to scale.
Fig. 4. Real part of the input impedance of the antenna from Fig. 3 for
different slot widths g.

Fig. 5. Photograph of the manufactured antenna.

Fig. 2. Current distribution in an 2 × 2 patch antenna configuration.

The input impedance of the antenna depends on the gap

lines are terminated with a open or short-circuit, a standing width. A plot of the real part of the two-patch slot antenna
wave pattern will be formed along the line. This standing wave from Fig. 3 is shown in Fig. rin. The input resistance is smaller
provides a time-varying electric polarisation, which is a source for a narrower slot.
of radiation [11]. Figure 2 shows the current distribution in a The antenna has been manufactured on high-impedance
four-patch antenna configuration. substrate. A photograph of the structure is shown in Fig. 5.
The guided wavelength in a slot line is given by [12] The results of the investigation of that type of antennas are
presented in [5], [6].
λ0
λg = q , (1) III. A NTENNAS ON T HIN S UBSTRATE
1
2 (εr,Si + 1)
The ChipFilm technology provides the possibility of man-
where λ0 is the free-space wavelength and εr,Si is the relative ufacturing integrated circuits on substrate with thickness as
permeability of the substrate. For a carrier frequency of low as 6 mum, as discussed in the introduction. Numerical
f = 60 GHz the corresponding wavelength is λg = 1.8 mm. experiments have been performed to assess the radiation
Therefore an open-circuited slot 1.35 mm long is a 3/4λg efficiency of antennas on thin substrate.
resonator. Simulations of a two-patch structure, presented in The antenna loss and the antenna radiation are both mod-
Fig. 3 show that the length of the slot should be a bit shorter, elled as resistors connected in series in the equivalent circuit
namely 1.1 mm, accounting for the effective slot elongation of the antenna. Therefore it is not possible to identify the
due to the stray capacitance of the open slot line end. The antenna efficiency by investigation of the input impedance.
current distribution, plotted in the figure, shows variation in The efficiency can be computed numerically by comparing
the current density along the gap. the quality factor of a lossless antenna model and the one of

18
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Antenna
Feeding network

Fig. 8. Return loss of the lossy integrated antenna.

Fig. 6. A top view of the experimental setup.

Fig. 9. A 3D model of the balun transformer.

Fig. 7. A photograph of a short-circuited antenna.

IV. E XPERIMENTAL S ETUP

a lossy antenna. The quality factor of any two-port is defined
The model of antennas on thin substrate is to be verified
as
by measurement. An experimental setup is prepared, which
PActive 1
Q= = , (2) allows the measurement of the input impedance of the slot
2πEStored ∆f antenna. The setup contains four antennas, each equipped with
or as a ratio between the active input power to the stored a feeding network, as shown in Fig. 6. There are two types
energy per cycle, which is inversely proportional to the −3 dB of antennas that will be investigated. The first type is the
bandwidth. Since the antenna input power is the sum of the slot antenna, described in the previous section. The second
radiated power and the power lost as heat, or type is the same slot antenna, but short-circuited at the end.
Such a short-circuit provides a DC connection between the two
PActive = Prad + Ploss , (3) patches, as they need to serve as a CMOS circuit ground plane.
A photograph of such antenna is shown in Fig. 7. Various
and the antenna efficiency is the ratio of the radiated to the antenna lengths for both antenna types have been designed.
input power, The feeding network contains a balun transformer, as the
antenna is symmetrical, so that the feeding line should be
Prad
ηAnt = , (4) balanced, whereas the measurement equipment provides un-
PActive balanced port connections. The balun is designed using a
we can compute the antenna efficiency as the ratio of the transformer. The offered technology provides two metallisa-
−3 dB bandwidth of a lossless and a lossy antenna, tion layers, so a stacked transformer design has been selected,
as the stacked transformers provide lower insertion loss than
∆flossless the single-layered ones [13]. Figure 9 shows a 3D model of
ηAnt = (5)
∆flossy the designed balun.
The calibration of the measurement system must be per-
The simulation results show antenna efficiency of ηAnt = formed using the Throuh-Reflect-Line (TRL) [14] technique
74.8 %. A plot of the return loss of the lossy antenna is shown in order to de-embed the balun characteristics from the mea-
in Fig. 8. surement results. This technique allows the calibration of

19
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

two-port measurement systems using three different, but not [3] C. Cao and K. K. O, “A 140-GHz fundamental mode voltage-controlled
standardised, port loads: direct connection of the measurement oscillator in 90-nm CMOS technology,” IEEE Microw. Wireless Com-
pon. Lett., vol. 16, no. 10, pp. 555–557, Oct. 2006.
ports at the reference plane (Through), connection of high- [4] T. Yao, M. Q. Gordon, K. K. W. Tang, K. H. K. Yau, M.-T. Yang,
reflection-factor load (Reflect) and connecting the ports with P. Schvan, and S. P. Voinigescu, “Algorithmic design of CMOS LNAs
a short line (Line). and PAs for 60 GHz radio,” IEEE J. Solid-State Circuits, vol. 42, no. 5,
pp. 1044–1057, May 2007.
Therefore calibration structures have also been designed on [5] H. Yordanov and P. Russer, “Integrated on-chip antennas for chip-to-chip
a different chip. The structures allow for the measurement of communication,” in Proceedings of the IEEE Antennas and Propagation
the input impedance of the antennas at their input ports, while Society International Symposium, 2008, San Diego, CA, 2008.
[6] ——, “Integrated on-chip antennas using CMOS ground planes,” in Pro-
de-embedding the influence of the balun transformers. ceedings of the 10th Topical Meeting on Silicon Monolithic Integrated
Circuits in RF Systems, New Orleans, LA, 2010, pp. 53–56.
V. S UMMARY [7] P. Russer, N. Fichtner, P. Lugli, W. Porod, J. Russer, and H. Yordanov,
This work discusses the design of on-chip integrated anten- “Nanoelectronics-based integrated antennas,” IEEE Microwave Maga-
zine, vol. 11, Dec. 2010.
nas with high radiation efficiency and low area requirements. [8] H. Rempp, J. Burghartz, C. Harendt, N. Pricopi, M. Pritscow, C. Reuter,
There are two possibilities for minimising the antenna losses— H. Richter, I. Schindler, and M. Zimmermann, “Ultra-thin chips on foil
one is using high-resistivity substrate and the other is using for flexible electronics,” in Proc. of IEEE Intl. Solid-State Circ. Conf.,
2008, San Francisco, CA, Feb. 2008, pp. 334–617.
very thin substrate. An experimental setup has been prepared [9] H. Richter, H. Rempp, M.-U. Hassan, C. Harendt, N. Wacker, M. Zim-
to verify the characteristics of integrated antennas on thin sub- mermann, and J. Burghartz, “Technology and design aspects of ultra-thin
strate. The antenna input impedance and radiation efficiency silicon chips for bendable electronics,” in Proc. of IEEE Intl. Conf. on
IC Design and Technology, 2009, Austin, TX, May 2009, pp. 149–154.
have been computed. [10] J. Burghartz, W. Appel, C. Harendt, H. Rempp, H. Richter, and M. Zim-
mermann, “Ultra-thin chips and related applications, a new paradigm in
ACKNOWLEDGEMENT silicon technology,” in Proc. of ESSCIRC, 2009, Athens, Nov. 2009, pp.
This work has been supported by the Marie Curie Pro- 28–35.
[11] P. Russer, Electromagnetics, Microwave Circuit and Antenna Design for
gramme. Communications Engineering, 2nd ed. Nordwood, MA: Artec House,
2006.
R EFERENCES [12] S. B. Cohn, “Slot line on a dielectric substrate,” IEEE Trans. Microw.
[1] H. Yordanov, M. T. Ivrlač, A. Mezghani, J. Nossek, and P. Russer, Theory Tech., vol. 17, pp. 768–778, Oct. 1969.
“Computation of the impulse response and coding gain of a digital [13] T. O. Dickson, M.-A. LaCroix, S. Boret, D. Gloria, R. Beerkens,
interconnection bus,” in 24th Annual Review of Progress in Applied and S. P. Voinigescu, “30-100-GHz inductors and transformers for
Computational Electromagnetics ACES, Niagara Falls, Canada, Apr. millimeter-wave (Bi)CMOS integrated circuits,” IEEE Trans. Microw.
2008. Theory Tech., vol. 53, pp. 123–133, Jan. 2005.
[2] K. K. O, K. Kim, B. A. Floyd, J. L. Mehta, H. Yoon, C.-M. Hung, [14] D. M. Pozar, Microwave Engineering, 3rd ed. Ney York, NY: John
D. Bravo, T. O. Dickson, X. Guo, R. Li, N. Trichy, J. Caserta, W. R. B. Wiley & Sons, 2005.
II, J. Branch, D.-J. Yang, J. Bohorquez, E. Seok, L. Gao, A. Sugavanam,
J.-J. Lin, J. Chen, and J. E. Brewer, “On-chip antennas in silicon ICs
and their application,” IEEE Trans. Electron Devices, vol. 52, no. 7, pp.
1312–1320, Jul. 2005.

20
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

SCP-RPSC - The New Technology for Microwave

Broadband Mobile Communications
Veselin Demirev1
Abstract – A retrospective review of a new technology, named different space directions, using simple one-channel
SCP-RPSC, is given in this report. It is particular useful as the receiver and signal processing techniques.
radio front end of the mobile microwave broadband The objectives stated above are achieved by a patented
communication systems with terrestrial and satellite positioning. method for radio communications, which proposes application
A list of possible applications is given too.
of additional pilot signal transmitted in the band of
Keywords – SCP, RPSC, RPSC-MA, broadband microwave information signals and available in the receiver by one of the
mobile communications. known methods for access. The SCP receiver terminal is
equipped with antenna array with random phase aperture
excitation. The phase shifts among the signals, coming from
the antenna elements, are random at the antenna output,
I. INTRODUCTION regardless of the information source direction. These random
phase spread signals correlate with the recovered pilot signal,
The terrestrial and satellite broadband mobile phase spread in the same manner, in a signal recovery unit
communications are currently a strong growth market, driven (Fig.1). The result of the correlation process between pilot and
chiefly by major projects to deploy vast regional or world- information signals is the recovered information signal at base
wide networks. They need new and wider frequency bands, band.
available at higher frequencies – up to millimetre wave
frequency bands. One of the biggest technical problems of
these mobile networks is the way of access to the satellite or
terrestrial base stations, particular the used antenna systems.
The need to change the polarization, to track Low Earth
Orbiting Satellites (LEO,s) or High Altitude Platforms
(HAPS), to select one of several Geo Stationary Orbit
Satellites (GEO,s) positions, as well as the requirements for
two way broadband mobile communications at low price and
mass market production leads to unsolved by traditional
antennas problems. Their solution needs entirely new
approach, which is subject of the last decade research activity
of the author. The name of the new technical solution is
Spatial Correlation Processing – Random Phase Spread
Coding (SCP-RPSC). A retrospective review of the research
step by step approach, used by the author, is given in the
report with the main bibliography for details.

II. SCP TECHNOLOGY Fig.1. Block scheme of a SCP-CDMA system

The main objectives of the SCP technology [1, 3, 5, 6, 7, One of the main parts of the SCP system is the random
10, 20, 21] are: phased antenna. In principle all kind of antenna arrays could
• To receive one or more radio signals coming from be used, but for Ku and Ka bands particular suitable is the
one or several spatially distributed sources (satellites, Radial Line Slot Antenna (RLSA). Until now it is used as
base stations), insuring high gain of the antenna systems phased array for fixed satellite reception.
and using fixed or mobile receiving terminals, equipped The main features of the SCP approach are:
with SCP signal processing systems. • Simple, cheap and flat passive radial line antenna,
• To ensure spatial selectivity high enough to cancel suitable for mass production in Ku and Ka frequency
the same frequency channel interference, coming from bands.
• One channel convenient microwave receiver with
1
Veselin Demirev is with the Faculty of Telecommunications at simple signal processing.
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, • Omni directional for the cooperative satellite or
Bulgaria, E-mail: demirev_v@tu-sofia.bg. terrestrial base station, but with high figure of merit G/T.
• Selection of different satellites and polarizations by
PN-codes.

21
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

• Applications in existing S-DVB systems with minor microwave receivers, leading to low active jamming
modifications of the ground transmitters, compatible with probability.
the existing satellite transponders. • The SCP-RPSC approach could be a breakthrough
• Multi-beam and soft handover features. technology, leading to unpredictable increase of the
frequency reuse factor in satellite and terrestrial
wideband networks. Close situated subscriber
III. RPSC TECHNOLOGY terminals could communicate with base stations,
using the same frequency channel without
The idea to use SCP principle in transmit mode [2, 9, 15, interference. The isolation between the terminals will
16] was born during the SCP project research. The be provided by their specific random phase spread
transmitting antennas, as well as the receiving random phase coding.
antenna arrays in SCP technology are pure passive, without
any active or nonreciprocal elements. The specific SCP
processing is situated in the receiver (Fig.2). According to the IV. SCP-RPSC APPLICATIONS
basic electromagnetic antenna lows the replacement of the
passive transmitting antenna with passive random phase A. Satellite Digital Video Broadcasting (DVB-S)
antenna array in the transmitter, and vice versa in the receiver
should not change the system working principles and system Proposal for a SCP-CDMA GEO satellite system,
parameters. The transmitted by the random phase antenna suitable for DVB-S communications in Ku-band for fixed and
array signals have specific phase spread. It can be considered mobile terminals, is given in [4]. The proposed algorithm for
as random spatial coding. system parameters evaluation, based on link budget
calculations, gives good results – Figure of merit (G/T) better
than 14 dBi/K for 60 cm antenna diameter at very low prize
(in order of several $). Similar proposal for quasi GEO
satellites at elliptical polar orbits is given in [8].

B. Space Links

• Inter Satellite Links (ISL)

The space segment of the future global satellite systems

for broadband communications can be designed in number of
ways, depending on the orbital type of the satellites and the
payload technology available on board. The use of different
satellite orbits to provide complementary services, each
optimized for the particular orbital type, is certainly feasible.
Satellites can be used to connect with each other and the
ground networks, through the use of Feeder Lines, Inter –
Satellite Links or Inter-Orbit Links, which when combined
Fig.2. Block scheme of a RPSC system
with on-board routing facilities, can be used to form a network
The main features of the RPSC approach are: in the sky. The unique properties of the SCP-RPSC approach
• Providing of full duplex radiocommunication system will give a new support for the future broadband LEO,s
with one simple and cheap transmit-receive antenna. communication systems in the service feeder lines, inter-
satellite and inter-orbit lines domain. The possible
• The transmitted random poly-phase spread signals
applications of the SCP-RPSC technology in these microwave
will not cause significant harmful interference to the
lines of several different types LEO,s constellations are
conventional satellites, using the same frequency
considered in the report [18]. A review of the possible
channels. The interference will be similar to that,
advantages, supported by a critical analysis, is given too.
caused by the sidelobes of an antenna array with
random elements distribution.
• Feeder Lines
• The transmitted random poly-phase spread signals
are uniformly radiated in the space above the
The company O3b, supported by Google, is building a
antenna. Several satellites, equipped with the same
new network of MEO satellites with steerable Ka-band beams
SCP receivers and providing space diversity, receive
to provide lower-cost, fibber-grade access for cellular
them. The knowledge of the receiving satellites
backhaul and IP backbone trunking in traditionally
positions for the transmitting equipment is not
underserved areas. The main O3b network parameters and
necessary (as it is for a conventional satellite earth
architecture are presented in the report [32]. Information
station).
concerning the satellite constellation and orbit, satellite and
• The transmitted random poly-phase spread signals
ground antenna beam steering, as well as inter-satellite
have low detection probability for the conventional
handover procedures are given. The possible applications of

22
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

SCP-RPSC technology in O3b MEO satellite system are

discussed too. A review of the mobile satellite communication systems
with military applications is given in report [11]. The profits
• SPS Lines of SCP-RPSC technology for such kind systems are listed and
analyzed too.
One of the most important use of satellite technologies
in the future will be in Solar Power Satellites [SPS]. The F. Telemedicine
concept of generating solar power in space for wireless
transmission to receivers on the ground has been discussed in The benefits for telemedicine systems, using SCP-
details during the last four decades. All of the sophisticated RPSC mobile satellite communications, are given in the report
SPS systems will need broadband wireless communications [12].
for telemetry and control purposes among the different parts
of their architectures. Another important problem of the future G. Aeronautical and Global Navigation Satellite Systems
SPS systems will be the transmission of video and telemetric (GNSS)
information among SPS mounting robots and satellite or
ground based control centers. The unique properties of the The benefits for aeronautical systems and GNSS, using
SCP-RPSC approach will give a new support for the future SCP-RPSC mobile satellite communications, are given in the
SPS mobile broadband communication systems [17]. reports [22, 23, 30].

C. High Altitude Platform Systems

A new radio technology to realize the last mile access to

V. IMPROVEMENTS OF REGULATORY STATUS OF
the broadband fixed networks, named High Altitude Platform SATELLITE SERVICE USING VEHICLE-MOUNTED
Systems (HAPS), is discussed in the reports [14, 24, 29]. Such ANTENNAS AND RPSC MULTIPLE ACCESS
a mode of service delivery offers advantages as coverage can TECHNIQUES
be rapidly set-up over any location and can be just as easily
removed or relocated; high elevation angles can be achieved Satellite connectivity while driving traditionally has been
to the mobile users; efficient frequency re-use schemes can be possible by using handheld personal terminal equipment with
employed to maximize network capacity; the round-trip delay low gain omni directional antennas. Recently, the new
is relatively short; the cost is considerably less than terrestrial satellite interactive broadband communication systems use
or satellite counterparts. The goal of the reports [14, 24, 29] is high gain satellite tracking antennas, installed on vehicles.
to discuss the possibilities and the advantages of the Vehicle-Mounted Earth Stations (VMES) currently can
implementation of SCP-RPSC technology in HAPS operate on conventional Ku-band frequencies (14 GHz
communications, particularly as subscriber terminal front end Uplink, 11-12 GHz Downlink) but only on a secondary basis.
equipment. The proposals deal with Line of Sight (LOS) mm- This means VMES can not claim interference protection from
wave propagation environment, which is accepted by the primary services such as fixed satellite systems and Earth
communication community as the only way to communicate Station on Vessels (ESV). A co-primary allocation of VMES
in these frequency bands. However, in high building city in the conventional Ku-band would be in the public interest,
environment most of the terminal links will be shadowed, as it would address a growing commercial demand for on the
which will need more and more new base stations. A possible move services. However, a co-primary allocation would also
solution is the Non-LOS mm-wave systems, working properly have to be conditioned on strict adherence to interference
in high building city environment. The possibilities of SCP avoidance mechanism, which in the best way obviously is
technology to create simultaneous several narrow virtual satisfied by the RPSC technology [26, 28].
antenna beams could be a good solution of the problem, The Radio-communication Sector of ITU is now seeking
leading to effective use of the reflected beams by gathering submissions from industry and governments on various
the signals in phase at baseband [31]. technical, regulatory and economic ideas in order to increase
the efficient use of satellite orbits and frequencies. The SCP-
D. WIMAX RPSC could be a breakthrough technology, leading to
unpredictable increase of the frequency reuse factor in
The goal of the reports [19, 25] is to discuss the satellite and terrestrial wideband networks. Close situated
possibilities and the advantages of the implementation of subscriber terminals could communicate with terrestrial or
SCP-RPSC technology in Wi-MAX communications. The satellite base stations, using the same frequency channel
implementation of this technology in subscriber terminals is without interference. The isolation between the terminals will
discussed first. After that the possible base station applications be provided by their specific random phase spread coding, due
are treated too. The applications of SCP-RPSC technology to their specific random design. We can consider this principle
simultaneous at base station and terminal stations are possible, of operation as a new multiple access approach, named by us
but they will need additional research and investigations. Random Phase Spread Coding - Multiple Access (RPSC-MA)
[27].
E. Military applications

23
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

VI. CONCLUSIONS [16] V. Demirev,, “SCP-RPSC technology – the new challenge in the
broadband satellite communications”, Caspian Telecoms,07,
Conference Proceedings, pp.178-181, Istanbul, Turkey, 2007.
The practical implementations of SCP-RPSC principles will [17] V. Demirev, “Application of SCP-RPSC Mobile
drastically change the existing paradigm in the mobile Communications in SPS Technology”, ISRSSP,07, Conference
microwave broadband satellite and terrestrial communication Proceedings,, pp. 129-132, Sofia, Bulgaria, 2007.
business in general. Many of the existing problems of the [18] V. Demirev, “SCP-RPSC Technology in the Feeder Lines of the
proposed systems, dealing with frequency and orbital resource LEO,s Communication Systems”, CEMA,07, Conference
sharing, beam pointing, beam shadowing, terrorist jamming Proceedings, pp. 1-5, Sofia, Bulgaria, 2007.
etc., will be solved successfully. [19] В. Демирев, “Приложение на технологията SCP-RPSC в
микровълновите Wi-Max системи”, Телеком, 07, Трудове на
конференцията, 29, Варна, България, 2007.
REFERENCES [20] V. Demirev, “Review of SCP Test Set-up and Results – I”,
CEMA, 08, Conference Proceedings, pp.7-10, Athens, Greece,
[1] V. Demirev, “Method and System for Space Diversity 2008.
Communications”, Patent WO/2003/013022, [21] V. Demirev, “Review of SCP Test Set-up and Results – II”,
PCT/BG2002/000016, 3.02.2003, H04B1/07 (2006.01), H04B CEMA, 08, Conference Proceedings, pp. 11-13, Athens, Greece,
7/02 (2006.01), Priority №105671/04.07.2001,BG. 2008.
[2] V. Demirev, “Radiocommunication Method and System with [22] В. Демирев, “Приложение на технологията SCP-RPSC в
poly-phase spread coding”, Patent WO/2003/107552, широколентовата мобилна спътникова система MOWGLY”,
PCT/BG2003/000027, 27.12.2003, H04B1/69 (2006.01), H04B Телеком, 08, Трудове на конференцията, стр.121-126 ,
7/06 (2006.01), Priority №106819/13.06.2002, BG. Варна, България, 2008.
[3] V. Demirev, V., “SCP technology – the new challenge of [23] В. Демирев, “Приложение на технологията SCP-RPSC в
broadband satellite communications”, ICEST’04, Conference широколентовите спътникови комуникации за въздушния
Proceedings, pp. 159-162, Bitola, Macedonia, 2004. транспорт”, Бултранс,10, Трудове на конференцията, стр.
[4] V. Demirev, A. Efremov “SCP-CDMA GSO’s system 34-37, Созопол, България, 2010.
proposal”, ICEST’04, Conference Proceedings, pp. 163-166, [24] В. Демирев, “Основни характеристики на микровълновите
Bitola, Macedonia, 2004. HAPS системи, базирани на технологията SCP-RPSC”,
[5] V. Demirev, “The probability theory with application in SCP Телеком, 10, Трудове на конференцията, стр.172-179 ,
technology”, ICEST’04, Conference Proceedings, pp. 167-168, София, България, 2010.
Bitola, Macedonia, 2004. [25] V. Demirev, “SCP-RPSC WI-MAX systems – the new
[6] B. Демирев, В. Витков, С. Kaменополски, A. Eфремов, approach for the next generation microwave access strategies”,
“Изследване на някои параметри на технологията SCP с Caspian Telecoms,08, Conference Proceedings, Istanbul,
МАТLAB симулации”, Телеком,04, Трудове на Turkey, 2008.
конференцията, том 1,стр.321 -326, Варна, България, 2004. [26] V. Demirev, “SCP-RPSC Technology – The Perfect Solution of
[7] V. Demirev, A. Efremov, “Some Important Parameters of the the Broadband MSS Problems”, Computer and Communication
SCP Technology”, ICEST’05, Conference Proceedings, V.2, Engineering, vol.4, N1, pp.43-46, 2010.
pp.516-518, Nis, Serbia and Montenegro, 2005. [27] V. Demirev, “Random Phase Spread Coding Multiple Access -
[8] R. Markov, V. Demirev, “Application of SCP technology in the New competitor of CDMA in the broadband wireless
Quasi-GEO satellite systems”, ICEST’05, Conference networks”, Journal of Applied Electromagnetism, vol.13,
Proceedings, V.2, pp.519-521, Nis, Serbia and Montenegro, Number 1(June 2011), pp. 26-32, 2011.
2005. [28] V. Demirev, “The regulatory aspects of SCP-RPSC technology
[9] V. Demirev, “Review of SCP-RPSC technology”, ICEST’05, – could they solve the VMES problems”, Journal of Applied
Conference Proceedings, V.2, pp.630-633, Nis, Serbia and Electromagnetism, vol.13, Number 1(June 2011), pp. 33-38,
Montenegro, 2005. 2011.
[10] B. Демирев, Д. Илиева, “Информационен капацитет по [29] В. Демирев, “Широколентови радиокомуникационни
Shannon на SCP система, ограничена по топлинен шум”, системи, базирани на летателни апарати”, Бултранс,2011,
Телеком, 05, Трудове на конференцията, том 1, стр.95-100, Трудове на конференцията, стр. 43-46, Созопол, България,
Варна, България, 2005. 2011.
[11] B. Демирев, “Спътниковите комуникации – тенденции и [30] В. Демирев, “Приложение на технологията SCP в системите
приложения”, CIO Journal, 11, стр.34-38, 2006. за спътникова навигация”, Бултранс,2011, Трудове на
[12] V. Demirev, “Broadband satellite SCP-RPSC Communications конференцията, стр. 47-50, Созопол, България, 2011.
– the new chance for the telemedicine”, CEMA, 07, Conference [31] V. Demirev, “SCP Rake Receiver - The Possible Solution of N-
Proceedings, pp. 17-21, Sofia, Bulgaria, 2006. LOS HAPS and WIMAX mm-Wave Networks”, CEMA,11,
[13] В. Демирев, A. Eфремов, E. Mихайлова, “Информационен Conference Proceedings, pp. 18-21, Sofia, Bulgaria, 2011.
капацитет по Shannon на ограничена по интерференция SCP [32] В. Демирев, “Приложение на технологията SCP-RPSC в
система”, Телеком,06, Трудове на конференцията, стр.272- спътниковата МЕО система О 3 b на Google”, Телеком,11,
277 , Варна, България, 2006. Трудове на конференцията, София, 2011.
[14] B. Демирев, B., M. Hиколова, “Приложение на технологията
SCP-RPSC в широколентови HAPS”, Телеком,06, Трудове
на конференцията, стр.283-288 , Варна, България, 2006.
[15] В. Демирев, “Приложение на вероятностната теория в
технологията SCP-RPSC”, Телеком,06, Трудове на
конференцията, стр.278-282, Варна, България, 2006.

24
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Study on Hybrid FSO/RF Systems Availability

Depending on the Meteorological Conditions
Tsvetan Mitsev1, Maxim Shupak2 and Boncho Bonev3
Abstract – The investigation for increasing the reliability of This paper presented analytical expressions and results for
operation of an FSO communication system through its estimation of the power budget of systems, operating in
integration into a hybrid with a wireless radio frequency system optical wavelength and millimeter frequency ranges. The
is done. By using numerical experiments, the systems’ potentialities of these systems are compared by using a
capabilities are compared, depending on the meteorological
conditions. An algorithm for switching the constitutive systems,
numerical experiment. An algorithm for switching of FSO and
depending on the specific meteorological condition is presented. RF subsystems depending on meteorological conditions is
It is based on the received power of each subsystem (FSO and suggested. Numerical and graphic results of the investigation
RF) of the hybrid FSO/RF communication system. are presented.

Keywords – Free space optics, FSO, radio frequency, hybrid

systems, telecommunications. II. THEORETICAL ANALYSIS
The electromagnetic waves (EM) are absorbed in
I. INTRODUCTION atmospheric gases, especially in water molecules, carbonic
dioxide, oxygen and ozone, when they propagate in Earth’s
In the last decade the Free Space Optics (FSO) atmosphere. For decrease the influence of this effect the
communication systems became alternative of the fibber optic wireless communication systems use frequencies
systems. In these years the technological progress leads to (wavelengths) situated in so-called “transparency windows of
their serious development that allows their wider usage. The the atmosphere”.
four channel FSO system works successfully on distance of The EM waves are also absorbed and scattered quite
more than 28 km with data rate 2.5 Gbps per channel or 10 frequency selectively by atmospheric aerosol, anthropogenic
Gbps overall [1]. or natural, like fog, cloud, smoke and hydrometeors (rain,
In USA, more than 90 % of business costumers are situated snow etc.). The attenuation for some wavelengths is so great,
on about one mile distance from some high-speed fiber-optic that sometimes the connection can interrupt completely even
network, but only 19% have access to it. That is mainly if the link distance is short. When a hybrid communication
caused by high cost of new networks installation. This gives system is used its link availability has to be higher than of
to FSO systems very big potential market. Their wider usage each containing subsystem.
in telecommunications, for example by Internet providers, is There are some main physical reasons for higher
limited by relatively low and dependent on distance link functionality of hybrid FSO/RF communication systems. In
availability. That low availability is connected with frequency range above 10 GHz radio waves are hardly
comparatively rapid and casual changes of atmospheric influenced by the hydrometeors and mainly by the rain,
transparency that are due to unpredictable changes of because it can be more intensive by the snow. Because rain
meteorological conditions during the time. If it is found a way drops size (0,1 mm to 7 mm) is in the same order as
to increase link reliability of FSO from typical 97-98 % to at wavelength, the radio wave attenuation in rain is considerable.
least 99.9 %, the problem will be solved. The link availability of a FSO system is influenced mainly by
Besides FSO systems, that operate in visible and near the fog, what is closely related with visibility Sm [2, 3]. The
infrared diapason, the wireless communication systems use fog particles have size of the order of microns and part of
millimeter radio waves also. A practical solution for microns (typical 1 µm to 20 µm), i.e. they are commensurable
increasing link availability of FSO is its integration whit a to optical radiation wavelength and therefore the optical wave
radio frequency (RF) communication system in a hybrid attenuation in presence of fog is hardly significant. The
FSO/RF system. The radio subsystem has lower data rate, but probability of simultaneously presence of fog and rain is very
has higher stability against bad meteorological conditions like small, for that reason it can be expected that for a hybrid
thick fog. FSO/RF communication system the disadvantages of its two
parts will compensate each other and the overall availability
1
Tsvetan Mitsev is with the Faculty of Telecommunications at of the system will increase significantly.
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, The functionality of the communication systems, working
Bulgaria, E-mail: mitzev@tu-sofia.bg. in wavelengths of millimeter and centimeter ranges, is
2
Maxim Shupak is with the Faculty of Telecommunications at determined by some factors as: the effective isotropic radiated
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000,
power – EIRP, that depends on transmitter power, feeder
Bulgaria.
3
Boncho Bonev is with the Faculty of Telecommunications at losses and antenna gain; propagation losses, including free
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, space losses, diffraction losses, attenuation caused by
Bulgaria, E-mail: bbonev@tu-sofia.bg. hydrometeors and gaseous molecules; receiver sensibility –

25
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

given as minimal power in the input of the receiver that  −0,5853 S m [ km ] 

guaranteed fixed value of BER (it is often chosen 10-6). 1  3,92  λ[µm]  
Φ PD = Φ Lτ tτ r exp− z[km]   .
The atmospheric gaseous absorption losses are well 2  S m [ km ]  0,55  
described in the literature and they depend mainly on
frequency [4]. The rain attenuation depends on frequency and
 R2 
rain rate and can be obtained by using the methods described 1 − exp − 2 2 r 
 ρ z (z ) 
in [5]. The scattering losses are due to troposphere’s scattering
.  , (5)
caused by troposphere’s density changes and rain scattering. 1 − exp(− 2 )
The power in the input of a radio receiver Pr [dBm] can be
calculated by expression where ΦL [W] is the laser power; τt and τr are respectively the
transmitter’s and the receiver’s optical systems transparency; z
Pr = Pt + Gt − Lt − Lf − La − Lr + Gr , (1) is the link distance; Sm is the meteorological visibility of the
atmosphere; λ is the wavelength; Rr [m] is the transmitter’s
where Pt [dBm] is the transmitter’s power; Gt (Gr) [dBi] –
transmitter (receiver) antenna gain; Lt (Lr) [dB] – transmitter’s antenna radius; ρz [m] is current Gaussian laser beam radius.
(receiver’s) tract losses; Lf [dB] – free space losses; La [dB] – When ρz is calculated the initial radius of laser beam ρ0, the
atmospheric losses (in atmospheric gases and hydrometeors) wavelength λ and the link distance z are used. The above
[4, 5]. mentioned correction is also applied.
The antenna gain G and free space losses can be calculated The signal-to-noise ratio in the output of the receiver can be
by relations calculated with taking in account quantum noises, thermal
noises and dark current of photo-detector by using the formula
4πηAg
G= (2) ( Φ PD . R D ) 2
λ2 SNR = . (6)
4. k B .T . ∆f
and 2.e . ∆f . I D + 2.e . ∆f . I S +
RL
2
 4πz  In eq. (6) RD [A/W] is the integral sensibility of photo-
Lf =   , (3)
 λ  detector; e [C] is the electron’s charge; ∆f [Hz] is the
frequency band of the photo-detector; ID [A] is the dark
where η is the coefficient of antenna area usage, that usually current of the photo-detector; IS [A] is the signal current; kB
has values from 0,55 to 0,75; Аg [m2] is geometrical area of
[J/K] is the Boltzmann constant; T [°K] is the temperature of
the antenna aperture; λ [m] is wavelength in free space; z [m] the detector; RL [Ω] is the load in the photo-detector’s circuit.
is distance between the transmitter and the receiver (link
In the graphics we use the logarithmical values of signal-to-
distance). noise ratio SNR [dB].
The signal-to-noise ratio SNR can be calculated by using
Very important part of the hybrid FSO/RF system
determined by Eq. (1) value of received power -
functionality is the procedure of subsystems switching and the
SNR = Pr − S r + A [dB] . (4) choice of a parameter and thresholds that will be used in this
switching. There are various variants. In dependence on the
We need to know the receiver’s sensibility Sr [dBm] and to chosen value of BER that is related to the applied modulation
choose a proper value of power reserve A [dB]. The sensibility technique, we can determine necessary SNR. Further by using
describes receiver’s internal noises for given frequency band this signal-to-noise ratio we can calculate the minimal value
(the manufacturer’s catalog value can be used). The power of the received power that guaranteed system’s functionality.
reserve is necessary for the cases when unpredictable losses The easiest way to realize a system, what takes decision for
appear. Usually its typical value is 10 to 15 dB. switching of subsystems is to use the received power. In this
The link budget for FSO system was described in our case there is not a need of additional signal transformations
previous works [6, 7]. In this work we present only final and calculations. Furthermore it is good to use two different
results what we use in the numerical experiments. For threshold levels and the level for switching RF→FSO has to
simplification of used mathematical model we assume that: be somewhat greater than the level for FSO→RF switching.
the region of Earth’s atmosphere that is part of the FSO In this way we avoid the more frequently subsystems
communication channel is homogeneous; the laser beam is switching. This can cause significant data losses, because of
with Gaussian distribution in transmitter’s aperture and the lower data capacity of RF system. After switching, the
conditions in the atmosphere are suitable for its preservation inactive subsystem has to be in waiting regime.
during the propagation; we introduce a correction for current The “International table for relation between visibility,
laser beam radius that takes account of additional extension of meteorological conditions and attenuation” [3] can be used for
the laser beam compared with theoretical; the noise due to the fast estimation of the attenuation in one hybrid FSO/RF
background radiation is neglected [9]; analog modulation of communication system in different weather conditions.
the laser optical beam is performed in the transmitter. Further the switching terms can be defined in one model
Therefore the average value of received signal optical FSO/RF system in dependence on the visibility [8]. The two
power in the input of photo detector is different thresholds can be defined.

26
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

In a hybrid FSO/RF system the only uncontrolled

variability in the link budget equation is the communication
channel transparency. It depends on concrete meteorological
conditions. A possible algorithm for switching and
functionality control of the subsystems is by estimation of this
variability. Then after comparison with previously given
values or with dynamic determined ones, obtained by
studying, the algorithm takes decision for switching. This is
the approach we choose, and the switching algorithm is given
on Fig.1.

Fig. 2. Hybrid FRO/RF system SNR dependence on the rain rate

Fig. 1. Subsystems switching algorithm

III. NUMERICAL EXPERIMENTS

For our numerical experiments we use the parameters’
values typical for the real communication system. We fix a
link distance z = 1,5 km because the typical hybrid
communication system operates on distances in order of 1 km
when the rain rate is from 2 mm/h to 100 mm/h or the Fig. 3. Hybrid FRO/RF system SNR dependence on the visibility
meteorological visibility is from 0,1 km to 20 km. When we
study the SNR dependence on the link distance z, we assume z
≤ 10 km in case of clear atmosphere (Sm = 20 km).
The other parameters are as follow: for the RF subsystem
we choose f = 2.4 GHz, with Pt = 5 dBm, Gr = Gt = 15 dBi,
Lr = Lt = 3 dB, Lf = 103,566 dB, Sr = − 90 dBm and
f = 60 GHz, with Pt = 10 dBm, Gr = Gt = 41 dBi,
Lr = Lt = 3 dB, Lf = 131,525 dB, Sr = − 85 dBm; for FSO
subsystem - λ = 532 nm, with ΦL = 0,5 mW, τt = 0,7, τr = 0,9,
Rr = 8 cm, ρ0 = 2 cm, θ = 2 mrad, Rd = 0,24 A/W, RL = 5 kΩ,
∆f = 1 GHz, Id = 1.5 nA and λ = 850 nm, with ΦL = 0,5 mW,
τt = 0,7, τr = 0,9, Rr = 8 cm, ρ0 = 2 cm, θ = 2 mrad,
Rd = 0,5 A/W, RL = 5 kΩ, ∆f = 1 GHz, Id = 2 nA. The
frequencies for RF are chosen because they aren’t object of
license.
Calculated values of SNR for two subsystems of the hybrid Fig. 4. Hybrid FRO/RF system SNR dependence on the link distance
FSO/RF system are shown on Fig. 2, 3 and 4.
Fig.2 shows that even very weak rain influences seriously Fig.4 shows that when the atmosphere is clear (Sm=20 km),
60 GHz RF subsystem functionality. In other way the FSO the FSO subsystem has serious advantage over the RF one
subsystem keeps its functionality even in heavy rain even for link distances from 4 km to 9 km.
conditions for two investigated wavelengths.
The drawings in Fig. 3 show that in fog conditions the FSO IV. RESULTS ANALYSIS AND CONCLUSIONS
subsystem performance decreases significantly without
dependence on wavelength even for visibility Sm in order of From applied drawings can be viewed that a heavy rain
1 to 2 km. In other hand the RF subsystem will work stable influences the FSO subsystem functionality much lower than
even when the visibility Sm is about 30 m, although in this the RF one. As expected the results show increase of optical
case SNR is relatively low. power through the photo detector’s aperture even in high

27
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

values of rain rate. When RF subsystem uses frequency of 2.4 ACKNOWLEDGEMENT

GHz the SNR is not affected both by the rain and fog but
despite this it stands relatively low. It varies significantly only This work was supported under Project Nr DUNK-01/03 –
with increases of distance. The calculations give us a 12.09 “University Scientific and Research Complex for
possibility to assess that in frequencies of 60 GHz the rain Innovation and Transfer of Knowledge in Areas of
attenuation is considerable, compared to other frequencies, Micro/Nano Technologies and Materials, Power Effectiveness
and it doesn’t allow high speed communication in long and Virtual Engineering”.
distances. However the fog attenuation in this frequency is
relatively low even when the visibility is very low. At the
same time, despite relatively low fog attenuation for REFERENCES
wavelength 850 nm according to 532 nm, we can’t speak for
independently working FSO system at these conditions. The [1] Johnston D., “Scientists complete laser link between Lab,
very high attenuation of optical waves in fog then the Mount Diablo,” Newsline, Lawrence Livermore National
Laboratory, 2002.
visibility Sm < 0,5 km, is obvious.
[2] Kolka Z., Wilfert O., Biolkova V., Reliability of Digital FSO
For distances up to 2 km RF system works satisfactorily Links in Europe, International Journal of Electronics, Circuits
with data transfer, but for communications on distances from and Systems, Volume 1, Number 4, p.236-240, 2007.
2 to 10 km, FSO systems are more suitable. We can remark [3] Willebrand H., Baksheesh G., Free-Space Optics: Enabling
that they show better functionality in high visibility Optical Connectivity in Today’s Networks, Sams Publishing,
conditions. Therefore a hybrid FSO/RF system will rely more Indianopolis, 2002.
on the FSO subsystem. The determination of concrete [4] ITU-R Recommendation P.676-7 (02/07) Attenuation by
intervals for work of each subsystem is object of further atmospheric gases
investigation. The desired system configuration depends on [5] ITU-R Recommendation P.838-3 (03/05) Specific attenuation
model for rain for use in prediction methods
the meteorological conditions and the wanted availability.
[6] Mitsev Ts., K. Dimitrov, B. Bonev, Influence of Laser Beam
It is obvious that a hybrid FSO/RF system contains parts Divergency on Free Space Optic Systems Functionality,
that compensate their disadvantages each other. Therefore the TELECOM 2008, 9-10 October, St. Constantine, Varna,
overall availability time for a hybrid system will be Bulgaria, 2008.
significantly greater than for each containing subsystem one [7] Mitsev Ts., K. Dimitrov, N. Kolev, Reliability Testing of Free
by one. Space Optical Systems in Laboratory Conditions, XLIV Intern.
In our work we have chosen to use a SNR as criterion for Scientific Conf. ICEST 2009, 26 to 27 June, Veliko Tarnovo,
link quality. This is motivated with our desire to avoid Bulgaria, 2009.
connection with used modulation technique. Nowadays when [8] Akbulut A., Ilk H.G., Arı F., Design, Availability and
Reliability Analysis on an Experimental Outdoor FSO/RF
digital systems are used, the more attractive quantity is BER,
Communication System, Ankara University, Faculty of
what allows development of this investigation. The theoretical Engineering, 2005.
model can be also developed with taking into account some [9] Bonev B., Relative Influence of Some Stochastic Factors on Bit-
specific effects and phenomena like type of fog – continental Error Rate of Ground-to-Ground Free Space Optics, XLII
or seaside for example. It is also possible to be studied a International Scientific Conference on Information,
hybrid system by using real statistical, meteorological data Communication and Energy Systems and Technologies, ICEST
and then to be formulated more accurately the interruption 2007, Vol. 1, pp. 203 – 206, June 2007, Ohrid, Macedonia.
intervals and link availability for the concrete geographical
region.

28
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Cylindrical Mesh TLM Model of Probe-Coupled Cavity

Loaded with Planparallel Dielectric Layers
Tijana Dimitrijevic1, Jugoslav Jokovic1 and Bratislav Milovanovic 1
Abstract – This paper describes an experimental verification of reflection and transmission characteristics are common
a compact wire model implemented into the 3-D TLM cylindrical parameters in the cavity exploration, input and output ports of
mesh for the purpose of an efficient analysis of a probe-coupled real microwave cavity devices are generally realized by
cylindrical microwave cavity loaded with planparallel dielectric coaxial probe that ensures coupling with corresponding
layers. An implementation of the compact wire model into a
cylindrical TLM mesh is based on wire structures parameters
electromagnetic (EM) field component [1]. For that reason,
calculation in conditions of variable cross-section of the TLM physical and electrical probe parameters form an integral part
nodes through which a wire conductor passes due to the nature of microwave technique studies regarding EMC
of a cylindrical grid along the wire path. Results reached by the (electromagnetic compatibility) problem of a probe-coupled
TLM based approach for characterization of a cylindrical cavity.
metallic cavity loaded with a planparallel dielectric of the The TLM enhancement in form of the compact model for
permittivity corresponding to the water have been compared wire structures has been developed [8, 9], yielding a
with the measured results. significant improvement in the required computer resources
compared to the traditional TLM method. The so-called TLM
Keywords – Cavity resonators, Electromagnetic analysis,
Dielectric layer, Probe antennas, Wire model, Cylindrical grid.
wire node has been implemented into the uniform TLM mesh
based on a rectangular grid, as mean cross-section dimensions
of nodes through which a wire conductor runs are always
constant, allowing to easily preserve distributed capacitance
I. INTRODUCTION and inductance of a wire per unit length. However, if a
rectangular uniform mesh is used to model a cylindrical
Extensive use of microwave energy in communication, structure [10, 11], a curved boundary would have to be
industry, science and medicine has led to the development of a described in a step-wise fashion which might result in a
number of different microwave devices based on microwave deviation of resonant frequency values as well as in excitation
metallic cavities [1-3]. Among them, the most popular ones of unwanted modes. A numerical error could be reduced by
are resonant applicators classified as either single or applying the TLM mesh of a higher resolution, which would
multimode cylindrical metallic cavities, partially loaded with result in increasing the simulation time. Moreover, a mesh
dielectric slabs, widely used in the processes of material resolution increasing is limited since an implementation of the
heating and drying. They come in various shapes and sizes compact wire model into the rest of the TLM mesh is based
based on the electromagnetic (EM) properties, geometry and on arbitrary ratio between radius of a wire and dimensions of
volume of dielectric materials. The knowledge of the mode nodes through which a wire conductor passes. Consequently,
tuning behavior in a cavity under loading condition (i.e. a higher resolution of an applied rectangular TLM mesh
physical and electrical parameters of the load) forms an enables a cylindrical cavity to be precisely modelled only if
integral part of the studies in microwave heating and can probes of a relatively small radius are used [10, 11].
significantly help in designing these applicators. These limitations of the rectangular TLM mesh for the
Electromagnetically-based numerical TLM (Transmission- purpose of the modelling of a probe-coupled cylindrical cavity
Line Matrix) time-domain method can be successfully used to were overcome by implementation of the TLM wire node into
investigate an influence of different EM and geometric the cylindrical TLM grid. This solution has enabled the
parameters of dielectric materials used as a load in microwave precise modelling of cylindrical cavity boundaries
cavity applicators on cavity’s resonant frequencies [4-7]. independently of a mesh resolution applied. However, in case
Desired mode distribution in the modelled cavity can be of probes that are radially placed, mean cross-section
established exciting a particular field component through an dimensions of cylindrical TLM nodes along the wire path are
impulse excitation. However, this way of setting up the variable from one node to another, leading to different wire
wanted TE or TM mode obviously differs from the network properties between nodes. For that reason, an
experimental procedure [7] where a probe, placed inside a additional connecting procedure for wire segments belonging
cavity, is used as an excitation. Consequently, numerical to TLM nodes with different cross-sections has been
results obtained in case of an impulse excitation would be implemented [12].
different from the experimental ones, in terms of resonant In this paper, an efficiency of the TLM wire model adapted
frequency values and an EM field level. Further, since the to cylindrical mesh has been verified on the example of a
probe-coupled cylindrical metallic cavity loaded with
1 planparallel dielectric layer placed at the arbitrary height from
Tijana Dimitrijevic, Jugoslav Jokovic and Bratislav Milovanovic are
a cavity bottom. The transmission coefficient based on the
with the University of Nis, Faculty of Electronic Engineering,
Aleksandra Medvedeva 14, Nis 18000, Serbia, coupling of two wire probes inserted into the cavity has been
E-mail: [tijana, jugoslav, bata]@elfak.ni.ac.rs considered numerically and experimentally.

29
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

II. MODELING PROCEDURE rLr , respectively, for a wire segment running along radial
direction are rCr  kCr rc and rLr  k Lr rc , where rc
In the conventional TLM time-domain method, an EM field represents a mean dimension of the node cross-section in r
 r  r 
strength in three dimensions, for a specified mode of

oscillation in a metallic cavity, is modelled by filling the field direction ( rc   i i1   z  / 2 , (where ri and ri+1 are
space with a network of link lines and exciting a particular  2 
field component through incident voltage pulses on lower and upper limits of the TLM wire node in radial
appropriate lines. An efficient computational algorithm of direction (Fig.3)), while rCr and rLr are factors empirically
scattering properties, based on enforcing continuity of the obtained by using known characteristics of the TLM network.
electric and magnetic fields and conservation of charge and Distributed capacitance and inductance per unit length,
magnetic flux [13], is implemented to speed up the simulation needed for modelling of wire segments, may be expressed as:
process. EM properties of different mediums in the cavity are
2 
modelled by using a network of interconnected nodes, a Cwr  , Lwr  ln rLr / rw  (6)
typical structure known as the symmetrical condensed node – ln rCr / rw  2
SCN [13]. Each node describes a portion of the medium where rw is a real probe radius.
shaped like a cubic (Cartesian rectangular mesh) or a slice An equivalent radius of the fictitious cylinder can be easily
(Non-Cartesian cylindrical mesh) depending on the coordinate kept constant along nodes column in a rectangular grid.
system applied. Additional stubs may be incorporated into the However, for a radial wire conductor in a cylindrical grid, as it
TLM network to account for inhomogeneous materials and/or is shown in Fig. 3, mean cross-section dimensions of TLM
electric and magnetic losses. nodes, through which a wire passes, vary making difficult to
When cylindrical structures are concerned, a non-Cartesian preserve distributed capacitance and inductance of a wire per
cylindrical mesh in the coordinate system (φ, r, z) can be used unit length. As result, admittance of the wire network link
for the modelling purpose. The coordinate system used and line, interposed over the existing network to account for wire
the port designations are shown in Fig. 1. Simulation proceeds presence, varies from one TLM node to another (Fig. 2).
exactly as for a SCN with stubs in a Cartesian grid. The only Therefore, an additional connecting procedure for wire
modification involves the calculation of stub parameters segments with different link-lines admitances has been
where account must be taken of the details of the new implemented into the existing TLM-based software [12].
geometry.
φj+1 Dφ j φj

w,inc + + w,inc
2Vi 2Vi+1
Vrpz - w -
Vrp Ve
z wire
Vznr Dri ri+1 Yiw w
Yi+1
ri
Vzn interface r
φ Region 1 Interface Region 2
Vnr
Vpz
a) b)
Vnz
Vpr Fig. 2. a) TLM nodes in rφ plane through which wire runs and b) an
interface between two nodes
Vzp y
r r
Vzpr 
Reflected voltages on both directions of the interface
Vrn
z
x
between nodes with different cross-section, which at the same
Vrnz time represent incident voltages respect to the node center for
the next time step, can be expressed as follows:
 Yw Yw
Vew  2Viw1,inc w i1 w  2Vi w ,inc w i w (7)
Yi  Yi1 Yi  Yi1
Fig. 1. A cylindrical SCN
Yi w  Yiw1 w,inc
Vi w,ref  Vi  Viw1,inc   Viw1,inc (8)
The TLM wire node in a cylindrical grid is based on a SCN Yi w  Yiw1
with one small modification in the form of additional link and Yi w  Yiw1 w,inc
stub lines interposed over the existing network to account for Viw1,ref  Vi  Viw1,inc   Vi w,inc (9)
increase of capacitance and inductance of the medium caused Yi w  Yiw1
by wire presence. The single column of TLM nodes, through where Vew is an equivalent voltage at the interface, Vi w,inc and
which a wire conductor passes, can be used to approximately Viw1,inc are the incident voltages.
form the fictitious cylinder which represents capacitance and
inductance of a wire per unit length [8]. Such compact wire model allows for simple incorporation
An equivalent radius of the fictive cylindre in a cylindrical of voltage/current sources and lumped loads and takes into
grid for calculating the capacitance and inductance, rCr and account the physical dimensions of wire probes [9],
determined only by TLM mesh resolution.

30
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

When modelling of cavities containing lossy loads is z z

concerned, implementation of losses in the TLM model is
carried out by introduction of stubs with losses in the nodes
a a
where scattering is going on. Stubs with losses may be
considered as infinitely long transmission lines, or
equivalently, as lines terminated with its characteristic h2 h2
impedance. They can be used to model either electric or h h
magnetic losses. In case of the symmetrical condensed node, d1 d2 d1 d2
stubs with losses are directly implemented in the scattering l h1 l h1
procedure, includinmg coupling with the corrsponding EM h2
field component. y y
If  ek and  mk represent effective electric and magnetic
x x
conductance, respectively, in k-direction, where k = (φ, r, z),
elements in the TLM node used for modelling of losses are Fig. 3. A probe-coupled loaded cylindrical cavity with
defined as: a) one water layer and b) two water layers
i j i j
Gek   ek , Rmk   mk (10) Obtained numerical results, representing transmission
k k coefficient in the frequency range of interest, have been
where (Δi, Δj, Δk) = (rΔφ, Δr, Δz). experimentally verified in both cases of a loaded cavity and
 ek  for variable lengths of probes. Figs. 4. and 5. present
Starting from  k*   o rk  j ,  k*  rk  j mk [13], it is
  comparative results in terms of varying of probe length and
possible to define a loss tangent at the appropriate frequency load conditions. Apparently, a very good agreement between
as: numerical and experimental results has been achieved,
  confirming that cylindrical mesh based TLM model can be
tan  ek  ek
, tan  mk  mk
(11) used for analyses of probe-coupled cavity loaded with
2 f 0 rk 2 f 0  rk
planparallel dielectric layers.
Finally, corresponding equations for reflected total voltages
and currents in corresponding direction have to be modified in
case of modelling of mediums with losses [13]. IV. CONCLUSION
This paper presents an efficiency of the compact wire
III. RESULTS AND ANALYSES model implemented into the 3-D TLM cylindrical mesh for
the purpose of the analysis of a probe-coupled loaded
The proposed TLM model based on the cylindrical grid and cylindrical microwave cavity. Due to cylindrical grid structure
enhanced with the TLM wire node has been used for and empirical nature of the compact model, this
modelling of a loaded probe-coupled cylindrical cavity with implementation has to take into account a change of wire
dimensions a = 7 cm and h = 14.24 cm, chosen to follow the parameters with a variable cross-section of the TLM nodes
experimental ones [7]. Wire probes, representing a feed and through which a radially placed wire conductor passes. The
receiving probe, were placed at the height l = 7.4 cm from the model accuracy has been experimentally verified on an
bottom of the cavity, along the radial direction and opposite to example of a probe-coupled cylindrical cavity containing one
each other (Fig. 3). In order to model real coaxial cable and two layers of water as a dielectric load, for different probe
characteristics, the probes were connected, through the TLM length values.
wire port, to the real voltage source and resistances of 50. In Considered configuration of a probe coupled cavity loaded
order to illustrate effect of probe length changing to EM field with dielectric layers placed at different heights from the
distribution, analyses have been carried out for different cavity bottom is of a great importance in a realization of
length of probes d1 = d2 = d = 4 and 5 cm, respectively. microwave resonant applicators, widely used for thermal
Two types of the loaded cavity have been considered. One processing of materials.
represents the cavity containing only one layer of water placed
at h1=10cm from the bottom of the cavity (Fig.3a), whereas
the other contains two water layers, one at the bottom and the ACKNOWLEDGEMENT
other at h1=10cm from the cavity bottom, (Fig. 3b). The
height of the each layer is h2=3cm. The permittivity of the This work was supported by Serbian Ministry of Education
water was taken into account (εr = 77 - j6). For modelling and Science within the project TR 32054.
purpose a cylindrical TLM grid (φr) = (3628) was used,
whereas in z-direction dimensions of nodes are different, due REFERENCES
to inhomogeneity of the medium inside the cavity. In order to
achieve time synchronization of network, the TLM node in [1] T. V. C. T. Chan, H. C. Reader, Understanding microwave
dielectric was set to be  r times less compared to the nodes heating cavities, Boston, London: Artech House, 2000.
[2] C. Balanis, Advanced Engineering Electromagnetics, New
dimension in the rest of the cavity filled with air. York, John Wiley & Sons, 1989.

31
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA
[3] S. Amnartpluk, C. Phongcharoenpanich, S. Kosulvit, and M. [8] V. Trenkic, The development and characterization of advanced
Krairiksh, "A power divider using linear electric probes nodes for TLM method, Ph.D. Thesis, University of
coupling inside conducting cylindrical cavity", Int. Symp. on Nottingham,1995.
Circuits and Systems 3, 2003, pp. 419-422. [9] V. Trenkic, A.J. Wlodarczyk, and R. Scaramuzza, "Modelling
[4] B. Milovanovic, N. Doncov, and A. Atanaskovic, "Tunnel type of coupling between transient electromagnetic field and
microwave applicator analysis using the TLM method", Proc. of complex wire structures", International Journal of Numerical
the 4th International Workshop on Computational Modelling: Electronic Networks, Devices and Fields, 12, 1999,
Electromagnetics in the Time Domain: TLM/FDTD and Related pp.257-273.
Techniques, CEM-TD 2001, Nottingham, United Kingdom, [10] J. Joković, B. Milovanović, and N. Dončov, "TLM analysis of
2001, pp.77-84. cylindrical metallic cavity excited with a real feed probe",
[5] B. Milovanovic and N. Doncov, "TLM modelling of the circular International Journal of RF and Microwave Computer Aided
cylindrical cavity loaded by lossy dielectric sample of various Engineering, John Wiley&Sons, USA, 16, 2006, pp.346-354.
geometric shapes", Journal of Microwave Power and [11] J. Joković, B. Milovanović, N. Dončov, "Numerical Model of
Electromagnetic Energy, A Publication of the International Transmission procedure in Cylindrical Metallic Cavity
Microwave Power Institute, VA, USA, 37, 2002, pp.237-247. Compared with Measured Results", International Journal of RF
[6] J. Joković, B. Milovanović, T. Ranđelović, "TLM Modelling of and Microwave Computer-Aided Engineering, Publication of
Microwave Applicator with an Excitation Through The the Wiley, 18, 2008, pp.295-302.
Waveguide", Microwave and Optical Computer Techonology [12] Bratislav Milovanović, Nebojša Dončov, Jugoslav Joković,
Letters, John Wiley&Sons, 48, 2006, pp.2320-2326. Tijana Dimitrijević, "EM Field Monitoring in Circular Cavity
[7] S. Ivkovic, B. Milovanovic, A. Marincic, and N. Doncov, Using Wire Compact Model Implemented in Cylindrical TLM
"Theoretical and experimental investigations of resonance Mesh", Proc. of the TELSIKS 2009 Conference, Niš, Serbia,
frequencies in loaded cylindrical microwave cavity", Proc. of October 7-9, 2009, pp. 339-342
the 3rd IEEE TELSIKS'97 Conference, Nis, Yugoslavia, 1997, [13] C. Christopoulos, The Transmision-Line Modelling (TLM)
pp.306-309. Method, Series on Electromagnetic Wave Theory, IEEE/OUP
Press, 1995.

Fig. 4. Transmission coefficient magnitude in the cylindrical Fig. 5. Transmission coefficient magnitude in the cylindrical
cavity loaded with one water layer cavity loaded with two water layers

32
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Implementation of pseudo random noise generator in

FPGA for Free Space Optics BER testing
Nikolay Kolev1 and Tzvetan Mitzev2
Abstract – The way to determine the quality of communication it depends of the case. The random bits or bytes are obtained
link is to measure its Bit Error Ratio (BER). BER most often is by reading registers in definite period of time. We decide to
measured by the so called brute force method. It is sending work with shift register method because of simple hardware
random bits through the system and calculating BER. The implementation. This technique has good statistical properties
devices which measure BER are called Bit Error Rate Testers.
and leads to very efficient hardware implementations.
They consist of pseudo random noise generator and comparing
scheme. This report presents the creation of FPGA based system FPGA are reconfigurable silicon chips which can be
for testing BER in Free Space Optics Systems. The paper offers programmed in various different hardware configurations [2].
block and principle realisation of PRNG algorithm. There is In this way we obtain completely different hardware
synthesized HDL code for Spartan 6 FPGA chip. Our generator implementations. The specific thing in this IC’s is the ability
is tested in real conditions with oscilloscope. There is shown to perform parallel processes unlike the MCU’s. Each
computer simulation and practical results. processing task is assigned to a different section of the chip,
and can function autonomously without any impact from other
Keywords – Bit Error Rate, Free Space Optics, Pseudo logic blocks. The performance of single part of the application
Random Noise Generator, FPGA
is not affected when we increase the number of processes. The
FPGA’s have a couple of advantages which are connected
I. INTRODUCTION with better performance, low cost and good reliability. Due to
the parallelism of these chips their computing power is almost
One of the most important parameter, for quality equal to the Digital Signal Processor’s computing power. The
assessment in telecommunication systems is BER. It is a response times of each I/O pin is faster.
measure of the percentage of bits that a system does not In this work FPGA implementation of PRNG is presented.
transmit or receive correctly. The experiments connected with This report is addition of our paper which presents
BER give us the information about the maximal bandwidth Experimental setup for BER measuring of Free Space Optical
and quality of the FSO communication link. The device System [3]. Due to the needs of increasing the throughput of
which measure BER is called Bit Error Rate Tester (BERT). our system we replace the existing discrete components based
There are a lot of companies that make BERTs such as generator with FPGA chip. The advantages of new system
Tektronix, Agilent, Rohde & Schwarz. Because of price of are: easy reprogramming, high frequencies (500+ MHz),
these commercial systems we decide to build our own BERT. minimizing the physical dimensions.
The main block of BERT is the pseudo random noise
generator which is explained in this work. II. SCHEMATIC DESIGN
Random number generators are mainly used in
telecommunications, cryptographic algorithms and
The block scheme of generator is shown in fig.1. The
authentication protocols. There are many different ways to
operating principle is explained in the next paragraph.
generate pseudo random bit sequence. Most popular are ADC
method , shift register method and software method [1]. The principle scheme of the pseudo random noise generator
ADC method can be achieved by feeding sinusoidal signal to can be seen on fig.2. It possible to consist of several D type
the ADC, while at the output of the IC there is a random flip flops connected in series, one XOR logic gate and one
digitized signal. With ADC can be achieved hi speeds, but fast
ADC’s are expensive. In Software method we are restricted
form computer’s communication ports. We choose to work
with shift register method because of the low price and hi
speed of FPGA. In this method we have a couple of D type
flip flops and one XOR logic element connected in series. The
clock generator moves the logic levels through the scheme. In
this generator the output is the last D type flip flop. In
software generators we increment fast one or couple registers,
Fig.1. Block scheme of random number generator
1
Nikolai Kolev is with the Faculty of Telecommunications at inverter. In shown case there are four D type flip flops which
Technical University, 8 Kl. Ohridski Blvd, Sofia 1000, Bulgaria, E- Q outputs are connected to D inputs of next flip flop. The two
mail: kolev@tu-sofia.bg. inputs of the logic gate XOR are connected to the last Q
2
Tsvetan Mitsev is with the Faculty of Telecommunications at output of the scheme, through inverter, and between two
Technical University of Sofia. E-mail: mitzev@tu-sofia.bg
middle flip flops.

33
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The four D type flip flops form 4 stages shift register. At use ieee.std_logic_1164.ALL;
every tick of the clock signal the logic level on every input is use ieee.numeric_std.ALL;
shifted one step left. In this situation the output of the XOR library UNISIM;
gate is a source of pseudo random bit sequence. The function use UNISIM.Vcomponents.ALL;
of inverter is to start the generator, because in the beginning at
every inputs and outputs there are only zeroes. The XOR gate
need different levels to produce ones at his outputs, that’s why entity tesALTYS is
there is inverter at the one of two inputs of the XOR gate [4]. port ( clk : in std_logic;
The period of repetition depends on number of D type flip Led5 : out std_logic;
flops. In particular case it can be calculated by formula 1, Led6 : out std_logic;
where “n” is number of D type flip flops. Led7 : out std_logic);
end tesALTYS;
2n  1 (1) …………..
end component;
In the shown scheme in fig.2 Feedback polynomial is (2), attribute CLKFXDV_DIVIDE of DCM_CLKGEN :
the period of repetition is 15 because 24 -1 = 15 component is "2";
x 4  x3  1 (2) attribute CLKFX_DIVIDE of DCM_CLKGEN :
component is "1";
For example if the number of bits are 16 then the period of attribute CLKFX_MD_MAX of DCM_CLKGEN :
repetition will be 65536, because 216-1 = 65535. If the component is "0.000";
numbers of bits are 19, the period of repetition is 524287 and attribute CLKFX_MULTIPLY of DCM_CLKGEN :
so on. So we can connect the necessary number of D type flip component is "4";
flops for desired period of repetition.
attribute CLKIN_PERIOD of DCM_CLKGEN :
For implementation of the generation algorithm it is
necessary to produce confirmation code for the FPGA chip. component is "0.0";
This code makes individual connections between the different attribute SPREAD_SPECTRUM of
parts in the chip. For code configuration tool we used Xilinx DCM_CLKGEN : component is "NONE";
ISE Web Pack tool which is freely accessed in Xilinx web site attribute STARTUP_WAIT of DCM_CLKGEN :
[5]. For top layer of our project we used schematic design component is "FALSE";
fig.2. attribute BOX_TYPE of DCM_CLKGEN :
component is "BLACK_BOX";
component FD
generic( INIT : bit := '0');
port ( C : in std_logic;
D : in std_logic;
Q : out std_logic);
end component;
attribute BOX_TYPE of FD : component is
"BLACK_BOX";
Fig.2. Principle realization of PRNG
component INV
port ( I : in std_logic;
O : out std_logic);
The HDL code is too big to be shown in paper, that’s why end component;
we will show a essential part of it in listing 1. In the beginning
attribute BOX_TYPE of INV : component is
of code there is definition of standard libraries like ieee.std,
ieee.numeric and so on. After that we have definition of inputs "BLACK_BOX";
and outputs of the chip. At the next lines there is the
configuration of the divider of the clock generator. After that component XOR2
we have defining inputs and outputs of the logic elements D port ( I0 : in std_logic;
flip flop, inverter and XOR gates of the scheme. At the end of I1 : in std_logic;
the code there is shown interconnections between the logic O : out std_logic);
elements and I/O pins. end component;
…………..
Listing1 HDL Code for Spartan 6 board. begin
Led5 <= Led5_DUMMY;
library ieee;
Led6 <= Led6_DUMMY;

34
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Led7 <= Led7_DUMMY; For our development we used evaluation board Altys
XLXI_7 : DCM_CLKGEN Spartan 6 from DIGILENT (fig.3) [6] . It is very powerful
………… board with rich peripheral devices such as LAN interface unit,
………… a couple of switch units and micro switches, LED’s, USB,
UART, HDMI, external RAM memory. We choose to work
XLXI_8 : FD
with ALTYS because of it high operating frequencies
port map (C=>XLXN_51, 500+MHz and existence of Gigabit Ethernet.
D=>XLXN_5,
Q=>XLXN_4);

XLXI_9 : INV
port map (I=>XLXN_4,
O=>XLXN_5);

XLXI_10 : FD
port map (C=>XLXN_4,
D=>XLXN_7,
Q=>XLXN_6);
XLXI_11 : INV
port map (I=>XLXN_6,
O=>XLXN_7);

XLXI_12 : FD
port map (C=>XLXN_6,
D=>XLXN_10, Fig.3 ALTYS Spartan 6 Board
Q=>XLXN_9);

XLXI_13 : INV
port map (I=>XLXN_9, III. EXPERIMENTAL SETUP
O=>XLXN_10);
The block scheme for measuring BER in FSO atmosphere
XLXI_14 : FD
channel can be seen at fig.4. It consists of Rx and Tx optical
port map (C=>XLXN_9, drivers, FSO atmosphere channel, counter device, BER
D=>XLXN_12, measuring scheme and pseudo random noise generator. The
Q=>XLXN_11); information is passed through atmosphere channels forward
and backward. At the end of the backward channel there is
XLXI_15 : INV BER measuring scheme, which compare signals from the
port map (I=>XLXN_11, generator and atmosphere channel. If signals are different the
O=>XLXN_12); scheme formed logical ones. Otherwise the scheme form logic
zeroes, which mean no error [3] .
XLXI_16 : FD
port map (C=>XLXN_11,
D=>XLXN_14,
Q=>XLXN_13);

XLXI_17 : INV
port map (I=>XLXN_13,
O=>XLXN_14);

XLXI_18 : FD
port map (C=>XLXN_13,
D=>XLXN_16,
Q=>XLXN_15);

Fig.4 FSO system

35
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

We plan to combine our FSO system fig.5 and FPGA based

PRNG for measuring BER in real conditions. We want to bind
the results for BER with atmospheric effects such as fog, rain,
snow. For reading the intensity of meteorological effects we
will use meteorological station.

Fig.6. Computer simulation of PRNG signal

V. CONCLUSION

The described algorithm can be used for learning the

dependence of the level of bit error in Free Space Optics
systems [8]. This system is easy reprogramming, it can
generate pseudo random bit sequence at frequency of
500Mhz, it have small physical dimensions. In future we plan
to realize all blocks in fig. 3 with FPGA.
Research results presented in this publication are funded by
Fig.5. Scheme for measuring BER in atmosphere channel
Internal competition of TU-Sofia-2011, contr. Nb112pd033-7.

IV. SIMULATION
REFERENCES
Fig.6 and Fig.7 shows computer simulations and real
oscillogram. In the upper part of the figures there is a PRNG [1] Beker H., P.Fred,The Protection of Communications. Wiley-
signal, while in the bottom of the figures there is a clock Interscience. p. 212. 1982.
signal [7]. [2] Pong P. Chu FPGA Prototyping by VHDL Examples: Xilinx
Spartan-3 Version (Feb 4, 2008)
[3] Kolev N., K. Dimitrov Y. Velchev T. Mitsev Experimental
Setup for BER Measuring of
Free Space Optical System , pp.214-217, Nis, Serbia , 2011.
[4] Anil K. Maini Defence Digital Electronics Principles, Devices
and Applications
Research and Development Organization (DRDO), India
[5] www.xilinx.com
[6] http://www.digilentinc.com/Products/Detail.cfm?NavPath=2,40
0,836&Prod=ATLYS
[7] Hartmann A. K. A practical guide to computer simulations
World Scientific Publishing Company; Pap/Cdr edition (March
30, 2009)
[8] Stamatios V. Kartalopoulos Optical Bit Error Rate: An
Estimation Methodology Wiley-IEEE Press 2004

Fig.7. Oscillogram of PRNG signal

36
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Experimental Estimation and Correction of the Methods

for Radio Waves Attenuation Prediction in Rain
Boncho Bonev1, Kliment Angelov2 and Emil Altimirski3
Abstract – An experimental estimation of main methods for
prediction of rain attenuation of radio waves is presented. Data Lrain = Lsp _ rain .r.d / 1000 , (1)
for attenuation values and rain rates from real radio links,
worked in Bulgaria are used for the study. Then the real and where Lsp_rain is the specific rain attenuation in dB/km and can
predicted attenuations are analyzed by using existing methods. be calculated by using the ITU model [4] as follows
Corrections in existed methods for increase their precision for
the territory of Bulgaria are suggested.
Lsp _ rain = k ( f ).RRα ( f ) . (2)
Keywords – Rain attenuation, gaseous attenuation, microwave,
prediction methods. In Eq. (2) RR is the rain rate in mm/h, k(f) and α(f) are the
frequency and polarization dependent constants.
In Eq. (1), the distance d is in meters and r is a correction,
I. INTRODUCTION
what renders an account of the fact that the rain falls only on
part of the link distance and can be calculated by the
The human race progress sets new challenges to the
communication technologies, related mainly with transfer of expression [4, 5]
much more information. The radio links that work in
frequency band from 2,4 to 5,6 GHz are not able to respond to r = (1 + d / 35000 exp(−0.015 RR) )−1 . (3)
these challenges. This requires the usage of higher frequencies
in the present-day communication systems. In this case the In our analysis we use these formulas to calculate
other problems, like hydrometeors and atmospheric gaseous theoretical rain attenuation for examined radio link. We also
attenuation exist. The losses caused by atmospheric gases and use them to obtain the value of specific rain attenuation based
hydrometeors are the main reason for link distance limitation on experimental data. Then by using of the smaller squares
when frequencies over 10 GHz is used [1]. This influence is method we obtain new coefficients for rain attenuation
significant especially for frequencies above 45-50 GHz when prediction model.
the link distance can reach only 1-2 km with the link
availability of 99,99% or greater [1]. III. RESULTS AND ALALYSIS
This paper presents an analysis of calculated and measured
rain attenuation and suggests a correction of coefficients used We examine two radio links working at frequencies
in rain prediction models [2]. The experimental data for radio 11,15 GHz and 19,15 GHz with vertical polarization on
wave attenuation are from 4-year measurements over radio distance of 7,1 km in Sofia region. For our study we use
link that has worked in Sofia region [3]. statistical data for rain rate in Sofia region based on 4-year
measurement. By using this data we calculate theoretical rain
II. THEORETICAL ANALYSIS attenuation with Eq. (1) – (3). The values of k(f) and α(f) are
obtained from [2] and are given in Table I.
The radio waves with frequencies above 10 GHz are
significantly absorbed and scattered by hydrometeors like TABLE I
rain, snow, fog, clouds etc. The attenuation when the heavy
Frequency, GHz k(f) α(f)
rain occurs is so great, that sometimes the connection can
interrupt completely even if the link distance is short. 11,15 0,018396 1,155685
The rain attenuation Lrain is caused mainly by rain
absorption and is expressed as [4] 19,15 0,0878735 0,991755

We also use the statistical data of rain attenuation for these

1
Boncho Bonev is with the Faculty of Telecommunications at radio links and compare and analyze theoretical and
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, experimental rain attenuation – yellow and red drawing on
Bulgaria, E-mail: bbonev@tu-sofia.bg. Fig. 1 and Fig. 2.
2
Kliment Angelov is with the Faculty of Telecommunications at
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000,
Bulgaria, E-mail: kna@tu-sofia.bg.
3
Emil Altimirski is with the Faculty of Telecommunications at
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000,
Bulgaria.

37
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The corrected values of rain attenuation are calculated and

Lrain, dB
18,00
are shown with the blue line on Fig. 1 and Fig. 2. The
corrected values of the coefficients are given in Table II.
16,00

14,00
TABLE II
12,00
Frequency, GHz k(f) α(f)
10,00 11,15 0,03485 1,04390
8,00 19,15 0,096502 1,00477
6,00

4,00 IV. CONCLUSIONS

2,00

0,00 From applied drawings can be viewed that in a heavy rain

0,00 20,00 40,00 60,00 80,00 100,00 120,00 140,00 conditions the calculated by ITU model rain attenuation and
Real Data ITU Model corrected RR, mm/h the real one are rather different. This can cause mistakes in the
link budget calculation. One of possible reasons is the rain
Fig. 1. Rain attenuation for 7,1 km radio link, f=11,15 GHz drops size distribution that can be different for the different
regions of a given country what is in some of climate regions
according ITU prediction model [8]. Therefore there is a need
35,00
Lrain, dB of specifying the rain attenuation prediction models and one
of possible ways is by defining more accurately the coefficient
30,00 in these models. Another way is to divide the climate regions
on sub-regions according to the specific meteorological
25,00
conditions. This demands more experimental data for rain
20,00 attenuation and rain rate in the typical regions like flat
country, mountain etc. and is the object of our further studies.
15,00

10,00 ACKNOWLEDGEMENT
5,00
This work was supported under Project Nr DUNK-01/03 –
0,00 12.09 “University Scientific and Research Complex for
0,00 10,00 20,00 30,00 40,00 50,00 60,00 70,00 80,00 Innovation and Transfer of Knowledge in Areas of
ITU Model Real Data corrected RR, mm/h Micro/Nano Technologies and Materials, Power Effectiveness
and Virtual Engineering”.
Fig. 2. Rain attenuation for 7,1 km radio link, f=19,15 GHz
REFERENCES
We assume the rain rate statistic is the same for the every
part of link distance, because the measurement period is long [1] V. Kvicera and M. Grabner, Results of long-term concurrent
enough. measurement of rain rate and rain attenuation at 38 GHz,
Fig.1 and Fig. 2 show the similar disposition of the Proceedings of 2002 URSI General Assembly, Maastricht,
drawings for theoretical and real rain attenuation of studied Netherlands, 17-24 August 2002.
radio link. Obviously the real attenuation is 1 dB for light rain [2] ITU-R Recommendation P.838-3 (03/05) Specific attenuation
and 3-4 dB for heavy rain bigger than the theoretical for two model for rain for use in prediction methods
studied frequencies. The correction obtained from Eq. (3) is [3] Alexandrova E., V. Sviatogor, V. Pozhidaev, A. Kavecki,
not the frequency dependent and the difference can be Results from experimental investigation on rain attenuation over
terrestrial radio links at frequencies 11,5, 19,3 and 29,3 GHz,
searched out in the specific rain attenuation calculation – Eq.
“Electropromishlenost i priborostroene”, vol. 7, 1989, pp. 15-19
(2). One of the possible reasons can be found in the rain drops [4] J. Seybold, Introduction to RF propagation, John Wiley & Sons
size distribution that is different for the rainfalls occur in Inc., Hoboken, New Jersey, 2005.
different time but with the same rain rate. That for example [5] R. L. Freeman, Radio System Design for Telecommunications,
causes serious differences in optical wave attenuation [6] and John Wiley & Sons Inc., Hoboken, New Jersey, 2007.
also can lead to different specific attenuation for the given [6] Colvero, C., M. Cordeiro, J. von der Weid, FSO Systems: Rain,
frequencies. That give us the ground to precise the Drizzle, Fog and Haze Attenuation from Different Optical
coefficients k(f) and α(f) in Eq. (2). Windows Propagation, International Microwave and
Optoelectronics Conference 2007, October 29 - November 1,
First it is calculated the specific attenuation value according
2007, Salvador, Brazil.
the real measured rain attenuation and then by using the same [7] Ferdinandov, E., B. Pachedjieva, Probability and Statistical
model based on Eq. (2) is obtained the new values of k(f) and Methods in Communications – Part 1, Siela, Sofia, 2005.
α(f). The smallest squares method is used [7]. [8] ITU-R Recommendation P.837-5. Characteristics of
precipitation for propagation modeling.

38
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Multiresolution Analysis of Multiple Reflections

in Transmission Lines
András Fehér1, Ádám Békefi2 and Szilvia Nagy3
Abstract – Wavelet analysis is applied to time-domain signals
of conducted measurements on cables with multiple reflections in
order to detect repeating patterns in the time-frequency domain.
 WFT
F jk  f   aWFT
jk  
  w t   j f t e ik t dt.  
In conducted radio frequency measurements the reflection is one
of the most varying components of the measurement uncertainty. Here w(t) is the window function, τ and ω are the time and
angular frequency of the transformed signal, and j and k are
Keywords – Reflection, wavelet analysis, measurement,
the indices of the values resulting from the discrete
measurement uncertainty.
transformation, belonging to τj = j ∙ τ0 and ωk = k ∙ ω0,
respectively. The reproduction, or synthesis of the original
signal from its continuous transformed counterpart can be
I. INTRODUCTION written as a double integral, similar to the original inverse
Fourier integral, whereas the synthesis of the discrete
In conducted radio measurements multiple reflections due windowed Fourier transformed signal from its coefficients is
to connections are always present, even if matched cables are to be calculated as
used, and this effect plays an important role in the
measurement uncertainty. In this article we present an f t    a WFT
jk  w jk t , (3)
j k
analyzing method based on wavelet transform to study the
nature of multiple reflections in a cable. The method is also
where jk(t) is the basis function of the transformation
tested with an artificially mis-matched cable, which has a part
with 75 Ω impedance between two 50 Ω impedance pieces,
without impedance matching.
 
w jk t   w t   j  eik t , (4)

with overline meaning the complex conjugation. Using this

II. ON WAVELET ANALYSIS notation, transformation (2) can be rewritten in a shortened
form
Wavelet analysis or multiresolution analysis [1–3] is a
widely used tool in data processing, especially in image WFT
F jk  f   aWFT
jk   w jk t  f t dt. (5)
compression and noise reduction [4,5], but it can be also used
for solving differential equations [6–8]. The results of the The wavelet transform has similar formula, except that the
wavelet analysis can also detect patterns. transformation function is different, i.e., in case of discrete
Let f be a function of the space of the square integrable transformation
functions L2( ). Wavelet analysis can be introduced e.g., as a
generalization of the windowed Fourier transform WaT
F jk  f   d jk   jk t  f t dt. (6)

 F WFT  f  ,    wt    f t e it dt ,   where the wavelets

and its discretised version 

 jk t   2 j / 2 2 j t  k .  (7)

Similarly to the window functions, there are many types of

wavelets, but once its type is chosen, all the wavelets are
generated as scaled and shifted versions of one function, the
mother wavelet ψ(t). This means that while in case of the
1
Andras FEHÉR, Széchenyi István University, Radio Frequency windowed Fourier transform, the envelop of the transforming
Test Laboratory, Egyetem tér 1, Győr, Hungary, H-9026 function remains the same (only shifted in time), and
E-mail: afeher@sze.hu. Web: http://rf.sze.hu increasing frequency manifests in more oscillations within the
2
Ádám BÉKEFI, Széchenyi István University, Department of window function, in case of the wavelet transform the shape
Telecommunications, Egyetem tér 1, Győr, Hungary, H-9026 of the transforming function remains similar, only shrunk with
E-mail: bekefiadam@index.hu. the increasing frequency. A demonstration for the basis
3
Szilvia NAGY, Széchenyi István University, Department of
Telecommunications, Egyetem tér 1, Győr, Hungary, H-9026
function shape can be seen in Fig. 1.
E-mail: nagysz@sze.hu. The inverse transformation takes the form

39
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

the high-pass part will belong to the actual wavelets, the low
pass one to the scaling functions; this can be further analyzed.
Downsampling is needed, thus the total number of expanding
coefficients remain constant after each step; cj-1 l and dj-1 l are
half as long as the starting cj l.

Fig. 2. Schematic diagram of one step in the wavelet analysis.

Filter coefficients gl and hl represent the high-pass and
low-pass filters, respectively; the circles mean downsampling by 2.
The starting vector cj k can be either the original sampled signal or
the low-pass output of the previous step.

The synthesis or reconstruction step is opposite to the

analysis step plotted in Fig. 2.

III. STUDY OF MULTIPLE REFLECTIONS IN COAXIAL

CABLES

In order to study multiple reflections we have prepared a

wrong connecting cable from Hirschmann KOKA 709 (75 Ω),
and H155 (50 Ω) low loss coaxial cables. 50 Ω instruments
were used for the measurement, and the connecting ends of
Fig. 1. Basis functions of a windowed Fourier transform (blue lines) the cable under test (CUT) were the H155 type lines, the
with Gaussian window (red lines) and a wavelet (Daubechies-4, middle part was substituted by 1.62 m of KOKA 709 line. As
green lines) at three frequencies or resolution levels. a reference high precision cable with attenuator was applied.
As a first step, the transmission characteristics of the cable
was determined by a network analyzer; its parameter S21can
  
f t    c0k 0k t     d jk jk t . be seen in Fig 3. A clear resonance valley is present at the first
(8)
k  j 0 k 
marker, near 105 MHz.

In this equation the first summation is necessary to set the DC

or low frequency components of the signal f(t), whereas the
second part makes the higher frequency refinements. The
basis elements ϕ0k(t) of the lowest frequency part are the
scaling functions. They behave similarly to wavelets in
Eq. (7). With increasing the frequency index j, the
corresponding frequency doubles, as it can be seen from (7)
and Fig. 1. It can also be observed in (7) that by increasing j
by 1, the shift distance is halved. This property is very
favourable if sharp edges, quick changes have to be
reproduced, as near the changes the wavelet coefficients djk
are large, and they are negligible in regions where smooth
changes are present only. Such functions are often present
both in image processing and in one dimensional data
analysis, and they cannot be treated properly with windowed
Fourier analysis, discrete cosine transform, etc., where the
window width is constant, and usually much larger than the Fig. 3. S21 of the studied cable measured by Rohde&Schwarz ZVL
edge which is studied by it. Network Analyzer (9 kHz to 6 GHz)
In practical applications, where a one or two dimensional
digital (sampled and quantized) signal is analyzed, the highest
frequency corresponds to the sampling frequency. During the As a test signal we have applied a carrier signal modulated
analysis the vector (matrix in 2D) is transformed according to by a 30 ns burst with a period of 200 ns. The carrier
Fig. 2, consecutively. The frequency domain is always halved, frequencies were near the 105 MHz point as the transmission

40
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

parameter varied there rather quickly, thus the modulated

 d jkn 
Nm
2
signals two sidebands had really different propagation
conditions. Wavelet analysis is efficient where the quick n 1
changes in the time domain are present, hence the usage of the d jk  , (9)
Nm
short bursts as modulator signals, practically the edges and
their near environments are interesting. The cable was and plotted in Fig. 6 for Nm = 10 and Nm = 100. The upper
measured by a 5 GS/s Tectronix oscilloscope, an example index n means the serial number of the measurements, Nm the
with its reference signal can be seen in Fig. 4. The modulator total number of measurements.
and modulated signals were not synchronized in order to be Fig. 7 contains the results for various resolution levels, i.e.,
able to study different relative phases, thus different shapes of for various frequency components. Ten analysing steps
(j = 0...9) were carried out, but only those are given where
interesting characteristics can be seen. The lower frequency

Nm = 10

Fig. 4. An example of the measured signal with a carrier frequency

of 120 MHz coming through the cable with reflection points (dashed
blue line) and a reference cable (continuous red line). The horizontal
axis is the number of the sampling.

signals.
Automatic measurement environment was prepared to
gather sufficiently large number of data vectors, at the
frequencies 93 MHz and 120 MHz. one hundred of
measurements were carried out and wavelet transformed. The Nm = 100
resulting coefficients varied a lot, as it is demonstrated in
Fig. 5.

Fig. 6. The effective wavelet coefficients for 10 (upper subfigure)

and 100 measurements. Data of cable with multiple reflection points
is plotted with blue dashed line, the reference with red continuous
line. Resolution index j = 1.

components do not differ significantly from the reference.

IV. CONCLUSION
The normalized wavelet coefficients of the reference and
the multiple reflected signals can be distinguished in higher
frequencies, however, the low frequency terms are in
Fig. 5. The 5th wavelet transforms of 10 different signals from the
cable with multiple reflection points. approximately the same, at least in average. The sampling
time ts = 400ps of the two-channel oscilloscope is just about 4
In order to see the trends, the square of the coefficients djk to 5 refinement steps away from the 93-120 MHz carrier
were summarized for all the signals. The result’s square root frequency’s characteristic time, thus the high frequency
was normalized by the number of measurements, effects of the reflections and nonlinearities should be found in
the first couple of steps, thus our results meet the
expectations. Also, ts becomes commensurable with the 30 ns

41
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 7. The effective wavelet coefficients after 100 measurements.

burst time after 6-7 refinement steps, thus the lower frequency [2] C. K. Chui, An Introduction to Wavelets, (Academic, San
terms will contain mostly the burst itself. Multiple reflection Diego, 1992).
in conducted measurements can be characterized by a [3] S. Mallat, A theory for multiresolution signal decomposition:
significant increment in the average fine resolution wavelet the wavelet representation; IEEE Trans. Pattern Anal. Mach.
Intel., vol. 11, pp. 674-693 (1989).
coefficients. [4] J. Kaftan, A. A. Bell, C. Seiler, T. Aach, Wavelet based
denoising by correlation analysis for high dynamic range
imaging; IEEE International Conference on Image Processing,
Cairo, pp. 3857-3860. (2009).
ACKNOWLEDGEMENT [5] C. Christopoulos, A. Skodras, and T. Ebrahimi, IEEE Trans.
Consum. Electron. vol. 46, p. 1103 (2000).
This work was supported by the Hegedüs Gyula fellowship [6] K. Urban, Wavelet Methods for Elliptic Partial Differential
and the projects TÁMOP-4.2.1./B-09/1/KONV-2010-0003, Equations, Oxford University Press, Oxford, (2009).
and TáMOP 4.1.1.A-10/1/KONV-2010-0005 of the Széchenyi [7] W. Dahmen, Wavelet methods for PDEs—Some recent
István University. developments, J. Comput. Appl. Math. vol. 128, p. 133 (2001).
[8] J. Pipek, Sz. Nagy, Refinement trajectory and determination of
eigenstates by a wavelet based adaptive method, J. Chem. Phys.
REFERENCES vol. 125, 174107 (2006).
[9] M. Kuczmann, A. Iványi, Finite Element Method in Magnetics,
[1] I. Daubechies, Ten Lectures on Wavelets, CBMS-NSF Regional Academic Press, Budapest, 2008, ISBN: 978 963 05 8649 8.
Conference Series in Applied Mathematics 61 (SIAM,
Philadelphia, 1992).

42
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Ad-Hoc Supported, Connection Fault-Tolerant Model for

Mobile Distributed Transaction Processing
Tome Dimovski1 and Pece Mitrevski2
Abstract – Mobile embedded systems increasingly use The main contribution of this paper is performance analysis
transactions for applications like mobile commerce, banking or of the Connection Fault-Tolerant Model [8] for mobile
commercial applications. We make a short review of the distributed transaction processing. We study the impact of the
Connection Fault-Tolerant Model for mobile distributed level of ad-hoc communication support on the mobile
transaction processing. We made simulation experiments study.
The performance analysis shows that ad-hoc communication
transaction commit rate.
supports considerably improve the transaction commit rate. The paper is organized as follows. Section II gives a survey
of related work. In Section III we present model of the mobile
Keywords – Mobile transaction, mobile computing environment, and in Section IV transaction model. In Section
environment. V we present our Connection Fault-Tolerant Model. In
Section VI, we present simulation results and analysis.
Section VII discusses conclusion.
I. INTRODUCTION
The increasing emergence of mobile devices contributes to II. RELATED WORK
rapid progress in wireless technologies. Mobile devices
interacting with fixed devices can support applications such as e- All Transaction Commit On Timeout (TCOT) [3] protocol
mail, mobile commerce (m-commerce), mobile banking etc. But is based on timeout approach for Mobile Database Systems,
there are many issues that are challenging and need to be which can be universally used to reach a final transaction
resolved before enabling mobile devices to take part in termination decision in any message oriented system. This
distributed computing. For distributed systems the transaction protocol limits the amount of communication between the
is a set of operations that fulfill the following condition: either participants in the execution of the protocol. TCOT does not
all operations are permanently performed, or none of them are consider mobile hosts as active participants in the execution of
visible to other operations (known as the atomicity property). transactions.
In the execution of transactions the key issue is the protocol The basic idea of the Two-Phase Commit Protocol for
that ensures atomicity. Mobile Wireless Environment (M-2PC) [4] is to adapt the 2PC
The mobile environment is characterized with mobile protocol for mobile systems with distributed transactions.
devices with limited resources like processing, storage, Mobile hosts are active participants in execution of
energy capacity and continuously varying properties of the transaction and they are sending confirmation that the work is
wireless channel. This increases the time needed for mobile done to the agent or to the fixed device in order to save
hosts to execute transactions and can even lead to execution energy. This protocol does not provide adequate management
failure. Mobile hosts (MHs) naturally show frequent and of mobility and failures caused by the network disconnection.
random network disconnections. The limitations and Fault-Tolerant Pre-Phase Transaction Commit (FT-PPTC)
characteristics of the mobile environment make it harder to [5] protocol provides mechanisms for dealing with
design appropriate and efficient commit protocols. A protocol disturbances in the systems in mobile environment. The
that aborts the transaction, each time the MH disconnects from protocol supports heterogeneous mobile databases. FT-PPTC
the network, is not suitable for mobile environments because it implements distributed transaction in two phases: pre-phase,
is part of the normal mode of operation. Disconnections need one which is covering the mobile hosts and the main phase
to be tolerated by the protocol. which refers to the fixed part of the network. Mobile hosts are
The two-phase commit (2PC) protocol [1] that allows the active participants in execution of transaction. No
involved parties to agree on a common decision to commit or mechanisms are developed for competition in the mobile
abort the transaction even in the presence of failures is the distributed transactions. FT-PPTC doesn’t provide adequate
most commonly used protocol for fixed networks but is management of mobility because when mobile hosts are
unsuitable for mobile environments. There are several other disconnected from fixed network for a long time they can
protocols, [3-5], [7], for transaction execution in distributed block resources on the fixed participants. This leads to an
mobile environment, but almost all consider limited number increased number of mobile transaction aborts.
of communication models.

1
III. MODEL OF THE MOBILE ENVIRONMENT
Tome Dimovski is with the Faculty of Technical Sciences at
University “St. Kliment Ohridski” of Bitola, Ivo Lola Ribar bb,
In this paper we consider system model for a mobile
Bitola 7000, Macedonia, E-mail: tome.dimovski@uklo.edu.mk.
2
Pece Mitrevski is with the Faculty of Technical Sciences at distributed environment consisting of a set of mobile hosts
University “St. Kliment Ohridski” of Bitola, Ivo Lola Ribar bb, (MHs) and a set of fixed hosts (FHs), presented in Fig. 1. The
Bitola 7000, Macedonia, E-mail: pece.mitrevski@uklo.edu.mk. model has two main parts: the fixed part of the network and a

43
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

mobile part of the network. Communication between the fixed V. CONNECTION FAULT-TOLERANT MODEL
part and the mobile part of the network is conveyed via
Mobile Support Stations (MSS). MSSs are connected to the
fixed part of the network via wired links. MHs can cross the A. Overview
border between two different geographical areas covered by
different MSSs. Some of the most frequent failures in mobile environments
are communication failures. When MHs are in motion, they
may exit the geographical area that is covered by some MSS
and the resources of the fixed participants may potentially be
blocked for an undefined period of time. If MHs do not
reestablish connection with any MSS the transaction is
aborted.
To minimize the number of mobile transaction aborts by
tolerating failures caused by network disconnections we
proposed Connection Fault-Tolerant (CFT) Model [8] for
distributed transaction processing in mobile computing
environment. The CFT Model ensures the atomicity property.
The CFT model considers two communication scenarios:
1. The first scenario is a Standard communication scenario
when MHs can directly connect to fixed network through
MSSs.
2. The second scenario is an Ad-hoc communication
scenario when MHs cannot directly connect to the fixed
network through any MSS. In this scenario MHs can ad-hoc
Fig. 1. Model of the Mobile Environment communicate with neighboring MHs to reach the fixed
network.
In the considered system model, first, MHs can In the Standard scenario, similar to [6], to minimize the use
communicate with the FHs through MSS via wireless of the wireless communication and conservation of the
channels only when they are located within the MSS coverage resources of MHs we assign a Mobile Host Agent (MH-Ag),
area. Second, the MHs can ad-hoc communicate with which we add in the fixed network, to each MH. We assume
neighboring MHs via wireless channels. When MHs enter a that in the execution of a transaction MH-Ag is representing
geographical area that is out of coverage of any MSS, to the MH in the fixed network and it acts as an intermediator
access database servers on fixed network they may connect between MH and the transaction CO. All communications
through a neighboring MH which is in the covering area of between MH and CO go through the MH-Ag. The MH-Ag is
any MSS. responsible for storing all the information related to the states
In our scenario, we consider mobile distributed of all MTs involving the MH. In the fixed network, a server or
environment where MHs can communicate with each other, or servers can be designated, where MH-Ag are created for each
with FHs through MSS, and MHs can ad-hoc communicate participating MH.
with neighboring MHs to reach fixed network. In our CFT model we define additional function to MH-Ag
We assume that database servers are installed on each FH, that we called Decision Algorithm (DAlg). DAlg is used
and each MH has a mobile database server installed. during the execution of a transaction when MH-Ag cannot
directly or ad-hoc communicates with its MH for a defined
period of time. DAlg’s task is to check if Transaction
IV.TRANSACTION MODEL Processing Fragment (TPF) function is write
(insert/update/delete) or read. If TPF function is write, DAlg
A distributed transaction where at least one MH participates saves the TPF in FIFO (First In First Out) queue list and
is called a Mobile Transaction (MT). We identify a MH where makes a decision for MH to send “Yes” vote to the transaction
a transaction is issued as a Home-MH (H-MH). Participating CO. When the connection between MH and corresponding
MHs and FHs in the execution of a mobile transaction are MH-Ag is reestablished, MH-Ag’s first task is to send all
called participant MHs (Part-MH) and participant FHs (Part- saved TPFs to the corresponding MH. If TPF function is read,
FH). DAlg will wait for connection reestablishment between MH
In our model we assume the existence of a Coordinator and corresponding MH-Ag, for a defined period of time. If the
(CO) which is responsible for coordinating the execution of connection is not reestablished in the specified time period,
the corresponding transaction. The CO is responsible for DAlg makes a decision for MH to send “No” vote to the
storing information concerning the state of the transaction transaction coordinator.
execution. Based on the information collected from the The second communication scenario is when MHs cannot
participants of the transaction, the CO takes the decision to directly connect to the fixed network, or MH-Ag cannot
commit or abort the transaction and informs all participants directly communicate with its MH through any MSS. In that
about its decision. The CO should be executed on the fixed case, they try to connect through ad-hoc communication with
host or hosts. That means logs will be kept more safely. any neighboring MH which is in the covering area of any

44
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

MSS. To allow this, we assign a MH-Relay Agent (MH-RAg) After the coordinator has received vote from every
to each MH. The MH-RAg is responsible for ensuring relay participant, it decides whether to commit or abort the
wireless link between neighboring MHs. This means that MH transaction. If, for any reason, even one of the participants
which is out of the coverage area can connect to his MH-Ag votes “No” or execution time–out expires, the coordinator
of the fixed network via MH-RAg of the neighboring MH decides to abort the transaction and sends “Abort” message to
which is in coverage area of any MSS. all participants. Otherwise, if all received votes are “Yes” and
execution time–out is not expired, the coordinator decides to
B. Connection Fault-Tolerant Model operation commit the transaction and sends “Commit” message, with
reset execution time–out to all participants. The participants
In this section we make a short review of the operation of need to acknowledge the coordinator decision before
the CFT model that has been presented in [8]. Fig. 2 execution time–out expires.
illustrates the execution of a mobile transaction for the
proposed model, but without the functions of the Decision VI.SIMULATION RESULTS AND ANALYSIS
Algorithm because in this paper our interest is concentrated on
the Ad-hoc communication impact on mobile transaction The simulation experiments study the impact of ad-hoc
execution. communication in CFT Model on the mobile transaction
If H-MH is connected to the fixed network through some commit rate. For the simulations, we used SimPy [9], a
MSS, it initiates a mobile transaction by sending transaction process-based discrete-event simulation package based on
processing fragment (TPF) to the coordinator through its standard Python programming language [10]. Table I
corresponding MH-Ag which acts as intermediator between summarizes our simulation parameters.
coordinator and MH.
Transaction coordinator computes the execution time-out TABLE I
( Et ), which is a time limit for all participants to complete the SIMULATION SETTINGS
execution of the TPFs and send a VOTE to coordinator. After Parameter Value
that coordinator sends Et and TPFs to all participating FHs Number of Part-MHs 5-10
and MH-Ags, asks them to prepare to commit the transaction, Fragment execution time (MH) 0.5s
and enters the wait state. MH-Ags try to send TPFs to their Fragment execution time (CO) 0.3s
participating MHs. If MH-Ag cannot communicate with a MH Transmission delay (wireless link) 0.4s
trough standard communication, it tries to connect through ad- Transmission delay (wireless ad-hoc link) 0.9s
hoc communication to any neighboring MH which is in the Transmission delay (wired link) 0.2s
covering area of any MSS. That function is assigned to the Disconnection Rate 0 - 95%
MH-Relay Agent. Ad-hoc support 0 - 80%

Hence, disconnection rate is defined as the ratio of time

where the participating MH is disconnected from the fixed
network, against the total simulation time. Ad-hoc support is
the ratio of time where ad-hoc communication is available
between MHs, against the total simulation time. It is hard to
quantify the level of ad-hoc support between MHs in mobile
distributed environment. In some parts of the wireless network
ad-hoc support can be lower compared to other. For that
reason, in our simulation we define three groups that represent
different parts of the wireless network with different level of
ad-hoc support. Every MH in the wireless network is a
member of one of the defined groups.
A simulation run is set to simulate 10 hours. Transactions
are generated with exponentially distributed in-between
intervals, with an average of 30 seconds. We assume that all
transactions are of similar length, but experience different
connection conditions.
Fig. 2. Mobile transaction processing
The main parameter for performance evaluation is the
mobile transaction commit rate. Figs. 3, 4 and 5 show mobile
When the participants receive the “prepare” message, they
transaction commit rate against different disconnection rates,
check if they could commit the transaction. If so, participants
and different ad-hoc support values. Transaction time-out is
send YES vote to the coordinator. MH sends vote to
set to 5 seconds. For results shown in Fig. 3 we assume that
coordinator through its corresponding MH-Ag via standard or
every MH is a member of the same group. It is evident that
ad-hoc communication.
ad-hoc support in the CFT model considerably improves the

45
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

transaction commit rate. The ad-hoc communication impact is of committed mobile transactions and shows resilience to
higher for networks where disconnection rate is higher. connection failures of the mobile devices.

Fig. 3. Impact of ad-hoc communication on Fig. 5. Impact of ad-hoc communication on

commit rate (single group) commit rate (three groups)

The performance analysis shows that ad-hoc support in the

CFT model considerably improves the transaction commit
rate. It is evident that the impact of ad-hoc support is higher
for less dynamic wireless networks.
In our future work we plan to evaluate the performance of
a model that includes a decision algorithm, as well, for which
we expect to have positive impact on mobile transaction
commit rate.

REFERENCES
[1] J. Gray, “Notes on Data Base Operating Systems”, Operating
Systems, An Advanced Course, 1978.
[2] N. Santos, P. Ferreira, “Making Distributed Transactions
Fig. 4. Impact of ad-hoc communication on Resilient to Intermittent Network Connections”, Proceedings of
commit rate (two groups) the 2006 International Symposium on on World of Wireless,
Mobile and Multimedia Networks. IEEE Computer Society,
Fig. 4 shows simulation results where MHs are classified in Washington, 2006.
two different groups that have different levels of ad-hoc [3] V. Kumar, “A Timeout-Based Mobile Transaction
support. It is evident that ad-hoc support improves the Commitment Protocol”, Proceedings of the East-European
transaction commit rate, but the percentage of improvement is Conference on Advances in Databases and Information
lower compared to the previous scenario where the level of Systems, 2000.
ad-hoc support was the same in each part of the wireless [4] N. Nouali, A. Doucet, H. Drias, “A Two-Phase Commit
Protocol for Mobile Wireless Environment”, Proc. 16th
network. Compared to the previous scenario, commit rate
Australasian Database Conference, 2005.
slightly decreases when the disconnection rate rises all the [5] B. Ayari, A. Khelil, N. Suri, “FT-PPTC: An efficient and fault-
way from 60 up to 95 seconds. tolerant commit protocol for mobile environments”, Proc. of
To present the influence of ad-hoc support in a highly SRDS, 2006.
dynamic wireless network, we classify all MHs in three [6] L. Xiang, Z. Yue-long, C. Song-qiao, Y. Xiao-li, “Scheduling
groups that have different levels of ad-hoc support. From the Transactions in mobile distributed real-time database systems”,
chart in Fig. 5, one can conclude that commit rate increment is Journal of Central South University of Technology, 2008.
not evident as before, e.g. the highest improvement of the [7] S.A. Moiz, M.K. Nizamudin, “Concurrency Control without
commit rate is about 11%. Locking in Mobile Environments”, First International
Conference on Emerging Trends in Engineering and
Technology, Nagpur, Maharashtra, 2008.
VII. CONCLUSIONS [8] T. Dimovski, P. Mitrevski, “Connection Fault-Tolerant Model
for distributed transaction processing in mobile computing
environment”, Information Technology Interfaces, 33rd
In this paper we made a short review of the operation of the
International Conference, Dubrovnik, Croatia, 2011.
Connection Fault-Tolerant Model [8] for mobile distributed [9] URL < http://simpy.sourceforge.net/>
transaction processing. It is developed to increase the number [10] URL < http://www.python.org/>

46
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fast Synthesis of High Order Microwave Filters by

Coupling Matrix Optimization
Marin V. Nedelchev, Ilia G.Iliev

Abstract: This paper presents optimization method for synthesis of They perform robust optimization, no matter about the
generalized high order microwave filters with arbitrary topology. The starting point. Unfortunately the global optimizers such as
method utilizes local optimizer for coupling matrix determination. genetic or stochastic have very slow convergence to the cost
The synthesis procedure converges very fast as for a initial point is function minimum.
used a vector based on the Chebyshev all pole filter for the same
degree of the filter. To validate the proposed synthesis method two
This paper presents optimization method for synthesis of
numerical examples for resonant filters are computed. The frequency high microwave filters with arbitrary topology. The method
responses from the synthesis procedure and the theoretical responses uses Nelder-Mead local optimizer for coupling matrix
show excellent agreement. determination. The synthesis procedure converges very fast as
Keywords: microwave filter, Chebyshev filter, Nelder-Mead for a initial point is used a vector based on the Chebyshev all
optimization, coupling matrix. pole filter for the same degree of the filter. The cost function
is based on amplitude of the transmission and reflection
I. INTRODUCTION coefficient zeros and their values at the cut-off frequencies
and the reflection coefficient maxima. To validate the
Microwave coupled resonator filters play important role in proposed synthesis method two resonant filters are designed
the modern communication systems. The constraint with asymmetrical responses. The frequency responses from
RF/microwave spectrum requires high attenuation in the stop the synthesis procedure and the theoretical responses show
band and low insertion loss in the passband of the filters. excellent agreement.
These requirements can be met only by cross-coupled
microwave filters, realizing attenuation poles on finite II. RESONATOR FILTER CHARACTERISTICS
frequencies. Cross-coupled resonator filters allow using
various topologies with variety of frequency responses. The The synthesis procedure starts with the low-pass prototype
microwave filter modelling is very important for the fast and with normalized angular frequency of passband ω = 1 . The
accurate design. transfer and reflection coefficients may be expressed as a ratio
Key point in the obtaining of the coupling matrix of two N-th degree polynomials as follows:
corresponding to the practical filter topology is to convert its P (ω ) F (ω )
transversal form to folded form using matrix rotations. Most S21 = N , S11 = N (1),
of the matrix rotation sequences are given in [4]. It is noticed EN ( ω ) ε EN (ω )
that this method for synthesis suffers from generality, because where ω is real angular frequency and
the matrix rotations cannot be derived for every one practical
filter topology. Some of the matrix rotation sequences cannot (
ε = 1 10 RL 10
)
− 1 . ( FN (ω ) PN (ω ) )
ω =1
, RL is the prescribed
converge in order to find the coupling matrix. Some of the value of the return loss in dB, in the passband of the filter. It is
disadvantages in this method are solved if arrow form of the assumed that all polynomials are normalized to their highest
coupling matrix is used [5] or Pfitzenmeir method is used [6]. degree coefficient.
In many practical cases, it is necessary to define the filter The method of computing the numerator of the reflection
topology in order to satisfy some manufacturing or space coefficient is outlined in [3].
requirements. In this case, the exact solution is hard to be
FN (ω ) = ( GN (ω ) + GN′ (ω ) ) ,
1
found utilizing the conventional synthesis methods. (2),
2
One possible general solution to the filter design for where both polynomials can be represented by two
G N ( ω ) = U N ( ω ) + VN ( ω )
arbitrary topology is to apply direct local optimization over
polynomials: and
the coupling matrix with successive starting point. In the basic
papers proposed optimization method for coupling matrix GN′ ( ω ) = U N (ω ) − VN (ω ) . Both polynomials U N (ω ) ,
VN (ω ) can be arranged according to the Cameron's recursive
synthesis [7,8], the starting vector is set to arbitrary values.
This makes the local optimization very unstable method for
cost function minimization. Another method is to use global procedure in [3]. Obviously the roots of U N (ω ) corresponds
optimization method for finding the coupling matrix for reflection zeros, and the roots of VN (ω ) correspond to the in-
certain filter topology.
band reflection maxima.
Marin Veselinov Nedelchev and Ilia Georgiev Iliev –are with It can be easily found that the transfer coefficient may be
Dept. of Radiocommunication and Videotechnologies in Faculty of
expressed in the following way[3]:
Telecommunication in TU –Sofia, N8, Kliment Ohridski bul., 1700
1
Sofia, Bulgaria. E-mail: mnedelchev@tu-sofia.bg, igiliev@tu- S212 (ω ) = (3)
sofia.bg. 1 + ε C N (ω )
2 2

where CN (ω ) is the filtering function. For general

Chebyshev characteristics, the filtering function is in the form:

47
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

 N  are the in-band maxima frequencies. In most papers

C N (ω ) = cosh  ∑ a cosh ( xn )  (4), concerning the optimization of the coupling matrix the last
 n =1  term of the cost function is missing. Because of the high order
ω − 1 ωn of the filter, the value of the transmission coefficient at the
where xn = , where ωn is the angular frequency of
1 − ω ωn prescribed zeros ω p is comparable to the precision of the
the prescribed transmission zero. computer. This make the optimization process hard to
In order to obtain the coupling matrix, it is necessary to converge at the global minimum of the cost function. The
consider the equivalent circuit of general coupled resonator global minimum is the Chebyshev solution for the microwave
filter shown on Fig.1. filter. As the values of the second term of the cost function
М2,n-1
M13 ML,n-1 needs to be weighted, in order to achieve comparable values
to the other terms of the cost function. Obviously there will
MS1 M12 M23
S L come up a problem with the choice of the weighted constant.
MS2
R1 R2 R3 Rn-1 Rn
For each filter topology and frequency response, a different
constant will be necessary. One possible solution for the
MS3
problem with the weights is to make each term of the cost
MSL
function in logarithmic scale with no weight coefficient.
Another solution is to add to the cost function another term
Fig.1. General coupled resonator filter
equalizing the reflection coefficient at its maxima to the ripple
The equivalent circuit consists of N series coupled factor ε . In this case the cost function contains all possible
resonators with frequency independent couplings M ij ( i ≠ j ), constraints for the filter response. The zeros for the
between the i-th and j-th resonators. The circuit is driven by transmission coefficient S21 are set at the prescribed
voltage source E with internal normalized resistance frequencies. The reflection coefficient must be zero at the
R1 = 1 and loaded to normalized impedance R2 = 1 . The
frequencies ω z , equal to ε ε 2 + 1 at the normalized cut-off
resonant frequency of each resonator f 0i is represented by the
frequencies ωcut − off = ±1 and equal to ε ε 2 + 1 at the
self-coupling coefficient M ii and the center frequency of the
filter. The transmission and reflection coefficients of a lossless frequencies ωm at the minimum of the cost function. The
filter of N-th order depend only of the coupling matrix М cost function may be modified with respect to the
transmission coefficient at the frequencies ωm . At these
(7):
frequencies S21 must be equal to ε , but the cost function will
S21 = −2 j [ A]N + 2,1 , S11 = 1 + 2 j [ A]11 , (6),
−1 −1

not be changed in its character.

where [ A] = − j [ R ] + ω [W ] + [ M ] , and [ R] is In this way it is possible to formulate the local optimization
problem for obtaining the coupling matrix.
( N + 2) × ( N + 2) matrix, which elements are zeroes except The starting point for optimization of the coupling matrix
is very important for the reaching of the global minimum of
R11 = RN + 2, N + 2 = 1 . [W ] is a ( N + 2 ) × ( N + 2 ) matrix, where the cost function (7). Having on mind that a local optimizer is
used, the starting vector should be close to the target value in
the main diagonal elements are unity except order to assure a fast convergence of the method. One of the
W11 = WN + 2, N + 2 = 0 . All remaining elements of [W ] are zeroes. possible starting coupling matrices is to set all self-coupling
couplings to zero ( M ii = 0 ) and all direct couplings to 1. The
[M ] is the coupling matrix, symmetrical around the main cross-coupling coefficients are all set to zero. The second
possible starting coupling matrix is to use classical Chebyshev
diagonal. filter from the same order. All self- and cross-couplings are
set to zero.
III. SYNTHESIS OF MICROWAVE FILTER WITH The investigation of the problem of high order filter design
COUPLING MATRIX OPTIMIZATION two numerical designs are investigated.
The cost function used in the optimization process is based
III. NUMERICAL RESULTS
on the zeroes and poles of the filtering function CN , assuming
that the number of poles is P and zeroes N [8]: For verification of the optimization method presented in
2
 ε  this paper, it is applied to an asymmetric resonator filters.
Cost = ∑ S11 (ω zi ) + ∑ S 21 (ω pi )
N P
+  S11 (ω = −1) −
2 2
 +
i =1 i =1  ε 2 +1  (7). A. Asymmetric 9 Resonator Passband Filter
2 2
 ε   N −1
ε 
+  S11 (ω = 1) −  + ∑  S11 (ω = ωmi ) − 2  The first numerical example is 9-th order CT filter sharing
 ε + 1  i =1 
2
ε +1  common resonator. This filter is of Chebyshev type and it has
In the cost function ω p are the prescribed transmission return loss more than 20dB in the passband. The transmission
coefficient prescribed zeros are placed on normalized
zeros, ω z are the zeros of the reflection coefficient, and ωm

48
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

frequencies ω p = [ −1.8, − 1.4, 1.3, 1.6] .The coupling 0 0.9842 0 0

0 0.0012 0.7589 −0.2868
0
0
0
0
0
0
0
0
0
0

0

0
0
0
 
diagram of the synthesized filter is shown on Fig.2. 0 0.7589 0.4095 0.5306 0 0 0 0 0 0 0 
 
0 −0.2868 0.5306 −0.0498 0.4665 0.2788 0 0 0 0 0 
S L 0 0 0 0.4665 −0.5535 0.4462 0 0 0 0 0 
 
R1 R3 R5 R7 R9 M = 0 0 0 0.2788 0.4462 0.0035 0.4844 0.2049 0 0 0 
0 0 0 0 0 0.4844 −0.4152 0.5049 0 0 0 
 
R2 R4 R6 R8 0 0 0 0 0 0.2049 0.5049 −0.0656 0.4767 −0.4048 0 
 
0 0 0 0 0 0 0 0.4767 0.5722 0.7031 0 
0 0 0 0 0 0 0 −0.4048 0.7031 0.0012 0.9842 
Fig.2. Coupling scheme of an asymmetric 9 pole filter  
0 0 0 0 0 0 0 0 0 0 0 

The reflection and transmission zeroes are calculated and The frequency response of the designed filter, calculated
summarized in Table 1. according to (6) and the coupling matrix derived in the
Table 1. Poles and zeros of asymmetric nine resonator filter optimization process, is shown on Fig.4. It is clearly seen that
№ Reflection zeros Transmission zeros Roots of VN the normalized cut off frequency is ωc = ±1 , while the
1 -j0.9880 -j1.8 -j0.9514 transmission zero frequencies are exactly
2 -j0.8888 -j1.4 -j0.7984
at ω p = −1.8, − 1.4, 1.3, 1.6 .The maximum value of the
3 -j0.6795 j1.3 j0.9566
4 j0.9893 j1.6 j0.8166 return loss is with the prescribed value of -20dB.
5 j0.8998 j0.5654
6 j0.7051 -j0.5331
7 j0.4006 j0.2164
8 -j0.3629 -j0.1752
9 j0.0211

The initial point for the coupling matrix elements for the
optimization procedure is to set the values of the all pole nine
resonator Chebyshev filter M S1 = M 9 L = 0.9876 ,
M 12 = M 89 = 0.9168 , M 23 = M 78 = 0.5870 ,
M 34 = M 67 = 0.5480 , M 45 = M 56 = 0.5372 .
Fig.4 Frequency response of nine resonator filter with asymmetric
response. Solid line-S21, dashed line- S11

B. Cascaded Quadruplet and Triplet Resonator

Passband Filter of 10-th Order
The 10-th order resonator filter is formed by cascade
connection of two trisections and one quadruplet section
between them (CQT filter). Each trisection realizes one
prescribed transmission zero and the quadruplet section
realizes two prescribed symmetrical transmission zeros. The
filter is of Chebyshev type and it has maximum return loss of
-20dB. The transmission zeroes are placed on frequencies
ω p = [ −1.2, ± 2, 1.6] .The coupling scheme of the filter is
Fig.3 Cost function value during optimization for CT nine
resonator filter shown on Fig.5.
S L
All self coupling and cross coupling coefficients are set to R1 R3
R4 R7 R8 R10
zero. The number of the independent values of the coupling
matrix is 23. R2 R5 R6 R9

After 139 iterations for the optimization coefficient, the

Fig.5. Coupling scheme of CQT filter of 10-th order
optimization procedure converges. The values of the cost
function vs the number of iterations is shown on Fig.3. The The roots of the polynomials in the numerator and
initial value of the cost function is 0.51 and the end value is denominator in (1) are shown in Table2. The starting point for
2.143.10 −8 . The optimization process stopped because of the optimization process is based on the Chebyshev coupling
reaching local minimum of the cost function (7). The final matrix elements M S1 = M 10 L = 0.9854 ,
coupling matrix is: M 12 = M 9,10 = 0.8130 , M 23 = M 89 = 0.5839 ,
M 34 = M 78 = 0.5444 , M 45 = M 67 = 0.5321 , M 56 = 0.5321 .

49
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Table 2. Poles and zeros of CQTfilter fast convergence of the cost function to a local minimum. In
№ Reflection Transmission zeros Roots of VN both cases this local minimum is found to be a global
zeros
1 j0.9892 -j2 j0.9566
minimum corresponding to general Chebyshev filter. In both
2 j0.9018 -j1.2 j0.8244 cases the starting point for the optimization process was the
3 j0.7246 j1.6 j0.6032 coupling matrix of classic Chebyshev filter. Starting from
4 -j0.9915 j2 -j0.9655 random initial point leads to a local minimum not
5 -j0.9201 -j0.8537 corresponding to Chebyshev filter.
6 -j0.7647 -j0.6530
7 -j0.5203 -j0.3696 IV. CONCLUSION
8 j0.4625 J0.3060 This paper presents optimization method for synthesis of
9 -j0.2057 -j0.0342
microwave filters with arbitrary topology of high order. The
10 j0.1385
method uses local optimization method for coupling matrix
The number of the independent values of the coupling determination. The synthesis procedure converges very fast as
matrix is 24 The optimization process converges fast in 238 for an initial point is used a vector based on the Chebyshev all
iterations of the optimizer with end cost function value pole filter for the same degree of the filter. To validate the
1.64519.10−7 . Fig. 6 shows the cost function value with respect proposed synthesis method two resonant filters are designed
to the iterations. with asymmetrical responses. Both presented examples show
fast convergence of the cost function to a local minimum. In
both cases this local minimum is found to be a global
minimum corresponding to general Chebyshev filter. The
frequency responses from the synthesis procedure are within
the expectations and found to be consistent with the
theoretical responses and given filter specifications.
REFERENCES
[1] A.E. Atia and A.E. Williams. "Narrow-Bandpass Waveguide
Filters." 1972 Trans. оn Microwave Theory and Techniques 20.4
(Apr. 1972 [T-MTT]): 258-265
Fig.6 Cost function value for asymmetric five resonator filter [2] Cameron, R., Advanced Coupling Matrix Synthesis Techniques
The coupling matrix derived in the optimization process is for Microwave Filters, IEEE Trans on MTT-50, Jan.2003, pp.1-10
given by (8). The corresponding frequency response [3] Cameron, R.J., General Coupling Matrix Synthesis Methods for
calculated by the coupling matrix and Eq.(6) is shown on Chebyshev Filtering Functions, IEEE Trans. On MTT, April 1999,
Fig.7. pp.433-442
[4] Rhodes, J.D., The Design and Synthesis of a Class of Microwave
Bandpass Linear Phase Filters, IEEE Trans on MTT 1969 pp.189-
204
[5] Macchiarella, G, An Analytical Technique for the Synthesis of
Cascaded N-Tuplets Cross-Coupled Resonators Microwave Filters
Using Matrix Rotations, IEEE Trans. On MTT May. 2005 pp.1693-
1698
[6] G. Pfitzenmaier, “Synthesis and realization of narrow-band
canonical microwave bandpass filters exhibiting linear phase and
transmission zeros,” IEEE Trans. Microwave Theory Tech., vol.
MTT-30, pp.1300–1311, Sep. 1982.
[7] Atia W.A., K.A. Zaki and A.E. Atia. "Synthesis of general
Fig.7 Frequency response of 10-th order CQT filter with
topology multiple coupled resonator filters by optimization." 1998
asymmetric response. Solid line-S21, dashed line- S11
MTT-S International Microwave Symposium Digest 98.2 (1998Vol.
As it is clearly seen from Fig.7, the transmission zeros are II MWSYM]): 821-824
placed on the prescribed values The maximum value of the [8] Amari, S., Synthesis of Cross-Coupled Resonator Filters Using
reflection coefficient is -20dB. Both presented examples show an Analytical Gradient-Based Optimization Technique, IEEE Trans
on MTT Sept. 2000, pp.1559-1564

 0 0.9832 0 0 0 0 0 0 0 0 0 0 
 0.9832 −0.0019 0.6244 −0.5149 0 0 0 0 0 0 0 0 
 
 0 0.6244 0.7140 0.4020 0 0 0 0 0 0 0 0 
 
 0 −0.5149 0.4020 −0.0802 0.5442 0 0 0 0 0 0 0  (8).
 0 0 0 0.5442 −0.0293 0.5266 0 −0.0427 0 0 0 0 
 
 0 0 0 0 0.5266 −0.012 0.5671 0 0 0 0 0 
M = 
0 0 0 0 0 0.5671 0.002 0.5267 0 0 0 0
 
 0 0 0 0 −0.0427 0 0.5267 0.0157 0.5421 0 0 0 
 
 0 0 0 0 0 0 0 0.5421 0.0563 0.5093 0.3348 0 
 0 0 0 0 0 0 0 0 0.5093 −0.4770 0.7368 0 
 
 0 0 0 0 0 0 0 0 0.3348 0.7368 −0.0019 0.9832 
 0 0 
 0 0 0 0 0 0 0 0 0 0.9832

50
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Random High Voltage Impulses

Modeling for EMC Testing
Kliment Angelov1 and Miroslav Gechev2
Abstract – In this paper an analytic method for random high steep climbing forefront and gradually fading rear front of the
voltage impulses generating for EMC testing has been described. current in time. Standards have been adopted for different
A statistical assessment of their distribution toward the electrical types of effects of lightning strikes, specifying the time
charge has been made in order to compare with the real physical duration of these fronts and amplitude value of current. Some
surges and electric discharges.
of these are:
Keywords – EMC, LEMP, Surge Protection. • 200 kA, 10/350 µs – for the first positive and negative
direct hit of lightning to the ground;
• 50 kA, 0,25/100 µs – for subsequent hits;
• 10 kA, 8/20 µs – for secondary lightning strikes and
I. INTRODUCTION switching surges.
Achieving high reliability of modern communication For mathematical models describing the shapes of pulses
systems imposes the requirement for trouble-free operation in exponential dependencies can be used. In work [3] the
harsh environmental conditions. Much of this environment is following formula is quoted:
dependent on natural factors. For example, a condition is
imposed to ensure a device temperature is maintained by air
conditioning systems or in certain cases, to take measures
against damage from moisture, strong wind, icing, I (t ) = I m k e −α t − e − β t
( ) (1)
earthquakes etc. For analytical studies of such processes is
appropriate to create relevant models and to implement and
comply with standards to integrate as well as technical
implementation and operation of the apparatus, and in where:
simulating the processes of change resulting from natural I(t) – current value of the current;
factors, especially in construction phase. Im – amplitude value of the current;
When it comes to protection of electronic communication k – normalized (scaled in amplitude) coefficient.
equipment from high voltage disturbances and effects of The shape of the curve is determined by the coefficients α
lightning strikes requirements are also established. Such are, and β.
for example, standards of the IEC [1] [2]. Different In this study an attempt is made to achieve the modeling of
mathematical models of high disturbing pulses are considered the stochastic nature of high-voltage disturbing pulses, as
[3] [4], in order to correctly simulate the processes “in case of using a step-exponential function of the type [6]:
LEMP” and achieve higher level of protection of electronic
communication devices [5].
The purpose of this article is to evaluate the statistical
ability to generate random high disturbing pulses with a I (t ) = I m .a.t b .e − c.t . (2)
specific mathematical model. The model could be used to
simulate the behavior of communication equipment in
situations when taking into account the stochastic nature of
the parameters of this type of interference. In this case the shape of the curve depends on the
coefficients b and c, and as using the coefficient a the
amplitude of the pulse can be scaled. By changing these
coefficients is possible to obtain the shapes of the curves of
II. DESCRIPTION OF THE PROBLEM the current corresponding to the standard 10/350 µs,
0,25/100 µs and 8/20 µs, which is discussed in other
A. Theoretical basis publications of this team [7].
To take into account the random nature of high-voltage
High-voltage disturbing pulses are characterized by very phenomena in nature a random number generators can be
used. These generators set values to the coefficients of the
1
Kliment Angelov is with the Faculty of Telecommunications at mathematical models. Changing the form of pulses, on one
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, hand leads to a difference in overall charges of each one, and
Bulgaria, E-mail: kna@tu-sofia.bg. on the other hand shifts the spectral distribution of the
2
Miroslav Gechev is with the Faculty of Telecommunications at harmonic composition.
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000,
Bulgaria, E-mail: miroslav.gechev@gmail.com

51
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

B. Test Implementation pulses that are unsymmetrically distributed to the mean value.
In such cases the value of the coefficient of variation is
Ability to generate random pulses can be used in cases of significantly greater than 30%, so the normal (Gaussian)
simulation study of the behavior of the electronic distribution is not applicable. Solution can be obtained by
communication equipment under close-to-real conditions. In using the distribution of Veybul, which can be described as:
such case it is necessary pulses to be generated with a
maximum close distribution as occurred in nature.
To check whether the model (2) is able to generate  0, q < qmin
trustworthy random pulses a statistical methods applied to the 
sample of the generated pulses can be used. Preliminary

f (q ) = 
. (8)
 k   q − q 
k1 −1
  q − q  k1 
studies performed on model (1) showed that it doesn’t give  .1 min
 . exp −  min
 , q > qmin
good results in a similar direction.  k 2   k 2    k2  
  
The generated pulses can be compared in their charge,
which can be determined by the following relation:
The values of coefficients k1 and k2 are determined by the
∞ following formulas:

Q= ∫ I (t ).dt ,
2
 q   q 
(3) k1 = 0,111186 + 0,835597  + 0,0759898  ; (9)
σ  σ 
t =0  q  q
1
1 n  k1

as obtained by (3) charge is in coulombs. For the purpose of k 2 =  ∑ q kj1  . (10)

comparative study we can work with normalized charge, so n  j =0 
the amplitudes of all pulses are equal. It can be assumed that:

The probability of occurrence of pulses defined or greater

amount of charge can be given with the dependency:
I (t ) max = 1 , (4)

 1, q < qmin
which can be achieved by using the coefficient a. In this case


the unit charge will be for 1 coulomb per ampere (C/A), but it P(q ) =   k1
 ;
can be denoted q. 
exp−  q − q  (11)
min
  , q q
   k 2  
A statistical evaluation of the resulting set of normalized > min
charges has to be made. For this purpose we calculate the
average value, the dispersion (variation) and the coefficient of
  
variation respectively on dependencies (5) (6) and (7), where
n is the number of random pulses in the sample.

III. RESULTS
n
1
q= ∑ qi
n i =1
(5)
Using the software environment MATLAB [8] simulations
of the probability distribution obtained with the dependence
(2) models of high-voltage disturbing impulses were made.
Using a random number generator values are given to the
1 n
σ q = ∑ [ (qi − q ) ] (6)
coefficients b and c, as Table I indicates the intervals of their
variation.
n i =1
TABLE I
σq COEFFICIENT VALUES
vq = (7)
q Coefficient 10/350 µs 8/20 µs Random

b 0,145 2,78 0,145 ÷ 2,78

To reflect plausible the actual distribution of occurrence of
impulses to their charge, the model should generate random c 0,00325 0,26 0,00325 ÷ 0,26

52
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

In Table I the values of the coefficients b and c, in case of electrostatic discharge, that breakthrough occurs when they
pulses corresponding to the standard 10/350 µs and 8/20 µs, reach a certain critical value of the charge.
are shown. They are the boundary and the distribution of
random values in these intervals is given by equal probability
law.
Fig. 1 shows a family of randomly generated curves, for
clarity of the image their number was reduced to 10. There are
variously shaped pulses with different slope of the forefront
and rear front respectively integrand area, respectively,
different relative charge.

Fig. 2. Probability Density Function

There is a cluster of pulses with relative charges in a

specific area, fairly close to the minimum, but there are a few
others with significantly higher charges than average. This
corresponds to the actual distribution of the disruptive signals
in nature - in most cases disturbing impulses are relatively
weak, but there also are extremely strong discharges (e.g.
Fig. 1. Random Generated Impulses direct hit by lightning), under which the charge is significant.
Fig. 3 shows a graph of the probability distribution function
To achieve credible data on the statistical analysis is of the relative charges. It can determine the probability of
necessary to sample a sufficiently large number of elements. occurrence of disturbing impulse with relative charge greater
In Table II are given values of average relative charge, than or equal to a certain value.
dispersion (variation) and coefficient of variation when
n = 10 000, and minimum and maximum value of the relative
charges.

TABLE II
STATISTICAL PARAMETERS

Parameter Value Unit

Mean Value 5,27 x10-5 , C/A

Standard Deviation 4,73 x10-5 , C/A

Coefficient of Variation 89,7 %

Fig. 3. Distribution Function
Number of impulses 10 000 -
By using these results a pattern of disruptive impulses
Minimum value 5,46 x10-6 , C/A generated by lightning activity in certain geographic areas can
be obtained. In Fig. 4. is presented a map showing the density
Maximum value 1,09 x10-3 , C/A of lightning activity worldwide, which was published in
George M. Kauffman’s [9] work about similar topic . A
similar map is shown in the corporate material of SPINNER
GmbH [10]. It shows the frequency of lightning falling,
Figure 2. shows the density of the probability distribution
measured in number of lightenings per square kilometer per
determined by formula (8) and the values of Table II. There is
year. It is seen that the highest density regions are Central
a minimum threshold of the relative charge below where no
Africa, the Himalayas, the Caribbean and Indochina, where
pulses are generated. From the perspective of a real physical
the density of lightning hits is in the range 50 ÷ 70
process, this can be explained by the nature of the occurrence
lightnings.km-2.year-1. For Bulgaria, this density is in the
of disturbing impulses. In most cases they are the product of
range 6 ÷ 10 lightnings.km-2.year-1.

53
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

probability of strong interference (e.g. direct hit by lightning)

is significantly less than that of the weaker ones. On the other
hand these Thunderstorms data have very general nature. It
strongly depends on topography, soil and other factors and for
more accurately determine the probabilities of occurrence of a
specific impulse with relative charge more accurate statistics
on lightning activity at the site should be used.

IV. CONCLUSION
From the foregoing it can be drawn:
Fig. 4. Worlds Lightning Strike Density - The mathematical model of high-voltage disturbing
pulse described by dependence (2) can be used to
Using the diagram of Fig. 3 the number of pulses with a generate random pulses to take into account stochastic
relative charge greater than or equal to a value per square character in nature;
kilometer within one year can be determined. In Table III are - The probability distribution of occurrence of disturbing
given the relative charges of the three main types of disturbing pulses to a particular relative charge meets the reality;
impulses. The same table gives the average number of - It is appropriate to experimentally confirm the
impulses exceeding that charge per square kilometer within simulation results, and compared with actual statistics
one year and the average time between the occurrence of two on lightning activity and its consequences.
pulses with exceeding relative charges per square kilometer.
These values are calculated for Bulgaria.

TABLE III
REFERENCES
YEAR LIGHTNING ACTIVITY
[1] IEC 60060, High-voltage test techniques.
[2] IEC 61312, Protection against LEMP.
10/350 µs 0,25/100 µs 8/20 µs [3] Maceika K., Lightning Protection of Electronic Data Processing
Systems, Scientific Proceedings of RTU. Series 7.
Relative Telecommunications and Electronics, 2003, vol.3.
charge, 4,399 1,326 0,166 [4] Gamlin M., Impulse current testing, Lightning Protection Forum
Shanghai, June 2004.
x10-4 C/A
[5] Hasse P., Overvoltage protection of low voltage systems, The
Institution of Electrical Engineers, London, 2000.
Number of [6] Каменов О., Висша математика, Част І, Издателство “Ciela”,
impulses for София, 2001.
0,002 0,7 5,8
1 km2 per [7] Angelov K., Model of Highvoltage Interferential Impulses,
year International Scientific Conference UNITECH 2010,
Proceedings vol. 1 (pp 286), 19-20 November 2010, Technical
Time between University Of Gabrovo.
[8] MATLAB IMAGE PROCESSING TOOLBOX. User's Guide,
impulses for "The Math - Works Inc.", 2000. www.mathworks.com.
182 500 521 62
1 km2, [9] Kauffman G.M., Using commercial lightning protectors in
days defense applications, RF Design Magazine, 2005, pp.16-21,
www.rfdesign.com.
[10] Spiner GmbH, Muenchen, Deutschland, www.spiner.de.

It is noted that the values for the period of occurrence of the

disturbing pulses is large. On one hand, due to the fact that the

54
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Optimum Divergence of the Transmitter Optical

Radiation in FSO Systems
Tsvetan Mitsev1, Nikolai Kolev2, Hristo Ivanov3 and Kalin Dimitrov4
Abstract – The determination of the optimum divergence of the coherent action in order to decrease the connection channel
transmitter optical beam θt,opt in FSO systems can largely length or increase the outage in case of poor weather
compensate for the negative impact of the change in the direction conditions. A typical way of overcoming this problem is the
of propagation of the optical radiation due to various random use of redundant power with a perfect optical setting of the
factors. Depending on the system parameters, the length of the
system and when there is a possibility that there are only
communication channel and the typical weather working
geometrical losses, which we have considered in our paper
conditions, the proper choice of θt can significantly increase the
reliability of information transmission and reduce the [1]. In it, we have derived an expression for calculating the
probability of outage. In this paper the influence of the optical maximum radial displacement of the receiver antenna center
output power of the transmitter and the length of the from the transmitter laser beam axis, depending on the initial
communication channel on the value of optimum divergence of Gaussian beam radius.
the laser beam after the transmitting antenna are shown. When This paper is a continuation of [1]. We have proven the
the divergence of the transmitter beam is set, the FSO system of significance of a transmitter optical antenna with adjustable
TU-Sofia can work reliably under conditions where the angle width of the transmitter diagram in order to increase the
deviations of the beam from its main direction exceed more than functioning reliability of the system. We have researched the
twice the deviations in the absence of adjustment.
impact of the optical radiation source power and the
Keywords – Communications, Free Space Optics, Laser Beam, connection channel length on the value of the optimum
Diverging Angle, Beam Wander divergence angle of the transmitter optical radiation. We have
indicated the basic parameters of the system and the
connection channel.

I. INTRODUCTION
II. OPTICAL PROPAGATION AND INTENSITY
The application of FSO systems is becoming more and DISTRIBUTION. DEFINITION OF THE PROBLEM
more frequent with specific connection conditions in the
contemporary communication systems and networks. This is In the selected location of the FSO system and a perfect
due to their wide bandwidth, tight radiation pattern of antenna, optical setting, that is a coincidence of the optical antennae
small size and weight, lower price, license free frequency axes of the opposite transmitter/receiver sets, angle θ = 0. The
band, that is, no frequency planning is necessary. The BER value with a perfect setting usually reaches values lower
increased interest in FSO systems, however, creates new than 10-20, when the values for normal functioning of the FSO
requirements for improvement of their characteristics, as well systems are within the range of 10-12 to 10-8. This allows,
as for optimizing some of their parameters, in particular those when the source power remains the same, for an increase of
of the divergence of the transmitter optical radiation [1-4]. the divergence of the transmitter θt, and in this case there is an
One of the reasons for decreasing the functioning reliability increase in the value of the maximum acceptable angle
of FSO systems are the random angle fluctuations of the deviations θmax of the laser beam from its main direction when
transmitter laser beam from the direction where the receiver is the condition is fulfilled that the received power Φr is bigger
placed. The main reasons for their existence are the turbulent than the threshold value Φr, min, respectively the minimal
fluctuations in the atmosphere and the mechanical movements average radiation intensity in the receiver aperture Ιr is bigger
of the bases on which the transmitter/receiver sets are placed than Ir, min (fig.1). With the further increasing of θt we reach
(or building sway) [5-7]. The phenomena mentioned have the maximum value of θmax when the installation and
parameters of the system are fixed, and then θmax starts
1
decreasing and we derive Φr < Φr, min , including the case
Tsvetan Mitsev is with the Faculty of Telecommunications at where the angle is θ = 0.
Technical University of Sofia, 8 Kl. Ohridski Blvd., Sofia 1000,
Bulgaria, E-mail: mitzev@tu-sofia.bg. As it is evident fig.1, for the derivation of the optimum
2
Nikolai Kolev is with the Faculty of Telecommunications at laser beam divergence of the transmitter θt, opt, where on
Technical University of Sofia, 8 Kl. Ohridski Blvd., Sofia 1000,
certain conditions we derive the maximum value of θmax, we
Bulgaria.
3
Hristo Ivanov is with the Faculty of Telecommunications at
need an intensity distribution model of the light of the source
Technical University of Sofia, 8 Kl. Ohridski Blvd., Sofia 1000, in the receiver antenna plane. This means that at a distance z
Bulgaria. from the transmitter in a plane transverse to the distribution
4 with a assumption for azimuthally beam symmetry, we have
Kalin Dimitrov is with the Faculty of Telecommunications at
Technical University of Sofia, 8 Kl. Ohridski Blvd., Sofia 1000, to derive the radial distribution of the plane density of the
Bulgaria, E-mail: kld@tu-sofia.bg. power I(ρ, z) ≡ I(θ, z). This distribution depends mainly on the

55
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

and typically Kθ ≥ 10.

-3
Ι(θ,z) [W/m2]

4,0.10 When the conditions (1) and (2) are fulfilled, the current
3,5.10-3 Gaussian radius is calculated by the formula

3,0.10-3 θt,1<θt,2<θt,3<θt,4 ρ z ( z ) = ρ 02 + (θ t,exp z ) 2 (6)

z = const
2,5.10-3 θt,2 and
2,0.10 -3
I ( ρ z , z ) = I ( 0 , z ).e −2
θt,3
1,5.10-3 is fulfilled.
1,0.10-3 θt,4 The losses in the light distribution between the transmitter
Ir,min and the receiver when there is a assumption of an uniform
5,0.10-4
volume extinction coefficient αe, are calculated by the
0 formula
0 0,2 0,4 0,6 0,8 1,0 1,2 1,4
θmax θ [mrad] τa = e −αez ,
(7)
−q
 λ [ µm ] 
Fig.1. Dependence of the radial distribution of intensity of optical
radiation I(θ, z) in the plane of the receiver (z = const) at different [
α e km −1 = ] 3, 92 



.
divergences of the transmitter optical radiation θt (θmax is the maxi- S M [ km ]  0 , 55 
mum acceptable angular variance of transmitter’s beam from its
main direction in case of θt, 4). In (7) τa is the transparence of the connection channel, SM is
the meteorological visibility of the atmosphere, and for the
typical atmospheric conditions the exponent q is calculated by
phase and amplitude distribution of the field in the emitting the formula
aperture At = πRt2. Rt is the aperture radius of the transmitting
antenna. In our model we will use synchronous phase and q = 0 , 585 3 S M [ km ] .
Gaussian amplitude distribution in the emitting aperture [1],
[8]. The maximum intensity value is along the beam axis, With the assumptions made, the optical radiation intensity
respectively in the center of the emitting aperture. When along the optical axis and its radial distribution are
ρ = ρ0, the light intensity decreases by e2 in relation to the 2 τ t τ a ( λ 0 , S M , z )Φ L
maximum and ρ0 is defined as an initial Gaussian beam I ( 0, z ) = ,
radius. In order to keep the Gaussian radial distribution in the πρ 2z ( z )
Fraunhofer zone 2 (8)
 θ 
− 2  

θ
z ≥ zc, exp (1) I ( ρ, z ) ≡ I ( θ, z ) = I ( 0, z ) .e  t,exp  .
it is necessary to fulfill the condition
In (8) the losses in the transmitter antenna optics have been
Rt ≥ 2ρ0 . (2) expressed by τt, ΦL is the power of the source radiation, θ is
the angle deviation of the transmitter optical beam axis,
zc, exp is calculated by the formula recognized from the case of perfect alignment, that is
10 ρ 02 coincidence of the optical axes of the transmitter and opposite
z c,exp = , (3) receiver optical antenna. The power ΦL is the laser power with
Kθλ0 a assumption for a digital communication system with On/Off
modulation (OOK) in the optical code impulse.
where λ0 is the central wavelength of the light source,
With the digital communication systems with OOK
Kθλ0 modulation, the system functioning quality is guaranteed by
ρ0 = , (4)
the low values of BER. With them, we calculate BER from
πθ t,exp
SNR again using an erfc function, which presupposes a great
and Kθ is the coefficient indicating the random fluctuations of slope in the changing of BER. A change by one order of SNR
the field phase in At. These fluctuations are due to different leads to a change up to ten orders of BER. Because of that it is
stochastic factors in the laser generation, and these factors more convenient to deal with and to represent graphically the
worsen the radiation coherence level and lead to a difference change of SNR from the different parameters of the system.
between the actual divergence and the theoretically defined To calculate SNR we need the optical beams at the input of the
receiver. The optical beam through the input aperture of the
one θt, teor,
receiver corresponding to the upper level of the optical code
θ t,exp = K θ θ t, teor , (5) impulse, is

56
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Φ pd ( θ , z ) = π. τ r . R r2 . I ( θ , z ) . (9) 1000

SNR
In (9) Rr is the aperture radius of the receiver telescope, τr is
the transmission coefficient of the optical receiver system. θt,1<θt,2<θt,3<θt,4
The above equation is true when the condition ρ z ( z ) >> R r 100
is fulfilled. θt,2
θt,3
The second in significance input optical beam, that is the
background one, is calculated by the formula SNRth θt,4
10
Φ B = π 2 τ r L λ ,B ( λ 0 ) R r2 θ 2r,exp ∆λ F , (10)

where Lλ, B is the spectral brightness of the background

radiation, and ∆λF is the transmission wavelength bandwith of
the interference filter before the photodetector, placed to 1
restrict the background radiation. 0 0,2 0,4 0,6 0,8 1,0 1,2 1,4
With the indication of the dispersion of the two main types θ [mrad]
of noise in the optical receivers, the thermal and the quantum θ1,max θ3,max
one, the expression for SNR calculation is Fig.2. Dependence of the signal to noise ratio from angular
R I ( λ 0 ) Φ pd ( θ , z ) deviation θ of the beam of the transmitter from its main direction at
SNR = .(11) different divergence of transmitter optical radiation θt.
 2 k T . A 
CI  B + eR I ( λ 0 ) [Φ pd ( θ , z ) + Φ B ]
 R Fb  beginning, and then it starts decreasing, as we have already
predicted.
The formula is true for an optical receiver with The dependence θmax(θt) for three values of the optical
preamplification and a p – i – n photodiode. radiation source power ΦL = [10, 15, 20] mW with connection
channel length z = 2 km has been shown in fig.3. It is evident
R I ( λ 0 ) = 8 , 06.10 5 η ( λ 0 )λ 0 is the integral sensitivity for
current of the photodetector, η(λ0) is the quantum efficiency 1,4
ΦL=20mW
θmax [mrad]

of the photodetector material, kB is the Boltzmann constant, e

is the charge of the electron, CI is the information throughput 1,2
of the digital communication system, and RFb is the value of ΦL=15mW
the resistor in the feedback of the preamplifier. 1,0

0,8 ΦL=10mW
III. SIMULATION RESULTS AND DISCUSSIONS
0,6
For the developed and implemented in TU-Sofia FSO
system [9], [10] we will determine the maximum divergence 0,4
θt, opt of the transmitter optical beam. The system works at a
wavelength λ0 = 850 nm with information throughput CI = 0,2
100 Mbps with power in the optical bit impulse ΦL = 10 mW.
Using a two-lens Kepler collimator, we gradually change the 0
beam divergence within the range of 1 mrad to 5 mrad. The 0 0,5 1,0 1,5 2,0 2,5 3
connection channel length is up to 2 km. The other system θt [mrad]
parameters necessary for the calculation using the method θt, opt
developed in ІІ, are: τt = 0,85; Kθ = 10; Rr = 5,5 cm; θr = 5
mrad; τr = 0,85; η(λ0) = 0,7; ∆λF = 10 nm; RFb = 1 kΩ; A = 5. Fig.3. Dependence θmax (θt) at z = 2 km for three values of ΦL.
For the calculations we choose values SM = 10 km, Lλ,B = 10-2 Determination of θt, opt (ΦL = 10 mW).
W/m2.sr.Å, T = 300 K, and the constants are kB = 1,38.10-23
J/K, e = 1,602.10-19 C. that when the power ΦL increases, θmax increases, too. From
the graphics it is evident that if we want to have the maximum
In fig.2, with an increasing divergence θt, exp of the value of θmax, it is necessary to change θt too, this means that
transmitter beam, we have shown the dependence SNR(θ). It its optimum value exists and it is θt, opt. When ΦL
is evident that, when we choose a minimal level for the increases two times and with an optimum value of the
signal/noise ratio SNRth = 11,2, which corresponds to BER ≈ transmitter optical beam divergence, the maximum possible
10-8, the maximum possible divergence of the beam θmax from angle beam divergence increases by 37%. It is also evident
the perfect alignment increases with the increasing of θt in the

57
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

from the graphics that θmax, depending on ΦL, undergoes more that the values θmax(θt, opt) increase when ΦL increases and
significant changes with the great values of θt. they decrease when z increases, and the connection channel
length z has a greater impact on them. When there is a
2,5 constant collimation of the transmitter beam, that is a constant
θmax [mrad]

value θt, the value of θmax is influenced to a much greater

z = 1km
extent by z and ΦL when the values of θt are big than when
2,0
they are small, for instance when θt ≤ 1 mrad. When there is
an optimum beam divergence setting, within the limits of the
1,5 research defined (ІІІ) it is possible to have a 121% increase in
z = 1,5km the acceptable value of the divergence θmax.

1,0 z = 2km
ACKNOWLEDGEMENT
0,5
This work has been supported by the Research Programme
of the Technical University of Sofia, Bulgaria, Internal
0 Projects 2011, contract Nb 112pd033-07.
0 1 2 3 4 5
θt, opt θt [mrad]
REFERENCES
Fig.4. Dependence θmax(θt) at ΦL = 10mW for three values of z. [1] Ts. Mitsev, K. Dimitrov, B. Bonev, “Influence of Laser Beam
Determination of θt,opt(z = 1 km). Divergency on Free Space Optic Systems Functionality”,
TELECOM’2008, Conf. Proc., Varna, Bulgaria, 2008.
[2] Zh. Zhao, R. Liao, Y. Zhang, “Impact of Laser Beam Deverging
In fig.4 is shown the dependence θmax(θt) for three Angle on Free-Space Optical Communications”, Aerospace
connection channel lengths z = [1, 1,5, 2] km when the optical Conference, IEEE, pp. 1-10, 2011.
[3] A. Farid, S. Hranilovic, “Outage Capacity Optimization for
radiation source power is ΦL = 10 mW. With the decreasing Free-Space Optical Links with Pointing Errors”, Journal of
of the distance z is necessary a significant readjustment of the Lightwave Technology, Vol. 25, Issue 7, pp. 1702-1710, 2007.
transmitter optical system, but as a result we can achieve a [4] Y. Ren, A. Dang, B. Luo, H. Guo, „Capacities for Long-
significant improvement of the functioning abilities of the Distance Free-Space Optics Links Under Beam Wander
system. When z is decreased 2 times, it is necessary to Effects“, Photonics Technology Letters, IEEE, vol. 22, issue 14,
increase θt by almost 3 times in order to maintain the optimum pp. 1069-1071, 2010.
setting of the system. As a result, however, the possibilities of [5] Shlomi Arnon, “Effects of Atmospheric Turbulence and
divergence of the beam from the main direction and keeping Building Sway on Optical Wireless Communication Systems”,
Opt. Lett., vol. 28, No. 2, pp. 129-131, 2003.
the functioning of the system, are more than 2,2 times greater. [6] E. Ferdinandov, B. Pachedjieva, B. Bonev, Sl. Saparev, “Joint
In the comparisons between fig.3 and fig.4 it is evident that Influence of Heterogeneous Stochastic Factors on Bit-Error
the functioning of the system is more sensitive to the change Rate of Ground-to-Ground Free-Space Laser Communication
Systems”, Optics Communications, vol. 270, issue 2, pp. 121-
in the connection channel length that it is to the optical
127, 2007.
radiation source power. When the values of the optical [7] Bonev B., Relative Influence of Some Stochastic Factors on Bit-
radiation divergence are θt < 1 mrad, the impact of the Error Rate of Ground-to-Ground Free Space Optics, ICEST
changing of z or of ΦL on θmax can be ignored. 2007, Vol. 1, pp. 203-206, Ohrid, Macedonia, 2007.
[8] E. Ferdinandov, B. Pachedjieva, K. Dimitrov, Optical
Communication Systems, Sofia, Technika, 2007.
IV. CONCLUSION [9] Ts. Mitsev, N. Kolev, K. Dimitrov, “Optical Wireless
Communication System”, XII-th International Scientific
This paper shows the possibility of a significant increase in Conference SMOLJAN-2010, Smoljan, Bulgaria, 26-27 June
the functioning and reliability of an FSO system with an 2010.
optimum optical radiation divergence setting of the transmitter [10] N. Kolev, “Selection of optimal settings depending on the FSO
system parameters", XIII International PhD Workshop OWD
θt, opt. Its value depends on the particular parameters of the
2011, Conf. Proc., vol. 29, pp. 467-472, Warsaw, 2011.
system and the communication channel. We have researched
the impact of the connection channel length z and the power in
the code impulse of the optical radiation of the source ΦL on
the maximum possible divergence θmax of the transmitter
beam from the perfect direction, that is when there is a
location on single optical axis of the opposite
transmitter/receiver antennae θ = 0 (fig.2). We have shown

58
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Body Bias Influence on Ring Oscillator Performance for

IR-UWB Pulse Generator in 0.18μm CMOS technology
Jelena Radic, Alena Djugova, Laszlo Nagy and Mirjana Videnovic-Misic
Abstract – A CMOS standard three-stage ring oscillator is bandwidth is subdivided into two bands: 3 − 5 GHz (lower-
examined in UMC 0.18μm technology. The ring oscillator band) and 6 − 10.6 GHz (higher-band).
performance dependence on the bulk (substrate) resistors One of the most critical components of an UWB system is
introduced into inverter PMOS and NMOS transistors is the pulse generator, which has to be designed with relatively
investigated. Simulation results showed that the ring oscillator
frequency is strongly dependent on the substrate resistor value.
low-complexity and low power consumption. Moreover,
This fact can be used to increase the ring oscillator frequency. As generated pulse train spectrum has to satisfy the FCC spectral
the ring oscillator is a part of an IR-UWB (Impulse Radio Ultra mask, making pulse generator design very challenging. There
Wide Band) pulse generator, its oscillating frequency determines are several typical techniques for designing it which usually
the central frequency of the pulse spectrum and have significant follow all-digital [5] − [6], analogue-digital [7] or all-analogue
effect on spectrum fitting within UWB FCC mask. [7] − [8] design approach. Digital solutions offer higher
integration, lower consumption and better controllability
Keywords – Body biasing, bulk resistor, CMOS process, while all-analogue techniques demonstrate circuit simplicity.
impulse radio ultra-wideband (IR-UWB), ring oscillator. As an essential part of an analogue-digital pulse generator
[5], a ring oscillator is studied in this work. To increase ring
oscillating frequency, resistors are introduced between the
I. INTRODUCTION bulk (substrate) transistors terminals and appropriate voltage
terminal (ground in case of NMOS transistor and Vdd in case
of PMOS transistors). Dependence of the ring oscillator
Impulse Radio Ultra-Wide-Band (IR-UWB) technology has
frequency on the value of the bulk resistors inserted in the
emerged as a potential solution for very high data rate short-
NMOS, PMOS, and both kind of transistors is examined. The
range communication, and low data rate communication
related to localization, targeting both low cost and low power ring oscillator is designed and simulated in 0.18μm CMOS
consumption [1] − [3]. It transmits extremely short pulses, on technology.
the order of a nanosecond or less, which occupy a bandwidth
up to several GHz. Additionally, IR-UWB technology offers II. STANDARD THREE-STAGE RING OSCILLATOR
high fading margin for communication systems in multipath DESIGN
environments [3].
The American Federal Communications Commission The pulse generator represents a key block in the impulse
(FCC) defines a signal as ultra-wideband if it occupies more UWB communication. As the pulse shape determines the
than 500 MHz of radio frequency spectrum or exhibits a spectrum characteristic of the UWB signal and effectively
fractional bandwidth of at least 25% [4]. As the FCC allocated dictates specific system requirements, its generation is one of
frequency spectrum for UWB technology is 3.1 − 10.6 GHz, the essential considerations in the UWB design. Fig. 1 shows
the power level from the UWB transmitter should be small the basic topology of an IR-UWB transmitter based on ring
enough not to interfere with the already existing oscillator as a part of the pulse generator.
communication systems such as WiMax, Bluetooth and GSM. It consists of a glitch generator, a switched ring oscillator, a
This requirement limits output power level of UWB TXs at buffer stage and a pulse shaping (band-pass) filter [7]. The
−41.3 dBm/MHz [4]. In the GPS band (0.96 – 1.61 GHz), glitch generator turns the ring oscillator on/off approximately
there is even more stringent regulation: less than
−75.3 dBm/MHz is needed to avoid interference problems.
The PSD (Power Spectral Density) in frequency interval from
1.61 GHz to 3.1 GHz depends on the type of application
(indoor, outdoor, GPS, wall & medical imaging, through-wall
imaging & surveillance system). In spite of these regulations,
there have been many reports of interferences with wireless
local area network (WLAN) systems operating in the
5 – 6 GHz band. Due to practical reasons, the UWB

Jelena Radic, Alena Djugova, Laszlo Nagy, and Mirjana

Videnovic-Misic are with the Department of Power, Electronic and
Telecommunication Engineering, Faculty of Technical Sciences,
Fig. 1. An IR-UWB transmitter based on ring oscillator as a part of
University of Novi Sad, Trg Dositeja Obradovica 6, Novi Sad 21000,
pulse generator.
Serbia, E-mail: {jelenar_, alenad, lnadj, mirjam}@uns.ac.rs.

59
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

defining the duration of its oscillation and thus the width of proportionally, while the oscillating frequency decreases
the pulse generated at the oscillator output. The switched ring (f0=1/T), and vice versa. For the smallest NMOS and PMOS
oscillator frequency defines position of the transmitted pulse transistors (transistor width/length: W/L=25μm/0.18μm), the
spectrum within the FCC mask [8]. Since in the IR-UWB oscillation frequency of 3.95 GHz has been obtained. To
communication the time domain pulse width roughly utilize the whole UWB band more effectively, the center
determines the width of the generated frequency spectrum [8], frequency of at least 6 GHz is required. Higher ring oscillator
it is important to design a pulse, which makes optimal usage frequency could be achieved without PMOS transistor M5.
of the available spectrum, within the limits imposed by the However, this transistor provides the identical oscillation start
FCC. The buffer isolates the ring oscillator from the pulse from the initial state, defined by connecting A’ (B) node to
shaping filter loading and improves the pulse generator Vdd at the falling edge of the on-off signal.
current driving capability. The band-pass filter additionally
accommodates the pulse in the allowed FCC spectral mask. A. Influence of the NMOS and the PMOS transistors bulk
The switched ring oscillator topology is shown in Fig. 2. It resistors
is composed of the three-stage ring oscillator (M1 – M3
inverters stages) and a pair of oscillation-enabling switches
The bulk (substrate) terminals of the transistors are floated
(transistors M4 and M5). Due to its simplicity and short
to improve the performance of CMOS SPST (single pole
start-up time, the ring oscillator is the most used architecture single throw) and SPDT (single pole double throw) switches,
in the IR-UWB transmitter applications. It has small especially the power-handling ability and insertion loss, in
resistance at each feedback nod which allows fast transient
[9] − [10], respectively. The series transistor particularly
response.
played an important role in the switch’s insertion loss, while
The oscillation-enabling switches, as their name says,
the shunt transistors enhanced the isolation when the switch
control the oscillation process. When the on-off signal
was turned off. To improve the ring oscillator performance,
(produced by the glitch generator) is high, M4 is turned on the method that uses the resistors placed in the inverter stages
(M5 is turned off), the inverters stages M1 – M3 outputs have
transistors bulk terminals is investigated in the paper. First,
initial voltage values determined by the size ratio of the
the bulk resistors connected to ground are inserted in the ring
corresponding PMOS and NMOS transistors. Due to the small
oscillator inverter NMOS transistors, shown in Fig. 3.
inverter reactance, the oscillation can start immediately.
Transistor M5 is turned on (M4 is turned off) at the on-off
signal low level, connecting the M1 stage output (the M2 stage
input) to Vdd, and effectively shutting down the oscillations.
As the ring oscillator is switched off by M4 transistor during
the inactive period of time, the power consumption is
minimized.

Fig. 3. The ring oscillator with NMOS bulk resistors architecture.

Fig. 2. The three-stage ring oscillator design.

III. THE RING OSCILLATOR PERFORMANCE

The proposed designs have been simulated in mixed

mode/RF 0.18μm CMOS technology using SpectreRF
Simulator from Cadence Design System. Supply voltage Vdd
of this technology is 1.8 V. The main problem with the
standard ring oscillator design was limited set of transistor
sizes available in the used technology as the ring oscillator
working frequency depends directly on transistors sizes. If the
transistors are larger, the period of the oscillation T rises Fig. 4. Dependence of the ring oscillator performance on NMOS
transistors substrate resistors values.

60
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Dependence of the ring oscillator performance on the TABLE I

NMOS transistor bulk resistors Rb value is depicted in Fig. 4. INFLUENCE OF THE NMOS/PMOS TRANSITORS BULK RESISTORS
It can be noticed that with resistor Rb value increase, the ring VALUE ON THE RING OSCILLATOR FREQUENCY

oscillator frequency f0 rises, while the peak-to-peak amplitude

Bulk resistors in Bulk resistors in
at the ring oscillator output changes negligibly. This can be
NMOS transistors PMOS transistors
explained by the two effects. First cause, describing the
changes in the NMOS transistor threshold voltage VtN by Rb (kΩ) f0 (GHz) f0 (GHz)
change in the bulk-to-source voltage VBS, is of less concern. 0.4 4.10 4.0
This phenomenon is sometimes called the “back-gate effect”, 0.8 4.15 4.10
since the body influences the threshold voltage when it is not 1.5 4.20 4.20
tied to the source, being considered as the second or the 3.0 4.25 4.25
“back-gate”. The NMOS body effect upon threshold voltage 5.0 4.25 4.25
VtN is given by [11]

VtN = Vt 0 + γ ( 2φ F − VBS − 2φ F , ) (1)

4,23

where Vt0 is threshold voltage for zero VBS voltage, ΦF is the

4,18
Fermi level deep in the bulk, γ = (tox / ε ox ) 2qε si N A is the

f 0 (GHz)
4,13
body effect parameter, tox is oxide thickness, εox is oxide
Bulk resistors in NMOS transistors
permittivity, εsi is silicon permittivity, NA is a doping
4,08 Bulk resistors in PMOS transistors
concentration, and q is the charge of an electron. With
increase in the NMOS transistor bulk resistors value, due to
4,03
the small substrate current the NMOS bulk voltage becomes
slightly positive with respect to the source (the NMOS 3,98
body-source junction is forward biased), resulting in a low VtN 1 2 3 4 5
voltage drop. This leads to increase in the current discharging R b (kΩ)
the ring oscillator output and internal node capacitances. As a Fig. 6. Dependence of the ring oscillator frequency on the NMOS
result, the period of reaching the peak values is decreased and and PMOS transistors substrate resistors value.
the ring oscillator frequency is increased. Dominant effect
represents reduction in the bulk parasitic capacitance current oscillating frequency was increased from 4.1 GHz to
caused by increase in the substrate resistors value. With 4.25 GHz in case of the NMOS substrate resistors, while f0
decrease in the bulk parasitic currents, more current is parameter was in the range from 4 GHz to 4.25 GHz for the
available for the inverter output capacitance discharging and PMOS bulk resistors method. It can be noticed that for lower
thus the f0 parameter rises. substrate resistor values, the higher oscillation frequency was
In the second case, the resistors tied to Vdd are introduced as obtained in the former technique. Furthermore, the f0 – Rb
the PMOS transistor substrate resistors, Fig. 5. Since the curves go into saturation for the bulk resistor values higher
influence on the ring oscillator performance was nearly the than 3 kΩ in both cases.
same as presented in the simulation above (increase in the
PMOS bulk resistors reduces the PMOS threshold voltage VtP B. Influence of the substrate resistors placed in each ring
and/or decreases the substrate parasitic capacitance currents oscillator inverter stage transistor
leading to increase in the current charging the ring oscillator
output and internal node capacitances), the output voltage To further increase the ring oscillator frequency the bulk
waveform are not given here. However, simulated f0 values resistors are introduced in each inverter transistor, shown in
for both approaches are summarized in Tables I, and Fig. 7. Influence of the Rb value on the ring oscillator
presented in Fig. 6. Varying Rb from 0.4 kΩ to 5 kΩ, the

Fig. 5. The ring oscillator with PMOS bulk resistors architecture. Fig. 7. The ring oscillator with bulk resistor in all inverter’s
transistors topology.

61
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

TABLE II changes can be attributed to the two facts: the NMOS/PMOS

INFLUENCE OF THE TRANSITORS BULK RESISTORS VALUE ON THE transistor threshold voltage drop due to minor increase in the
RING OSCILLATOR FREQUENCY bulk voltage, and/or the parasitic currents reduction, with
increase in the bulk resistor value. Both effects cause increase
Rb (kΩ) f0 (GHz)
in currents available to charge/discharge the load capacitance
0.1 4.0 and thus the ring oscillator frequency. However, the latter
0.2 4.05 effect is much more dominant.
0.4 4.15
0.8 4.35
1.5 4.50 ACKNOWLEDGEMENT
3.0 4.55
5.0 4.60 This work was supported in part by the Ministry of
Education and Science, Republic of Serbia, on the project
number TR-32016.

4,58
REFERENCES
4,48
[1] M. Ghavami, L. B. Michael, and R. Kohno, Ultra Wideband
4,38 Signals and Systems in Communications Engineering, John
f 0 (GHz)

Wiley&Sons Ltd, 2004.

4,28 [2] K. Siwiak and D. McKeown, Ultra-Wideband Radio
Technology, John Wiley&Sons Ltd, 2004.
4,18
[3] J. R. Fernandes and D. Wentzloff, “Recent Andvances in IR-
UWB Transceivers: An Overview”, IEEE Int. Conf. on Circuits
4,08
and Systems, Conference Proceedings, pp. 3284−3287, Paris,
France, 2010.
3,98
0 0,8 1,6 2,4 3,2 4 4,8
[4] First Report and Order: Revision of Part 15 of the
Commission’s Rules Regarding Ultra-Wideband Transmission
R b (k)
Systems Federal Communications Commission (FCC), ET
Fig. 8. Dependence of the ring oscillator frequency on the substrate Docket 98-153, Adopted February 14, 2002, Released Apr. 22,
resistors value. 2002.
[5] V. V. Kulkarni, M. Muqsith, K. Niitsu, H. Ishikuro, T. Kuroda,
frequency is presented in Table II, and shown in Fig. 8. The “A 750 Mb/s, 12 pJ/b, 6-to-10 GHz CMOS IR-UWB transmitter
same f0 parameter dependence on the bulk resistor as in the with embedded on-chip antenna”, IEEE Jour. of Solid. State
previous simulations can be observed. Nevertheless, it should Circuits, vol. 44, no. 2, pp. 394-403, 2009.
be emphasized that the achieved ring oscillator frequency is [6] H. Kim, Y. Joo, S. Jung, “A tunable CMOS UWB pulse
considerably higher comparing to the initial fo value generator”, IEEE Int. Conf. on Ultra-Wideband, Conference
Proceedings, Waltham, MA, pp. 109-112, 2006.
(3.95 GHz, obtained in the topology presented in Fig. 2), and [7] S. Sim, D.W. Kim, S. Hong, “A CMOS UWB pulse generator
the values achieved in previous approaches for the same Rb for 6–10 GHz applications”, IEEE Microwave and wireless
values, as expected. components letters, vol. 19, no. 2, pp. 83-85, 2009.
[8] O. Novak, C. Charles, “Low-power UWB pulse generators for
biomedical implants”, IEEE Int. Conf. on Ultra-Wideband,
IV. CONCLUSION Conference Proceedings, pp. 778-782, Vancouver, BC, 2009.
[9] R. Xu, Y. Jin, C. Nguyen, “Power-Efficient Switching-Based
Standard three-stage ring oscillator topology has been CMOS UWB Transmitters for UWB Communications and
analyzed in 0.18μm CMOS technology. Dependence of the Radar Systems”, IEEE Transaction on Microwave Theory and
ring oscillator performance on the bulk resistors inserted in Techniques., vol. 54, no. 8, pp. 3271–3277, 2006.
the inverter NMOS, PMOS, and both kind of transistors has [10] M.C. Yeh, Z.M. Tsai, R.C. Liu, K.Y. Lin, Y.T. Chang, H.
been investigated. Simulations confirmed strong dependency Wang, “Design and analysis for a miniature CMOS SPDT
switch using body-floating technique to improve power
of the ring oscillator frequency f0 on the substrate resistor
performance”, IEEE Transaction on Microwave Theory and
value. The maximum frequency (4.6 GHz) obtained in case Techniques., vol. 51, no. 1, pp. 31–39, 2005.
the bulk resistors were introduced into each inverter stage [11] Thomas H. Lee, The design of CMOS radio-frequency
transistors is remarkably (16.5 %) higher than value achieved integrated circuits, Cambridge University press, 2004.
with the standard ring oscillator architecture. The f0 parameter
[12]

62
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Comparative Performance Studies of Laboratory WPA

IEEE 802.11b,g Point-to-Point Links
José A. R. Pacheco de Carvalho1,

Cláudia F. F. P. Ribeiro Pacheco2, Hugo Veiga3, António D. Reis4

Abstract –Wireless communications using microwaves are based on an access point, AP, which permits communications
increasingly important, e.g. Wi-Fi. Performance is a very crucial of Wi-Fi electronic devices with a wired based LAN through a
issue, resulting in more reliable and efficient communications. switch/router. Wi-Fi has penetrated the personal home, where
Security is equally very important. Laboratory measurements a WPAN allows personal devices to communicate. Point-to-
are made about several performance aspects of Wi-Fi IEEE
802.11 b,g WPA point-to-point links. A contribution is given to
point and point-to-multipoint configurations are used both
performance evaluation of this technology under WPA indoors and outdoors, requiring specific directional and
encryption, using available equipments (DAP-1522 access points omnidirectional antennas. Wi-Fi uses microwaves in the 2.4
from D-Link and WPC600N adapters from Linksys). Detailed and 5 GHz frequency bands and IEEE 802.11a, 802.11b,
results are presented and discussed, namely at OSI levels 4 and 7, 802.11g and 802.11n standards [1]. The 2.4 GHz band is
from TCP, UDP and FTP experiments, permitting measurements intensively used and is having increasing interferences.
of TCP throughput, jitter, percentage datagram loss and FTP Therefore considerable attention has been focused on the 5
transfer rate. Comparisons are made to corresponding results GHz band where, however, absorption increases and ranges
obtained for open links. Conclusions are drawn about the are shorter.
comparative performance of the links.
Nominal transfer rates up to 11 (802.11b), 54 Mbps (802.11
Keywords – WLAN, Wi-Fi, WPA Point-to-Point Links, IEEE a, g) and 600 Mbps (802.11n) are specified. The medium
802.11b, IEEE 802.11g, Wireless Network Laboratory access control is CSMA/CA. There are studies on wireless
Performance. communications, wave propagation [2,3], practical
implementations of WLANs [4], performance analysis of the
effective transfer rate for 802.11b point-to-point links [5],
802.11b performance in crowded indoor environments [6].
I. INTRODUCTION Performance has been a very important issue, resulting in
more reliable and efficient communications. In comparison to
Contactless communication techniques have been traditional applications, new telematic applications are
developed using mainly electromagnetic waves in several specially sensitive to performances. Requirements have been
frequency ranges, propagating in the air. Wi-Fi and FSO, pointed out [7]. E.g. requirements have been quoted as: for
whose importance and utilization have been recognized and video on demand/moving images, 1-10 ms jitter and 1-10
growing, are representative examples of wireless Mbps throughput; for Hi Fi stereo audio, jitter less than 1 ms
communications technologies. and 0.1-1 Mbps throughputs.
Wi-Fi is a microwave based technology providing for Wi-Fi security is very important. Microwave radio signals
versatility, mobility and favorable prices. The importance and can be very easily captured as they travel through the air.
utilization of Wi-Fi has been growing as it complements Therefore, several security methods have been developed to
traditional wired networks. It has been used both in ad hoc provide authentication such as, by increasing order of
mode and in infrastructure mode. In this case a WLAN arises security, WEP, WPA and WPA2. WEP was initially intended
to provide confidentiality comparable to that of a traditional
1
wired network. A shared key for data encryption is involved.
José Pacheco de Carvalho is with the Remote Detection Unit and The communicating devices use the same key to encrypt and
the Physics Department at the University of Beira Interior,
decrypt radio signals. The CRC32 checksum used in WEP
R. Marquês d’Ávila e Bolama, 6201-001 Covilhã, Portugal,
E-mail: pacheco@ubi.pt.
does not provide a great protection. However, in spite of its
2
Cláudia Pacheco is with the Remote Detection Unit at the weaknesses, WEP is still widely used in Wi-Fi
University of Beira Interior, R. Marquês d’Ávila e Bolama, 6201-001 communications for security reasons, mainly in point-to-point
Covilhã, Portugal, E-mail: a17597@ubi.pt. links. WPA implements the majority of the IEEE 802.11i
3
Hugo Veiga is with the Remote Detection Unit and the standard [1]. It includes a MIC, message integrity check,
Informatics Centre at the University of Beira Interior, R. Marquês replacing the CRC used in WEP. Either personal or enterprise
d’Ávila e Bolama, 6201-001 Covilhã, Portugal, E-mail: modes can be used. In this latter case an 802.1x server is
hveiga@ubi.pt. required. Both TKIP and AES cipher types are usable and a
4
António Reis is with the Remote Detection Unit and the Physics
group key update time interval is specified.
Department at the University of Beira Interior, and with the
Department of Electronics and Telecommunications/Institute of
Several performance measurements have been made for 2.4
Telecommunications, at the University of Aveiro, 3810 Aveiro, and 5 GHz Wi-Fi open [8-9] and WEP links [10], as well as
Portugal, E-mail: adreis@ubi.pt. very high speed FSO [11]. It is important to find the effects of
WPA encryption on link performance. Therefore, in the

63
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

present work new Wi-Fi (IEEE 802.11 b,g) results arise, using The results were obtained in batch mode and written as
personal mode WPA, through OSI levels 4 and 7. data files to the client PC disk. Each PC had a second network
Performance is evaluated in laboratory measurements of WPA adapter, to permit remote control from the official IP
point-to-point links using new available equipments. University network, via switch.
Comparisons are made to corresponding results obtained for
open links.
The rest of the paper is structured as follows: Chapter II
III. RESULTS AND DISCUSSION
presents the experimental details i.e. the measurement setup
and procedure. Results and discussion are presented in The access point and the PC wireless network adapter were
Chapter III. Conclusions are drawn in Chapter IV. manually configured, for each standard IEEE 802.11 b, g,
with typical fixed transfer rates (1, 2, 5, 11 Mbps for 802.11b;
6, 9, 12, 18, 24, 36, 48, 54 Mbps for 802.11g). For every fixed
II. EXPERIMENTAL DETAILS transfer rate, data were obtained for comparison of the
laboratory performance of the WPA and Open links at OSI
The measurements used a D-Link DAP-1522 bridge/access layers 1 (physical layer), 4 (transport layer) and 7 (application
point [12], with internal PIFA *2 antenna, IEEE 802.11 layer) using the setup of Fig. 1. For each standard and every
a/b/g/n, firmware version 1.31 and a 100-Base-TX/10-Base-T nominal fixed transfer rate, an average TCP throughput was
Allied Telesis AT-8000S/16 level 2 switch [13].The wireless determined from several experiments. This value was used as
mode was set to access point mode. The firmware from the the bandwidth parameter for every corresponding UDP test,
manufacturer did not make possible a point-to-point link with giving average jitter and average percentage datagram loss.
a similar equipment. Therefore, a PC was used having a At OSI level 1, noise levels (N, in dBm) and signal to noise
PCMCIA IEEE.802.11 a/b/g/n Linksys WPC600N wireless ratios (SNR, in dB) were monitored and typical values are
adapter with three internal antennas [14], to enable a PTP link shown in Fig. 2 and Fig. 3, for open and WPA links, and
to the access point. In every type of experiment, interference 802.11b, g, respectively.
free communication channels were used (ch 8 for 802.11b,g). The main average TCP and UDP results are summarized in
This was checked through a portable computer, equipped with Table I, both for WPA and open links. In Fig. 4 polynomial
a Wi-Fi 802.11 a/b/g/n adapter, running NetStumbler software fits were made to the 802.11b, g TCP throughput data for
[15]. WPA personal encryption was activated in the AP and WPA links, where R2 is the coefficient of determination. It
the PC wireless adapter using AES and a shared key with 26 was found that the best TCP throughputs are for 802.11 g, for
ASCII characters. The experiments were made under far-field every link type. The 802.11 b, g average data are reasonably
conditions. No power levels above 30 mW (15 dBm) were close for both link types. The best average 802.11g TCP
required, as the access points were close. throughput is for open links. In Figs. 5-7, the data points
A laboratory setup has been planned and implemented for representing jitter and percentage datagram loss were joined
the measurements, as shown in Fig. 1. At OSI level 4, by smoothed lines. Concerning jitter it was found that, on
measurements were made for TCP connections and UDP average, the best jitter performances are for 802.11 g for both
communications using Iperf software [16]. For a TCP link types. For each standard, jitter performances agree
connection, TCP throughput was obtained. For a UDP reasonably well within the experimental errors. However
communication with a given bandwidth parameter, UDP average jitter for 802.11 b is slightly higher for WPA (5.5+-
throughput, jitter and percentage loss of datagrams were 0.2 ms) than for Open links (5.3+-0.3 ms), meaning that in
determined. Parameterizations of TCP packets, UDP this case increasing security leads to a minor degradation of
datagrams and window size were as in [10]. One PC, with IP jitter performance. Fig. 7 shows percentage datagram loss
192.168.0.2 was the Iperf server and the other, with IP data. Except for 802.11 g, where the highest value is for
192.168.0.6, was the Iperf client. Jitter, representing the WPA, no significant sensitivities were found for most data
smooth mean of differences between consecutive transit (1.4 % on average), within the experimental errors, either to
times, was continuously computed by the server, as specified standard or link type.
by the real time protocol RTP, in RFC 1889 [17]. The scheme At OSI level 7 we measured FTP transfer rates versus
of Fig. 1 was also used for FTP measurements, where FTP nominal transfer rates configured in the access point and the
server and client applications were installed in the PCs with PC wireless network adapter for IEEE 802.11 b, g as in [10].
IPs 192.168.0.2 and 192.168.0.6, respectively. The server PC The results show the same trends found for TCP throughput.
also permitted manual control of the settings in the access Generally, except for 802.11g TCP throughput, 802.11 b
point. jitter and 802.11g percentage datagram loss, the results
The server and client PCs were HP nx9030 and nx9010 measured for WPA links were found to agree, within the
portable computers, respectively, running Windows XP. They experimental errors, with corresponding data obtained for
were configured to optimize the resources allocated to the Open links.
present work. Batch command files have been written to
enable the TCP, UDP and FTP tests.

64
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

TABLE I
Average Wi-Fi (IEEE 802.11 b,g) results; WPA
and Open links.
Link
WPA Open
type
Parameter/
IEEE 802.11b 802.11g 802.11b 802.11g
standard
TCP
2.9 13.4 3.0 14.5
throughput
+-0.1 +-0.4 +-0.1 +-0.4
(Mbps)
UDP-jitter 5.5 2.3 5.3 2.3
(ms) +-0.2 +-0.1 +-0.3 +-0.1
UDP-%
1.2 1.8 1.2 1.2
datagram
+-0.2 +-0.2 +-0.2 +-0.1
loss
FTP
280.2 1450.6 289.9 1526.9
transfer rate
+-11.2 +-58.0 +-11.6 +-61.1
(kbyte/s)
Fig. 1- Laboratory setup scheme.

Fig. 2- Typical SNR (dB) and N (dBm); 802.11b;WPA and open Fig. 4- TCP throughput versus technology and nominal transfer rate;
links. WPA links.

Fig. 3- Typical SNR (dB) and N (dBm); 802.11g; WPA and open Fig. 5- UDP - jitter results versus technology and nominal transfer
links. rate; WPA links.

65
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

ACKNOWLEDGEMENT
Supports from Universidade da Beira Interior and FCT
(Fundação para a Ciência e a Tecnologia)/PEst-
OE/FIS/UI0524/2011 (Projecto Estratégico-UI524-2011-
2012) are acknowledged.

REFERENCES
[1] Web site http://standards.ieee.org Web site; IEEE 802.11a,
802.11b, 802.11g, 802.11n, 802.11i standards.
[2] J. W. Mark, W. Zhuang, Wireless Communications and
Networking, Prentice-Hall, Inc., Upper Saddle River, NJ, 2003.
[3] T. S. Rappaport, Wireless Communications Principles and
Practice, 2nd ed., Prentice-Hall, Inc., Upper Saddle River, NJ,
2002.
Fig. 6- UDP - jitter results versus technology and nominal transfer [4] W. R. Bruce III, R. Gilster, Wireless LANs End to End, Hungry
rate; open links. Minds, Inc., NY, 2002.
[5] M. Schwartz, Mobile Wireless Communications, Cambridge
University Press, 2005.
[6] N. Sarkar, K. Sowerby, “High Performance Measurements in
the Crowded Office Environment: a Case Study”, In Proc.
ICCT’06-International Conference on Communication
Technology, pp. 1-4, Guilin, China, 27-30 November 2006.
[7] E. Monteiro, F. Boavida, Engineering of Informatics Networks,
4th ed., FCA-Editor of Informatics Ld., Lisbon, 2002.
[8] J. A. R. Pacheco de Carvalho, P. A. J. Gomes, H. Veiga, A. D.
Reis, ”Development of a University Networking Project”, in
Encyclopedia of Networked and Virtual Organizations, Goran
D. Putnik, Maria Manuela Cunha, Eds. Hershey, PA
(Pennsylvania): IGI Global, pp. 409-422, 2008.
[9] J. A. R. Pacheco de Carvalho, H. Veiga, P. A. J. Gomes, C. F.
Ribeiro Pacheco, N. Marques, A. D. Reis, “Wi-Fi Point-to-
Point Links- Performance Aspects of IEEE 802.11 a,b,g
Laboratory Links”, in Electronic Engineering and Computing
Technology, Series: Lecture Notes in Electrical Engineering,
Fig. 7- UDP – percentage datagram loss versus technology and Sio-Iong Ao, Len Gelman, Eds. Netherlands: Springer, 2010,
nominal transfer rate; WPA links. Vol. 60, pp. 507-514.
[10] J. A. R. Pacheco de Carvalho, H. Veiga, N. Marques, C. F.
Ribeiro Pacheco, A. D. Reis, Wi-Fi WEP Point-to-Point Links-
IV. CONCLUSION Performance Studies of IEEE 802.11 a,b,g Laboratory Links, in
Electronic Engineering and Computing Technology, Series:
A new laboratory setup arrangement has been planned and Lecture Notes in Electrical Engineering, Sio-Iong Ao, Len
implemented, that permitted systematic performance Gelman, Eds. Netherlands: Springer, 2011, Vol. 90, pp. 105-
measurements of new available wireless equipments (DAP- 114.
[11] J. A. R. Pacheco de Carvalho, N. Marques, H. Veiga, C. F.
1522 access points from D-Link and WPC600N adapters from
Ribeiro Pacheco, A. D. Reis, ”Experimental Performance
Linksys) for Wi-Fi (IEEE 802.11 b,g) in WPA point-to-point Evaluation of a Gbps FSO Link: a Case Study”, Proc. WINSYS
links. 2010- International Conference on Wireless Information
Through OSI layer 4, TCP throughput, jitter and percentage Networks and Systems, pp. 123-128, Athens, Greece, 26-28 July
datagram loss were measured and compared for WPA and , 2010.
open links. Generally, except for 802.11g TCP throughput, [12] Web site http://www.dlink.com; DAP-1522 wireless
802.11 b jitter and 802.11g percentage datagram loss, where bridge/access point technical manual.
increasing security encryption was found to degrade [13] Web site http://www.alliedtelesis.com; AT-8000S/16 level 2
performances, the results measured for WPA links were found switch technical data.
[14] Web site http://www.linksys.com; WPC600N notebook adapter
to agree, within the experimental errors, with corresponding
user guide.
data obtained for Open links. [15] Web site http://www.netstumbler.com; NetStumbler software.
At OSI layer 7, FTP performance results have shown the [16] Web site http://dast.nlanr.net; Iperf software.
same trends found for TCP throughput. [17] Network Working Group. “RFC 1889-RTP: A Transport
Additional performance measurements either started or are Protocol for Real Time Applications”, http://www.rfc-
planned using several equipments, security settings and archive.org
experimental conditions, not only in laboratory but also in
outdoor environments involving, mainly, medium range links.

66
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Customer Satisfaction based Demand Analysis of Mobile

Services
Aleksandar Tsenov1
Abstract – The rapid technology growth of the mobile customer satisfaction with a defined number of services. The
networks has led to the situation where the competition for each experiment included a wide study over three groups of mobile
one customer became a great importance. The customer services users, conducted at the Technical University of Sofia.
satisfaction is increasingly attended to the quality, to the The three target groups were the university teachers, the
performance and to the usefulness of the services. These are in
nature different characteristics and their complex evaluation
administration and students. These groups are being chosen
requires implementation of new methods and tools such as fuzzy because of the differences in age, in job, in activity of service
logic and fuzzy evaluation schemes. The paper proposes an use etc.
approach for analyzing the customer demand on mobile services
according their satisfaction with these services. The demand
analysis is object of fuzzy evaluation approach based on the II. CUSTOMER VALUE HIERARCHY AND
customer value hierarchy. ATTRIBUTES DEFINITION

Keywords – Customer Satisfaction, Fuzzy Evaluation, Mobile

Services, Significant Attribute Definition. A. The mobile customer value hierarchy

Based on the complete chain of customer value layers, the

I. INTRODUCTION first step in the procedure is to shift the layers from the
individual perspective to the aggregate perspective of a group
There are many known attempts for creating a methods and of customers. Based on the mobile customer investigation, an
models for meeting the improved customer requirements model for constructing the mobile customer value hierarchy is
according the service quality. All of them requires introduced [1].
considerable preparation before the evaluation schemes are
implemented. The first step of the whole evaluation process
B. Attributes definition
is the definition of an appropriate customer experience model
in order to find out the customer’s goal and purpose, the
In order to apply the fuzzy evaluation of the customer
desired consequences in use situation and the desired
satisfaction the following investigation where performed:
products/ services attributes and performances [1]. In the same
Three different groups of mobile users where asked about the
work an exemplary “mobile customer value hierarchy” was
services mentioned above – teachers, administration staff and
defined. In [2] the author introduces an overall demonstration
students. Each group covers 100 people. The questionnaire
of the applicability of fuzzy evaluation method for Service
includes four questions according each of the services:
Level Management metrics.
There are not many similar researches in the field of the
customer centric evaluation. In [3] the authors use the fuzzy 1. Do You use the service …….?
similarity approach for clustering the QoS opinions for Web 2. How could You evaluate the service?
services. In [4] a fuzzy oriented approach for clustering of the 3. Could You evaluate service parameters?
services attributes and for definition of the most significant of
these attributes is introduced. The research is based on the The answers on the questions 1 and 3 are “Yes” and “No”.
customer demand in personalized services. In [5] the authors The answers on the question 2 include different evaluation
apply fuzzy evaluation of SLA (Service Level Agreement) levels. The evaluation is defined with 5 evaluation levels –
oriented quality metrics in NGN. All mentioned research from 1 to 5. Table I shows the results of the answers on the
works are based on hypothetical values, and not on real data first question and the corresponding coding for each service.
and are only used for proving the applicability of the fuzzy The values give the number of the group members which are
logic in such complex evaluation problems. engaged with the corresponding service.
This work is an extension of [6]. In the work mentioned, the In [6] these definitions were used for calculating the
authors attempt to apply the fuzzy evaluation approach to the customer satisfaction score - S. For the three groups being
customer experience hierarchy in order to evaluate the studied it was obtained:
Satisfaction Score S (Teachers) = 3.0333;
Satisfaction Score S (Administration) = 2,994157;
1
Aleksandar Tsenov is with the Faculty of Telecommunications at Satisfaction Score S (Students) = 3,005901.
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, That means – the group “Teachers” has the higher
Bulgaria, E-mail: akz@tu-sofia.bg. satisfaction score with the services have being studied.

67
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA
TABLE I TABLE II
EVALUATION METRICS SERVICE USAGE SERVICE QUALITY PARAMETERS

No. Attributes Votes Code No. Objectives Attributes Code

1 Voice mail box Teachers – 57 S1 1 Voice mail Message duration S11
Administration – 76 box (S1) Number of messages to be stored S12
Students - 87 Number of rings before start
2 SMS Teachers – 78 S2 VoiceMailBox S13
Administration – 89 Broken messages S14
Students - 92 Speech quality S15
3 Voice call Teachers – 100 S3 2 SMS (S2) Message length S21
Administration – 100 Message validity S22
Students - 100 Number of messages to be stored S23
4 Conference call Teachers – 14 S4 Multi-user send message S24
Administration – 2 Service coverage S25
Students - 34 3 Voice call Service coverage S31
5 Routine Services Teachers – 82 S5 (S3) Speech quality S32
Administration – 91 Second voice call during
Students - 89 conversation S33
6 E-Bank Teachers – 8 S6 4 Conference Service coverage S41
Administration – 11 call (S4) Speech quality S42
Students - 13 Number of participants S43
7 Data Service Teachers – 23 S7 Service control S44
Administration – 8 5 Data Service Down speed S71
Students - 47 (S7) Up speed S72
8 Mobile Purchase Teachers – 26 S8 Lost data S73
Administration – 18 Service coverage S74
Students - 55
9 Internet browsing Teachers – 42 S9 Table 2 is produced with applying the proposed in this
Administration – 32 work Attribute – objective map. This means: uncover the
Students - 70 relationship between quality parameters that the customers
10 Mobile movie Teachers – 4 S10
engaged (attributes layer) and the used services (objective
Administration – 28
Students - 65
layer). That approach enables the mobile provider to identify
11 Mobile games Teachers – 4 S11 the customer’s goal from their consume history. So we reduce
Administration – 12 the layers in the customer value hierarchy to an attribute –
Students - 75 objective map. We discovered attributes that are equal to more
12 MMS Teachers – 34 S12 than one objective – for example: Speech quality.
Administration – 52 So we obtained 21 service quality parameters. The main
Students - 78 goal of the following analysis is to find out the most important
13 Travel Info Teachers – 5 S13 parameters that are significant not only for the corresponding
Administration – 11 service but for the overall quality grade of the delivered
Students - 23 services too.
14 Entertainment Teachers – 6 S14
Info Administration – 14
Students - 19 C. Significant Attributes Analysis
15 Finance Info Teachers – 9 S15
Administration – 32 The significant attributes of customer value hierarchy are
Students - 11 the key attribute variables of the attribute layer which
distinctly correlate to the objective layer. Because of the large
In this work we will follow other way – further numbers of mobile telecommunication products/services and
decomposition of the attributes defined above into most the relatively small percentage of the mobile services/products
important quality parameters of each one service, represented engagement, the original data of customer value hierarchy is
as attribute in the value hierarchy. The number of parameters high dimensional sparse feature data. This paper adopts the
for each service is limited on 5, but it can be higher or fuzzy cluster analysis method [7] to find the significant
respectively lower. attribute.
Table II represents 5, randomly chosen services and their According to the rough set theory, data of the customer
quality characteristics. We choose only 5 services in order to value objective layer and attribute layer can be defined as S=
make the work more understandable and clear. We suppose (U, A, V, f). Here: U= {u1, u2,…, un}: the set of customers
that when the approach is proven with smaller number of where n is the total number of customers. A={a1, a2,…, am}:
services, it will become applicable for great amount of the set of variables of the objective layer and of the attribute
services too. Each characteristic is presented with unique layer. A = C ∪ D , where C is the characteristics set of the
code. attribute layer, and D is the characteristics set of the objective

68
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

layer. V is the set of the customer attribute parameters. The TABLE III
value of f (uj, ai) indicates the value of uj about ai. INVESTIGATION RESULTS – GROUPS “TEACHERS”,
The significant attributes analysis is solved by fuzzy “ADMINISTRATION”, “STUDENTS”
clustering [8]. The process of the analysis includes the
Teachers Administration Students
following steps: N Objecti Attribu Objecti Attribut Object Attribute
Step1. Calculate the similarity matrix for the attributes. The o. ves tes ves es ives s
pair-wise comparison method is used to obtain the values of 1 S1 - 57 S11 – 31 S1 - 76 S11 – 44 S1 - 87 S11 – 54
the corresponding element aij, where (i=1,2, …,k and j=i+1, S12 – 11 S12 – 21 S12 – 43
i+2,…,i+(k-1)). The values of aji are obtained as 1-aij. Here k S13 – 5 S13 – 31 S13 – 56
S14 – 24 S14 – 25 S14 – 44
is the number of the attributes for the corresponding objective. S15 – 55 S15 – 48 S15 – 48
Step2. Calculate the fuzzy similarity matrix R. As shown in 2 S2 - 78 S21 – 77 S2 - 89 S21 – 69 S2 - 92 S21 – 88
equation (1) the research adopts the cosine distance measure S22 – 65 S22 – 75 S22 – 74
as the method of similarity measurement of the study objects. S23 – 71 S23 – 77 S23 – 70
m 2 m m
S24 – 45 S24 – 58 S24 – 66
(1) S25 – 76 S25 – 82 S25 – 87
r =
ij ∑
k =1
( a aik ) /
jk a 2∑
i=1
a 2
ik ∑
j=1
jk
3 S3 – 100 S31 – 99 S3 – 100 S31 – 90 S3 – 100 S31 – 100
S32 – 93 S32 – 75 S32 – 98
During the study the calculation of the fuzzy similarity S33 – 45 S33 – 80 S33 – 100
matrix R using Euclidian Distance measure was performed as 4 S4 - 14 S41 – 7 S4 - 2 S41 – 1 S4 - 34 S41 – 13
S42 – 11 S42 – 2 S42 – 12
well. The results obtained where almost the same and will be S43 – 8 S43 – 0 S43 – 14
not shown here. S44 – 6 S44 – 1 S44 – 12
Step3. Calculate the fuzzy transitive closure t(R) of the 5 S7 - 23 S71 – 15 S7 - 8 S71 – 5 S7 - 47 S71 – 31
fuzzy similarity matrix R with the square method [9]. If the S72 – 16 S72 – 8 S72 – 30
S73 – 21 S73 – 4 S73 – 18
fuzzy similarity matrix can be expressed as R = (rij)nxn, then S74 – 19 S74 – 3 S74 – 42
R o R = (tij)nxn max 
min  ,   (2)
weight of the attribute values:
δ −δ
   e j i
If  o  =  − δi ), (3)
−1
, then the fuzzy transitive closure a ij = δ −δ
= log it (δ j
1 + e j i

[t(R)] =  .
where δk is the scale location of object k and logit-1 is the
Use the cluster method to analyze t(R) with intercept λ and
inverse logit function.
determine the significant attributes set.
Here the calculations for the group “Teachers” are shown:
So the fuzzy set A is obtained:
III. DATA ANALYSIS 1 0.56218 0.57444 0.54157 0.37754
0.43782 1 0.51250 0.47918 0.32082)
 (
A = 0.42556 0.48750 1 0.46672 0.31003( (4)
The investigation gave 300 questionnaires out to the 0.45843 0.52082 0.53328 1 0.33924(
individual mobile customers. The questionnaire enumerates 0.62246 0.67918 0.68997 0.66076 1 '
the quality parameters of the attribute layer corresponding to a
given service of the objective layer. For each service on of the Then, on Step 2, the fuzzy similarity matrix R is calculated:
1 0.95463 0.94595 0.95094 0.92991
objective layer the number of the customers that use the 0.95463 1 0.98009 0.96527 0.92442)
service is given and on the attribute layer - the number of  (
users that have evaluate the corresponding attribute. The R = 0.94595 0.98009 1 0.95725 0.90856( (5)
results obtained by the investigation of all three groups are 0.95094 0.96527 0.95725 1 0.92817(
0.92991 0.92442 0.90856 0.92817 1 '
shown in the following Table III. The number of answers for
each attribute gives the relative importance of the After that the transitive closure [t(R)] for the set of
corresponding attribute in the group. This relative importance parameters is calculated according equation (2). In all cases,
is used as a weight of the attribute for the calculation of the being studied, the intercept λ is chosen from the values
fuzzy set values. obtained for [t(R)].
The calculation procedure is as follows: 1 0.95463 0.95463 0.95463 0.92991
 )
0.95463 1 0.98009 0.96527 0.92991
Step 1. For each one group, after partitioning A into C and  (
D the membership degree of each one attribute is calculated. [t(R)] = 0.95463 0.98009 1 0.96527 0.92991( (6)
For example: For the service 1 (S1) in group “Teachers” and 0.95463 0.96527 0.96527 1 0.92991(
the corresponding service attributes the following set A is 0.92991 0.92991 0.92991 0.92991 1 '
obtained: There are 57 (N = 57) positive answers on the above Here we can have the following values for λ. For each λ we
question 1. This is equal to 0,19 (K = N/Nall) of all can define the corresponding clusters of parameters:
participants in this study. Then the number of the positive λ=1 {S11},{S12},{S13},{S14},{S15}
answers on question 3, related to K is calculated. So the λ=0.98009 {S12, S13},{S11},{S14},{S15}
weight of each one attribute value is defined. λ=0.96527 {S12, S13, S14},{S11},{S15}
The elements in set A are calculated according the pair- λ=0.95463 {S11, S12, S13, S14},{S15}
wise[10] comparison of the attribute value with respect to the λ=0.92991 {S11, S12, S13, S14, S15}

69
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Then we build the dynamic cluster diagram, from which the the most significant parameters for all groups and services
most significant parameters can be obtained. As shown on the will be shown (Table IV).
figure below, for the group “Teachers”, the most significant
parameters of the service S1 – Voice mail box are the
parameters coded as S12 and S13. All others parameters are
IV. CONCLUSION
concatenated one after another to the base cluster, built from
these two parameters. In this paper a fuzzy evaluation approach is introduced,
This dynamic cluster diagram can be implemented later as a used for definition of significant service parameters, that
model of deductive database for easier search of significant defines the customer satisfaction with the mobile services.
attributes also in cases of greater amount of parameters or in The study has to be continued in order to evaluate the already
case of deeper investigation of the significant service defined significant attributes from the mobile operator point
parameters. An appropriate method for cost effective search in of view.
such database structures is the 2P-Method introduced in [11]. The correspondence of the evaluation results will be a good
starting point for development of appropriate approaches,
methods and tools for improving the grade and the
effectiveness of the Customer Relationship Management and
the Customer Experience Management.

REFERENCES

Fig. 1. Dynamic cluster diagram [1] Al. Tsenov, “Customer Experience Hierarchy Model”,
Proceedings of the XLV-th International Scientific Conference
Following the same steps the following calculation and ICEST 2010, Ohrid, Macedonia, June 24 – 26, vol.1, pp. 77-80,
results are achieved for all other services in correspondence to 2010.
the services being studied. [2] Al. Tsenov, G. Yoncheva, E. Stoyanova and Al. Pavlov, “Fuzzy
Evaluation of Service Level Management Metrics”, XLVI-th
TABLE IV International Scientific Conference ICEST 2011, Nish, Serbia,
SIGNIFICANT PARAMETERS June 29 – July 01, presented paper, 2011
Servi Group λ Parameter clusters Significant [3] W. Lin, C. Lo, K. Chao, and N. Godwin, “Fuzzy Similarity
ce Parameters Clustering for Consumer-Centric QoS-aware Selection of Web
{S11},{S12,S13}, S12, S13 Services”, in Proc. CISIS 2009, pp. 904-909, 2009
S1 Teachers
0.91577 {S14},{S15} [4] S. Yajing et al., “Customer Value Hierarchy Based Customer
Administra {S11},{S12,S14}, S12, S14 Demand Analysis In Personalized Service Recommender
tion 0.96807 {S13},{S15} System”, International Journal of Simulation, Vol. 7 No 7, ISSN
Students {S11},{S12,S14}, S12, S14 1473-0831 print, pp. 77-84, 2007
0.95929 {S13},{S15} [5] Y. Danfeng, Y. Fangchun, “Fuzzy evaluation of SLA-oriented
Final S12, S14 QoSM (The Quality of Service Management) in NGN”,
S2 Teachers {S21},{S22,S24}, S22, S24 Proceedings of IC-BNMT, ISBN 978-1-4244-4591-2, pp. 301-
0.96113 {S23},{S25} 305, IEEE, 2009
Administra {S21,S24},{S22}, S21, S24 [6] Tsenov Al., I. Ivanov, T. Poparova, S. Neykov, L. Ivanova, M.
tion 0.96956 {S23},{S25} Gadjeva, “Fuzzy evaluation of customer satisfaction with
Students {S21},{S22},{S23,S24}, S23, S24 mobile services”, 10th International Conference TELSIKS
0.95954 {S25} 2011, vol. 2, pp. 665 – 668, October 5-8, Nish, Serbia,
Final S24 [7] H. Jiang, “Fuzzy Evaluation on ERP System Implementing Risk
S3 Teachers 0.93725 {S31, S33},{S32} S31, S33 Based on Membership Degree Transformation New Algorithm”,
Administra {S31}, {S32, S33} S32, S33 Second International Symposium on Electronic Commerce and
tion 0.85246 Security, Nanchang, China, May 22-May 24, pp 409-416, 2009
Students 0.86190 {S31, S33}, {S32} S31, S33 [8] Jordanova and V. Mladenov, “Fuzzy sets and computational
Final S31, S33 intelligence”, Lecture notes, Sofia University, 2005
S4 Teachers 0.95469 {S41, S44}, {S42}, {S43} S41, S44 [9] Sun Peide, “ Novel Fuzzy Clustering Algorithm to Predict Gas
Administra {S41, S43, S44}, {S42} S41, S43, Outburst Zone”. Proceedings of the 2006 International
tion 0.98709 S44 Symposium on Safety Science and Technology, October 24-27,
Students 0.94694 {S42, S44}, {S41}, {S43} S42, S44 2006 Changsha, Hu’nan, China
Final S41, S44 [10] David, H.A. (1988). The Method of Paired Comparisons. New
S7 Teachers 0.95217 {S71, S72},{S73},{S74} S71, S72 York: Oxford University Press.B. Milovanovic, Z. Stankovic, S.
Administra {S71},{S72},{ S73, S74} S73, S74 Ivkovic and V. Stankovic, "Loaded Cylindrical Metallic
tion Cavities Modeling using Neural Networks", TELSIKS'99,
0.95968
Conference Proceedings, pp.214-217, Nis, Yugoslavia, 1999.
Students 0.86853 {S72},{S71, S73},{ S74} S71, S73
[11] Iltchev V., Cost Evaluation of Methods for Query Processing in
Final S71, S73
Deductive Database Systems, Information Technologies and
Here, from all other calculations, only the highest grade of Control, Vol. 5, No 1, 2007, pp. 10-20, ISSN:1312-2622,
λ, the corresponding parameter clustering and the final result– Bulgarian Union of Automation and Informatics, Sofia,
Bulgaria,

70
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Investigate common work of software phone systems in

virtual environments and real switching systems
Borislav Rozenov Necov1 Krasen Zhivkov Bankov2 Mario Petrov Georgiev3

Abstract - This paper discusses the common work of In practice, only 10% of servers capacity in a continuous
Linux - based IP phone systems Trixbox and Elastix, build mode is used. A large data center could save electricity
in virtual environments. Their interconnection with real equivalent to the consumption of 200,000 households under
switching systems is applied. Advantages of using loading up to 50% of its capacity. A serious problem for the
virtualization software in communications are shown using ecological balance is large carbon dioxide emissions.
VoIP software analyzators. Experiments are made, According to calculations, covering 50.000 data centers,
revealing the advantages of using VoIP telephony and emissions will exceed 10 million tons by 2013. These figures
virtualization, leading to much easier maintenance and leave no doubt that information technology bear their share of
more flexible services. responsibility for environmental protection [4].
Optimized use and better distribution of work is achieved By
Keywords – Voice-over-IP, Virtualization, Phone systems application virtualization, which in turn leads to fewer servers
and to shorten the prolonged periods in which they operate
without load. Integrated approach in the management of
I. INTRODUCTION information technology is beneficial not only for climate but
also for budget of the companies. It enables them to reduce
Transmission of data and voice via VoIP is extremely hardware investments and make large cost savings for
attractive to business users, service operators and for home electricity [5].
users because it allows Internet and data networks, already
established in offices[1], enterprises and administrative areas, Two experiments are made to examine the virtualization.
to transmit voice calls, video conferences to support and other Telephone software systems Trixbox and Elastix are used as
real-time applications. well as softphones and VoIP analizators.
VoIP telephone applications such as Asterisk-based
Trixbox can be uploaded to a virtual environment, which II. EXPERIMENTAL RESULTS
means reducing the number of physical machines and leads to
less power consumption, simplifying IT infrastructure, a much Virtual TRIXBOX and ELASTIX are loaded in the virtual
easier maintenance and much greater service flexibility [2]. machine.
Installation of multiple virtual environments (operating
systems and applications) in one physical server
(homogeneous hardware) is more economically reasonable Different software phone systems TRIXBOX and
than provision and maintenance of each physical server for ELASTIX are used when conducting the experiments. Some
each application [3]. of them are in virtual environment, while the rest are real
Figure 1.1. shows the qualitative leap made in the period 2007 working (not in virtual environment). For the purposes of the
to 2009, when the total number of installed physical servers study some other phone systems can be used such as 3CX and
stops to increase and even decreases at the expense of the different hardware switching systems. Test computers have
rapid growth of virtual machines. In the U.S. over the next 2-3 1GB RAM operating memory and processor Intel Pentium 4.
years the number of installed physical servers will be reduced Emulating virtual programs can be VMWare products
by 20%. (Workstation, VMWare Player), Microsoft products, Oracle
(Virtual Box) etc.

2.1. Virtual Trixbox - Virtual Elastix

At the start of the call as well as at its termination, SIP
signaling protocol is used.

Fig.1.1. Increasing use of virtual servers

71
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig.2.4. Quality of Service audio

2.2 Virtual Trixbox - real Trixbox

In this experiment, TRIXBOX is loaded in the virtual machine
Fig.2.1. Signalling on call
and real built PBX server TRIXBOX is used.
The graph in Fig. 2.5 shows the number of RTP packets per
Figure 2.2. shows the number of packets (axis Y) per unit time
unit time. Again some hesitation is noticed, because the length
for 120 seconds. The number of RTP packets vary over time
of each packet is not the same and therefore requires much
because the length of each packet is not the same and for one
time to transmit if more bytes are used.
unit of time different number of packets can be transmitted.

Fig.2.2. Distribution of RTP packets per unit time - IO Graphs Fig.2.5. Distribution of RTP packets per unit time

In fig. 2.3. with black the values of jitter are shown, the The values of jitter (black) and the delay between packets (in
average is about 3ms. In red delays between packets are red) have improved by about 2ms (Fig. 2.6) compared to the
marked. first investigation.

Fig.2.3. Jitter and delay of packets in graphical form Fig.2.6. Jitter and delay of packets in graphical form

Fig. 2.4 shows service quality of the audio signal (Audio Below is shown the total activity of the network - network
QoS). Reports for 30 seconds back have been noted. R factor activity, VoIP activity, conducted conversations, provided
(R Factor) is within the allowable (ranging from 0 to 120). packet activity, protocol activity. Everything is within normal
Packet loss in unusual circumstances (Burst Packet Loss limits.
Rate%) is 0.102% at source and destination at 0.055%.
Rejected packs (Discarded Packets) are respectively 2.105 and
26.

72
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

III. CONCLUSION

The advantages of virtualization are undisputed. It goes faster,

both in business and in everyday life. In common work with
communication systems and applications it has abilities that
will undoubtedly make its use more widely.
Fig.2.7. Network Monitor

The main parameters that are subject to change in a better

direction are shown in fig.2.8 and fig.2.9.
Source R Factor is improved by 10 units and MOS score by
0.5 as the result become almost 3.1. Rejected packets and the
loss possibility are less than previous experiment. The average REFERENCES
jitter value is significantly reduced in the destination as it
becomes 0,415 ms, and in sourse - 3,5 ms. The average delay [1]Свилен Иванов, „Изграждане на виртуални машини с
between packets is again about 20ms. Xen‖,
http://0101.nccdn.net/1_5/2e6/367/236/Xen.pdf
[2]Microsoft, ―Desktop Virtualization Strategy‖, White Paper
[3] Kerry Garrison, ―trixbox CE 2.6: Implementing,
managing, and
maintaining an Asterisk-based telephony system‖, Packt
Publishing, 2009
[4] Наньо Нанев, „Виртуализацията - що е то?‖,
http://itanalyses.blogspot.com/2009/01/blog-post.html
[5] „Виртуализацията - на прага на версия 3.0‖,
Fig.2.8. Quality of Service audio
http://cio.bg/3653_virtualizaciyata__na_praga_na_versiya_30

Fig.2.9. Audio Details

Contacts:
2
Krasen Zhivkov Bankov, student in . KТТ in FE of TU-Varna
„ Studentska” № 1 str, e-mail: banki4@abv.bg
1
Borislav Rozenov Necov, student in . KТТ in FE of TU-Varna,
„ Studentska” № 1 str, e-mail: sharkiller@mail.bg
3
Mario Petrov Georgiev, student in . KТТ in FE of TU-Varna,
„ Studentska” № 1 str, e-mail: markata88@gmail.com

73
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Analysis of current methods and technologies for

encoding, distribution and consumption of IPTV services
Jordan Kanev1 and Stanimir Sadinov2
Abstract – Internet Protocol television (IPTV) is a system picture expert group MPEG. In addition to this H.264/MPEG-
through which television services are delivered using the Internet 4 is used for HDTV, while older standard MPEG-2 is used for
protocol suite over a packet-switched network, instead of being SDTV. Another solution is presented by Society of Motion
delivered through traditional terrestrial, satellite signal, and Picture and Television Engineers (SMPTE) and implemented
cable television formats.
by Windows Media 9 (WM-9) is named VC-1. It has similar
Keywords – IPTV, VoD, MPEG4 AVC, H.264 features to MPEG-4, but has better integration with PCs.

I. INTRODUCTION
A new method of delivering and viewing television
channels over IP network and high speed broadband access
technology is called IPTV, short for internet protocol
television and also known as broadband TV. It is not only a
distribution method, but also bring new interactive features
and changes the way we watch TV. Ability to pause, resume
and fast forward TV shows gives freedom to every user to
make individual custom program schedule is only a small part
of innovations presented by interactive television. Internet
protocol television differs from Internet television as Internet Fig.1. IPTV architecture
TV is streaming video content over public Internet while Some of the most popular video compression standards are
IPTV is streaming dedicated video content via private shown in Table I below:
managed network with quality of service (QoS). If service
provider is delivering three services: Broadband internet, TABLE I
voice over IP (VoIP) and IPTV the technology is called Triple VIDEO COMPRESSION STANDARDS
play and if wireless mobility is added then it becomes Quad
play. World largest IT companies like Microsoft, Cisco and Published Standard Organization
Google are involved in developing end-to-end IPTV solutions: 1993 H.261/MPEG-1 ISO, IEC
Microsoft TV, Google TV and Cisco Content Delivery 1995 H.262/MPEG-2 ISO, IEC, ITU-T
System. 1999 MPEG-4 ISO, IEC
2003 H.264/MPEG-4 AVC ISO, IEC, ITU-T
II. IPTV ARCHITECTURE 2006 VC-1/WM9V ISO, Microsoft
2008 VC-2 Dirac ISO, BBC
End-to-end IPTV infrastructure consist of three major
components: originate (head-end), distribute (delivery Originally MPEG-1 is used for Video-CD (VCD) 120 mm
network) and consumption (user-end). An overview of typical optical disc and became the first format for distributing films.
IPTV architecture is presented on fig.1 MPEG-2 is internationally accepted standard for digital
television and is widely used for digital video broadcast
As IPTV is transmitted over IP it needs proper compression (DVB) systems.
techniques to compress the video prior to its transmission MPEG-4 is used for Internet and mobile video, standard-
depending on the available bandwidth. Researchers from ITU- definition (SDTV) and high-definition television (HDTV).
T have found MPEG compression the best possible solution MPEG-4 uses up to 50 % less bandwidth than MPEG-2 with
for this challenge. ITU-T has standardized H.264 and that is bit rates from 5 Kbit/s to 10 Mbit/s depending on quality and
equivalent to MPEG-4 (part 10) standardized by Moving video screen resolution.
To deliver video content over IP network IPTV system
1
Jordan Kanev is with the Faculty of Electrical Engineering and needs interactive services. IPTV services can be divided in
Electronics, Technical University of Gabrovo, 4 H. Dimitar St., 5300 three main groups: streaming TV, video on demand (VOD)
Gabrovo, Bulgaria, E-mail: danchokanev@gmail.com. and time-shifted TV. According to ETSI specification [6]
2
Stanimir Sadinov is with the Faculty of Electrical Engineering NGN integrated IPTV include following IPTV services:
and Electronics, Technical University of Gabrovo, 4 H. Dimitar St., Broadcast TV; Content on Demand (nCoD, pCoD); Personal
5300 Gabrovo, Bulgaria, E-mail: murry@tugab.bg.

74
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Video Recording (cPVR, nPVR); Pay Per View (PPV); VoD content can be encrypted, scrambled and embedded
Interactive TV (iTV); User Generated Content (UGC); with a tag to avoid content piracy enforcing Digital rights
Profiling and personalization; Content Recommendations management (DRM).
(CR); Advertising (Ad) and Targeted Advertising (TAI); Then encoded video content is sent in an MPEG transport
Messaging services; Notification services; Personalized stream and depending on service it can be forwarded to
channel; Bookmarks or Content Marking (CM). service provider delivery network in case of live TV or stored
IPTV services provide users with more control over viewed on VoD server in case of video on demand service.
TV channels and give them ability to make decision what TV
content to choose and when to watch it. User can watch IPTV
with a number of network-addressable devices like laptop,
IV. DISTRIBUTION VIA IP NETWORKS
personal computer, set-top box and TV, smartphone, tablet,
gaming consoles, etc. Video originated at Head-end must be preserved and
transported across IP network to reach end users. Broadband
networks used for Internet (data) and voice services are ideal
III. OBTAINING VIDEO AT HEAD-END for adoption of IPTV. Physical layer (Layer 1) can be copper,
fiber optic, combination of both types and even broadband
Service providers need to acquire and encode video content. wireless (IEEE 802.11n). Technologies used for data link
They can use several different video sources and analog, layer (Layer2) are MetroEthernet, xDSL, FTTx, ATM,
digital and IP technologies to do that. Streaming video can be WiMAX, LTE, DVB-H, etc. Depending on cable types and
obtained from following sources: physical topology IP networks include intermediate devices
• Satellite - DVB-S (SDTV) / DVB-S2 (HDTV) like routers and switches to forward user data across the
network. Additional mechanisms and protocols are used to
• Cable – DVB-C / DVB-C2 ensure reliable delivery of IPTV traffic. IPTV traffic is
• Terrestrial – DVB-T / DVB-T2 sensitive to losses and delay. Quality of Service (QoS) is
• Analog TV - obsolete responsible for prioritizing video over other traffic on the
network. QoS Standards include IEEE 802.1p QoS, IEEE
TV signal acquisition needs different equipment like 802.1q VLAN, CEA2007 VLAN Mapping.
satellite dishes, TV antennas, cables modems, DVB receivers, Additional control software (or middleware) is required to:
Set-top boxes (STB) depending on source type and gain user access privileges, manage video content, to protect
technologies of distribution systems. There are also a variety intellectual property via Digital Rights Management (DRM),
of modulation methods and schemes, some of them are shown billing software and offer services like electronic
in Table II. programming guide (EGP), VoD catalog, web services,
TABLE II mobile applications, domain name system (DNS), Network
MODULATION SCHEMES Time Protocol (NTP), Dynamic Host Configuration Protocol
(DHCP), nCoD, nPVR, advertising, etc.
IPTV is delivered via IP multicast in case of streaming TV
Standard Modulation schemes or via IP unicast in case of video on demand. More complex
DVB-S QPSK,8PSK,16-QAM distribution method is available with combination of both
DVB-S2 QPSK,8PSK,16APSK,32APSK unicast and multicast delivery mode depending on underlying
DVB-C QAM : 16- to 256- QAM IP network topology of service provider delivery network.
DVB-C2 COFDM: 16- to 4096-QAM Internet Group Management Protocol (IGMP) is used to join
DVB-T COFDM(OFDM) : QPSK, 16QAM, 64QAM and leave multicast streams. IP Multicast Standards include
DVB-T2 OFDM: QPSK, 16QAM, 64QAM, 256QAM IGMPv1, IGMPv2 and IGMP Snooping.
Analog AM(VSB),FM,QAM According to ETSI specification [7] the transport streams
should be encapsulated with Real-Time Transport Protocol
Video on demand can be recorded live video stream for (RTP). Real Time Streaming Protocol (RTSP) is responsible
later review or uploaded video content from various sources to for control over delivery of video content. Real-Time
a dedicated VoD server or media library. Sources can be live Transport Control Protocol (RTCP) is used to send statistics
TV, Video CDs, DVDs, Blu-ray discs, cameras and various and control information for QoS.
multimedia devices.
After video acquisition it must be processed by video V. USER-END
encoder device. Typically live video content is compressed
using H.264/MPEG-4 AVC codec. Other codecs can also be User devices for IPTV reception include: traditional set-top
used instead. The most common codecs used for VoD are box (STB), IP STB, PC, smartphone, tablet and any internet
MPEG-2, MPEG-4 and VC-1. In case video is acquired browsing device able to play video content.
directly in older MPEG-2 or other video format transcoder
device can be used to convert data into desired video codec.
VI. CONCLUSION

75
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

This paper describes current state of IPTV technology.

Main goal is to deliver better video quality at the lower bit
rate, offering virtually unlimited number of TV channels.
Adding new features and services is very easy and
straightforward. Services can be adopted with minimal
investments and with use of existing IP network infrastructure
delivered to users.

ACKNOWLEDGEMENT
This paper is supported by Experimental Research Unit
(UTzNIT) at the Technical University of Gabrovo.

REFERENCES
[1] O’Driscoll, G., “Next Generation IPTV Services and
Technologies”, John Wiley & Sons, Inc, 2008.
[2] Cooper, W., Lovelace, G., “IPTV.Guide”, informitv, 2006.
[3] Microsoft, “Microsoft TV IPTV Edition”, available online at
http://www.microsoft.com/TV.
[4] Cisco, “Cisco IPTV Solutions”, available online at
http://www.cisco.com/go/iptv.
[5] IEEE 802-2001: “IEEE Standards for local and metropolitan
area networks: overview and architecture”, 2002.
[6] ETSI TS 182 028: TISPAN NGN Release v.3.5.1: “NGN
integrated IPTV subsystem Architecture”, 2011.
[7] ETSI TS 102 034 v1.4.1 “Transport of MPEG-2 TS Based DVB
Services over IP Based Networks”, 2009
[8] Конов, К., “Цифрово радио и телевизионно разпръскване”,
Диос, 2011.

76
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Average SIR Comparison for SC Systems Using Different

Decision Algorithms in the Presence of Interference
Aleksandra Panajotović1, Dragan Drača2 and Nikola Sekulović3
Abstract – Ascertaining the importance of the selection Rayleigh statistics [5,6].
combining (SC) as the most efficient diversity technique, average There is set of performance criteria that allow the system
signal-to-interference ratio (SIR) at the output of dual SC designer to evaluate the performance of wireless systems and
receiver is investigated in the paper as important performance investigate influence of key system parameters. The most
criterion. The diversity system operates in microcell environment
and applies desired signal power algorithm. Numerical results
popular first order performance criteria are outage probability
are presented to show effects of fading severity and level of (OP), average bit error probability (ABEP), channel capacity,
correlation. Moreover, they are used to compare performance of average output SNR/SIR, etc [7-11]. In this paper, analytical
dual SC systems using different decision power algorithms. expression for average SIR at the output of dual SC receiver
operating over correlated Rician fading channels in the
Keywords – Cochannel interference, Fading, Selection presence of Rayleigh CCI is derived for the case when
combining. receiver applies desired signal power algorithm. Numerical
results illustrate the influence of system and channel
parameters on the system performance. Moreover, our
previous published result [12] is used to compare average
I. INTRODUCTION output SIR of considered SC system with SC system applying
SIR decision algorithm.
The performance of wireless system is severely affected by
fading and cochannel interference (CCI). Fading is result of
multipath propagation, and CCI is result of frequency reuse II. SYSTEM AND CHANNEL MODEL
[1]. Space diversity techniques, which combine input signals
from multiple receive antennas (diversity branches), are the If diversity system is applied in small terminal, correlation
well known techniques that alleviate the deleterious effects of arises between diversity branches due to insufficient antenna
fading and CCI. The most popular diversity techniques are spacing. In that case, desired signal envelope, r1 and r2, on
maximal-ratio combining (MRC), equal-gain combining two diversity branches follow correlated Rician distribution
(EGC) and selection combining (SC). Among of these types whose probability density function (PDF) is given by [13]
of diversity technique, SC has the least implementation
complexity since it processes only one of diversity branches rr  r 2  r 2  2b2 1    
pr1r2  r1, r2   12
exp  1 2 2 
[2]. Usually, SC receiver chooses the branch with the highest  (1   )  2 1   2  
4 2

signal-to-noise ratio (SNR), which corresponds to the
strongest signal if equal noise is assumed among the diversity   rr    br   br 
k Ik  2 1 2 2  Ik  2 1  Ik  2 2  , (1)
branches. In some systems, where CCI is more significant   1       1       1    
k 0
 
than noise, SC receiver can employ one of following decision
1, k  0
power algorithms: the desired signal power algorithm, the k  
total signal power algorithm and the signal-to-interference 2, k  0
power ratio (SIR) algorithm [3].
where  is branch correlation coefficient and Ik (·) is modified
There are few statistical models used to describe fading in
Bessel function of the first kind and k-th order. Rice factor
wireless environment. Rician statistical model is typically
observed in the line-of-sight (LoS) path of microcellular urban and average desired signal power are defined as K  b 2 /  2 2 
and suburban land mobile environment [4]. In such and    2 1  K  , respectively.
environment, CCI experiences significantly deeper fading We assume that there is a single dominant interferer,
than the desired signal. Therefore, the different fading models independent of the desired signal, subjected to Rayleigh
for the desired signal and CCI have to be used. Cochannel fading [14]. PDF of its envelope is expressed by
interference from distant microcell may be modelled by
a  a2 
pa  a   exp   2 
, (2)
1
Aleksandra Panajotović is with the Faculty of Electronic a 2
 2 a 
Engineering at University of Niš, Aleksandra Medvedeva 14, Niš
18000, Serbia, E-mail: aleksandra.panajotovic@elfak.ni.ac.rs.
2
where  a 2 is average CCI power.
Dragan Drača is with the Faculty of Electronic Engineering at
University of Niš, Aleksandra Medvedeva 14, Niš 18000, Serbia, E- The considered dual SC receiver uses the desired signal
mail: dragan.drača@elfak.ni.ac.rs. power decision algorithm. Actually, it selects the branch with
3
Nikola Sekulović is with the School of Higher Technical the largest instantaneous desired signal power, i.e.
Professional Education, Aleksandra Medvedeva 20, Niš 18000,
Serbia, E-mail: sekulani@gmail.com.

77
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

r 2  max r12 , r2 2  . The instantaneous SIR at its output is given  2K    2 p  k K n  l  k   0.5 

  exp     k
by   max r12 , r2 2  / a 2  r 2 / a 2 .  1    k , p ,n,l 0 n !l ! p !  p  k  1   l  k  1
k  n  l 1.5
S 1   
 k  n  l 1.5
III. AVERAGE OUTPUT SIR   n  k  1 1  K 1    (5)
  p  n  k   p  k  l  i  1.5  
Average output SIR is important performance criterion for   p  n  k  !   p  k  l  1.5    
  i 0 2 p  k l i  2 i ! 
wireless system operating in interference-limited
environment. The average SIR envelope,  , at the output of  p  l  k   p  k  n  j  1.5   

  p  l  k !   p  k  n  1.5    p  k  n j  2  ,
SC system can be obtained by averaging the instantaneous  j 0 2 j!  
SIR envelope,     r a , over its PDF [15]
where S is average input SIR defined as S   /  a 2 .

    p   d  , (3) The average SIR envelope at the output of SC system
0 applying SIR decision algorithm can be evaluated using
Numerical Integration in program package Mathematics after
where pµ (µ) for SC system with desired signal power substitution pµ (µ) [12]
decision algorithm was derived in [16] as
 2 K   2 k K p l  k
p     exp    
 1  r  k , p , n , l , m0 
2 k  2 n  p l  2
p  k 1
 2K    2 p  k K n  l  k 1  K 
p     exp     k   n  l  m  k  2  ra2 m r 2 n  k  4 n  4 k  2 p  2 l 3
 1    k , p ,n,l 0 n !l ! p !  p  k  1   l  k  1 
  m  1   l  k  1   n  k  1   p  k  1
S  2 p  2 k 1
  K  1
2 k  2 n  p l  2
  n  p  m  k  2
  n  k  11   
p
1   2 k  p  n l 
n ! p !m !l !1  r  1  r 
2 n  k 1 2 p  2 l  2 n  3 k 1

 
  
  
  p  n  k  !1  K   1    2  p  k  l  1 !
l 2l n

1  ra2 
n  l  k 1 m
   a2 n  2l  2 k  2 m
1   l   p  k l  2
   S  1  K    
 2 
n p k m 2
   2  K  1 
   1   2   1
    
   a2 1  ra2   1  r 2  
  
 F1  n  l  k  m  2, n  l  k  1, n  l  k  2,  2 
pnk  p  l  k  i  1!1  K i  2i  
2

  p  k l i  2


 n  l  k  1
i 0 iS 1 K  2 
2 
p  k  l  i 1
i! 1    
2

2 1  2



 
2 F1  n  p  k  m  2, n  p  k  1, n  p  k  2,  2 
(6)
  
 n  p  k  1

 
 p  l  k !1  K n  2n 1   l  2  p  k  n  1! 
 pk n2 1 r  2 n  p  k 1 m
 a2 n  2 p  2 k  2 m 
1   n   ,
a

S  
1 K  2  n l  k  m 2
   2  K  1  
 1   2   2
1
  
   a 1  ra   1  r  
2 2

 (4)   

p l  k  p  n  k  j  1!1  K  j  2 j  in (3). The parameters ra and r are branch correlation
   .  a2 1  ra2   K  1
pk n j 2  coefficients and
 
j 0 jS 1  K   
2 
 1  r 2 
.
 
2 p  k  l  j 1 j ! 1   2  
2 1  2 


  
IV. NUMERICAL RESULTS
Substituting Eq. (4) into Eq. (3), the average output SIR Previous proposed mathematical analysis is complemented
envelope is obtained in the analytical form using [17, Eq. in this section through illustration of influence of system and
(3.194(3))] channel parameters on the average output SIR as important
system performance criterion.
Figure 1 shows normalized average SIR (  / S ) at the
output of dual SC system applying different decision

78
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

algorithms in function of Rice factor. The performance sums in (5) and (6) converge for any values of branch
curves are evaluated for different values of correlation correlation coefficient and Rice factor. As is shown in Table I,
coefficient in order to show influence of distance between the number of terms that need to be summed to achieve the
diversity branches on considered performance criterion. desired accuracy strongly depends on branch correlation.
Regardless of applied decision algorithm, diversity gain
decreases with increase of branch correlation coefficient. Also TABLE I
diversity gain is greater for environment with light fading than NUMBER OF TERMS SUMMED TO ACHIEVE THREE-SIGNIFICANT-
for environment with severe fading. Comparison of Fig.1 (a) FIGURE ACCURACY OF AVERAGE OUTPUT SIR (DESIRED SIGNAL/SIR
ALGORITHM)
and Fig. 1 (b) shows advantage of SIR signal power decision
algorithm because it provides better diversity gain. That  = 0.1 10/11
advantage is more noticeable for greater values of Rice factor.  = 0.2 9/13
In environment with severe fading it is better to use SC  = 0.3 9/12
system with desired signal power algorithm because it  = 0.4 10/9
requires less complicate receiver and gives almost the same  = 0.5 11/16
diversity gain as SC system with SIR decision algorithm.

2.43
2.42  = 0.1
V. CONCLUSION
2.41  = 0.2
 = 0.3 In this paper the performance of dual SC system operating
Normalized average output SIR

2.40
 = 0.4 in interference-limited microcell environment has been
2.39
 = 0.5
2.38
studied. Actually, average output SIR as important
2.37
performance criterion has been derived in infinite series form
2.36
for the case when SC system using desired signal power
2.35
decision algorithm. Presented numerical results have
2.34
described influence of fading severity and correlation
2.33
coefficient on considered performance criterion. Moreover,
2.32
evaluated results have been compared with results obtained
2.31
for SIR decision algorithm. The general conclusion of this
2.30
paper is that SC diversity system with SIR algorithm provides
0 1 2 3 4 5 6 better diversity gain regardless of working conditions.
K [dB]

ACKNOWLEDGEMENT
a)
This work has been funded by Serbian Ministry for
2.43
Education and Science under the projects TR-32052, III-
2.42
44006, TR-33035.
2.41
Normalized average output SIR

2.40
2.39 REFERENCES
2.38
2.37 [1] W. C. Jakes, Microwave Mobile Communications, New York,
2.36
Wiley, 1974.
[2] M. K. Simon, M. –S. Alouini, Digital Communications over
2.35
r = ra = 0.1 Fading Channels, 2nd ed., New York, Wiley, 2005.
2.34
r = ra = 0.2 [3] L. Yang, M. –S. Alouini, “Average Outage Duration of Wireless
2.33 Communications Systems”, ch. 8, Wireless Communications
r = ra = 0.3
2.32 r = ra = 0.4 Systems and Networks, Springer, 2004.
2.31 r = ra = 0.5 [4] K. A. Stewart, G. P. Labedz, and K. Sohrabi, “Wideband
2.30 channel measurements at 900 MHz”, VTC’95, Conference
0 1 2 3 4 5 6 Proceedings, pp. 236-240, Chicago, USA, 1995.
K [dB] [5] Y. –D. Yao, A. U. H. Sheikh, “Investigations into cochannel
interference in microcellular mobile radio systems”, IEEE
b) Trans. Veh. Technol., vol. 41, no. 2, pp. 114-123, 1992.
[6] Y. –D. Yao, A. U. H. Sheikh, “Outage probability analysis for
Fig.1. Average output SIR versus Rice factor for several values of microcell mobile radio system with cochannel interferers in
correlation coefficient: Rician/Rayleigh fading environment”, Electron. Lett., vol. 26,
a) desired signal power algorithm; b) SIR power algorithm no. 13, pp. 864-866, 1990.
[7] I. Petrović, M. Stefanović, P. Spalević, S. Panić, and D.
The main problem in the infinite series expression of the Stefanović, “Outage Analysis of Selection Diversity over
average output SIR is its convergence. The nested infinite Rayleigh Fading Channels with Multiple Co-Channel

79
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA
Interferers”, Telecomm. Sys., accepted for publication, DOI: Rician Fading Channels in the Presence of Cochannel
10.1007/s11235-011-9438z. Interference”, AEÜ, pp. 1061-1066, vol. 63, no. 12, 2009.
[8] I. Petrović, Z. Nikolić, M. Stefanović, S. Panić, P. Spalević, and [13] M. K. Simon, Probability Distributions Involving Gaussian
Dj. Bandjur, “Multiple Co-Channel Interferers Influence on Random Variables – a Handbook for Engineers, Scientists and
Selection Combining over Correlated Weibull Fading Mathematicians, New York, Springer, 2002.
Channels”, Frequenz, accepted for publication. [14] H. Yang, M. –S. Alouini, “Outage Probability of Dual-Branch
[9] Dj. Bandjur, M. Stefanović, and M. Bandjur, “Performance Diversity Systems in the Presence of Co-Channel Interference”,
Analysis of SSC Diversity Receiver over Correlated Ricean IEEE Trans. Wireless Commun., vol. 2, no. 2, pp. 310-319,
Fading Channels in the Presence of Co-Channel Interference”, 2003.
Electron. Lett., vol. 44, no. 9, pp. 587-588, 2008. [15] G. K. Karagiannidis, “Performance Analysis of SIR-based dual
[10] Z. Popović, S. Panić, J. Anastasov, P. Spalević, and M. Selection Diversity over Correlated Nakagami-m Fading
Stefanović, “Performance Analysis of Selection Diversity over Channels”, IEEE Trans. Veh. Technol., vol. 52, no. 2, pp. 1207-
Exponentially Correlated -µ Fading Environment”, Intern. J. 1216, 2003.
Commun. Syst., vol. 24, no. 7, pp. 925-937, 2011. [16] A. Panajotović, N. Sekulović, M. Stefanović, and D. Drača,,
[11] S. Jovković, S. Panić, M. Stefanović, P. Spalević, and D. Krstić, “BEP Comparison for Dual SC Systems Using Different
“Performance Analysis of SSC Diversity Receiver over Decision Algorithms in the Presence of Interference”,
Correlated Ricean Fading Channels in the Presence of Co- TELSIKS’11, Conference Proceedings, vol. 2, pp. 467-470, Niš,
Channel Interference”, EURASIP J. Wireless Commun. Net., Serbia, 2011.
vol. 2010, DOI: 10.1155/2010/583093. [17] I. S. Gradshteyn, I. M. Ryzhik, Table of Integrals, Series, and
[12] A. Panajotović, M. Stefanović, and D. Drača, “Performance Products, CD version 1.0, Academic Press, 1996.
Analysis of System with Selection Combining over Correlated

80
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Optimization of Traffic Distribution Coefficients in

IP Radio-Relay Network with Path Diversity
Dragana Perić1, Miroslav Perić2, Branislav M. Todorović3,
Milan Šunjevarić4, Miroslav Popović5
Abstract - This paper presents results of optimization of majority of time (typically about 99.9% of time) it operates
traffic distribution coefficients between primary and backup with full capacity. In smaller percentages of time 0.01% to
route in IP radio-relay network. For generation of primary and 0.09% link operates with reduced capacity.
backup route NOPHY algorithm is used which ensures that they When smaller percentage of unavailability is needed, the
don't have common links. Results of network performance
only solution is to use diversity techniques in radio network,
simulation with random link degradation are given.
when signal is transmitted over independent links. Switching
Keywords - load balancing, backup routes, fading, traffic between paths could be done by simple rerouting [11], when
protection unavailable links are not used for IP packet transmission.
However, for rerouting network convergence process might
I. INTRODUCTION not be fast enough to track network changes caused by fading
[7]. In order to overcome this problem, cross-layer routing
techniques could be used [8].
Radio networks can be used for transmission of IP traffic One of such methods, described in [9],[10], exploits traffic
between computer networks. Achievable throughput of distribution on primary and backup routes, that can serve both
individual point-to-point link that is used for router as unavailability protection and network performance
interconnection in these networks is greater that 1 Gbit/s, improvement method. It is shown that pre-calculated backup
especially in the case of millimeter wave links [1]. This routes could be good protection mechanism and algorithms
solution is competitive to fiber optics and free space optics for for backup route generation are explained. Network
short distances in urban areas, and its advantage is quick performance improvement is achieved because traffic
installation and reconfiguration. Main drawback of high distribution over primary route and backup route can also
capacity radio link is its susceptibility to additional signal serve as load balancing and therefore decrease possibility of
attenuation (fading) caused by multipath propagation or rain link congestion [11].
attenuation, that can cause link outages [2]. In order to reduce This paper is focused on traffic distribution between
influence of fading on an individual IP link various techniques primary route and backup route. It is shown that it could be
of tradeoff between link capacity and signal level are used both for traffic protection and load balancing.
performed like: adaptive rate, adaptive coding and adaptive
modulation [3]. Such radio link could be described as a II. MODEL OF IP RADIO NETWORK
function of link capacity vs. receiver signal level. Using
propagation model described in [4] for each individual link, Network consists of N nodes with IP routers, nodes are
percentages of time when it operates at full capacity, reduced connected with E links, with capacities C={c(e)}, c(e)
capacity or when it is unavailable (in link down state) could =1,...,E. Traffic demands between two nodes are denoted by
be calculated. In the procedure of radio link planning, radio h(i,j), i=1,2..,N, j=1,2,..,N. It is assumed that traffic demand
link parameters like transmitter output power and antennas exists for each pair of nodes in the network, and they define
gains are chosen to meet link performance and availability traffic demand matrix H={h(i,j)}, h(i,j)>0, i≠j, i h(i,j)=0, i=j .
objectives. Usually telecom operators accept link performance
objectives given in ITU-R recommendations [5], [6].
Depending on link class, unavailability objectives are in range  0 h(1, 2) .. h(1, N ) 
from about 0.01% to 0.05% (about several tens of minutes to  h(2,1) 0 .. h (2, N )  (1)
several hours at annual level). When link is available, the H=
 .. .. 0 .. 
 
1
Dragana Perić is with "IMTEL Komunikacije a.d.", Institute of  h( N ,1) .. h( N − 1, N ) 0 
Microwave Techniques and Electronics, 165b M. Pupina Blvd., 11070
Belgrade, Serbia, E-mail: dragana@insimtel.com
2
Miroslav Perić is with “VLATACOM d.o.o.”, Research and Development Each traffic demand hij can be served by different flows
Center, 5 M. Milankovića Blvd., 11070 Belgrade, Serbia, E-mail: x(i,j,k), k=1,..,Nk(i,j). Number of flows Nk(i,j), can be different
miroslav.peric@vlatacom.com.
3
Branislav M. Todorović is with RT-RK, Institute for Computer Based for every pair of nodes (i, j), with Nk(i,j)≥1.
Systems, Narodnog Fronta 23A, 21000 Novi Sad, Serbia, E-mail: Routing algorithm assigns route r(i,j,k) to each flow
Branislav.Todorovic@rt-rk.com x(i,j,k). Routes are then memorized in the routing table R. For
4
Milan Šunjevarić is with RT-RK, Institute for Computer Based Systems,
path r(i,j,k), vector Dr(i,j,k) indicates if link belongs to path is:
Narodnog Fronta 23A, 21000 Novi Sad, Serbia, E-mail: micosun@eunet.rs
5
Miroslav Popović is with Faculty of Technical Sciences, Trg Dositeja
Obradovića 6, 21000 Novi Sad, Serbia, E-mail: Miroslav.Popovic@rt-rk.com

81
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Dr (i , j , k ) = {δ (i , j , k ,1),..., δ (i , j , k , e ),..., δ (i , j , k , E )} A bottleneck in the network is the link with maximum

(2)
i = 1, 2,.., N , j = 1, 2,.., N , k = 1, 2,.., N k (i , j ) load, so criteria for network performance measure is obtained
by maximization of normalized link loads.
where: Lmax = max L(e) (11)
e∈{1,.., E }
1, if link e belongs to route r(i, j,k)
δ (i , j , k , e ) =  (3) Links that have Ll(e) equal to Lmax are actually bottleneck
 0, otherwise of the network because they are most susceptible to
for each link e=1,…,E. congestion. Apart from maximum load, we can monitor
Using this notation, maximum number of flows using link average load:
e is defined by capacity of that link ce [11]: 1 E
N N Nk (i , j )
Lave = ∑ L (e)
E e =1
(12)
∑∑ ∑
i =1 j =1 k =1
δ (i, j , k , e) ⋅ x(i, j , k ) ≤ ce , e = 1,..E (4)
In both cases, smaller values of Lmax, Lave indicate that
i≠ j network has better performances, i.e. it has small possibility of
It is assumed that each individual link is realized by point- congestion in real traffic conditions. In cases when traffic
to point microwave link, which capacity varies with receiver demand cannot be served because it is directed to zero capacity
signal level: link, criteria become infinity.
C = C(nR) (5)
IV. TRAFFIC DISTRIBUTION ALGORITHMS
In absence of fading the link operates with full capacity
C0., while in fading condition it could be reduced by capacity
reduction factor cR, or be in link down state, when its capacity Even load balancing scheme implies that traffic is equally
is equal to zero. distributed among available paths. This scheme acts as load
balancing and is implemented in many commercially available
III. PERFORMANCE CRITERIA BASED ON LINK LOAD routers [12]. According to it, service of traffic demand h(i,j) is
equally split between flows x(i,j,k). Load balancing
Two performance criteria are defined, both based on link coefficients b(i , j , k ) equal between nodes i and j by x(i,j,k)
load, with purpose to monitor the highest link load in the are defined by expression:
network. Traffic demand h(i,j), i=1,2,..,N, j=1,2,..,N, i≠j is
served by load balancing among flows x(i,j,k), k=1,...,Nk(i,j). b(i, j , k ) equal = 1/ N k (i, j ), i = 1,..., N ,
(13)
Balancing factor b(i,j,k) is calculated as ratio of the flow
j = 1,..., N , k = 1,..., N k (i, j )
x(i,j,k) and the demand h(i,j) and it is the element of traffic
load balancing matrix B. where Nk(i,j) is number of flows serving traffic demand h(i,j).
x(i, j , k ) Simplicity of implementation is a very strong advantage
b (i , j , k ) = , k = 1,...,N k (i, j ) (7)
of this scheme, but it has two drawbacks.
h (i , j )
The first drawback is that coefficients calculation is based
It is clear that sum of balancing factors for all flows equals
only on number of flows that serve traffic demand, without
one:
Nk (i , j )
consideration of other traffic demands in network. That is why
∑
k =1
b(i, j , k ) = 1, b(i, j,k) ≤ 1 (8) traffic is unequally distributed in network.
The second drawback of this method is that if one route
Using elements of the matrix B, the vector of link loads is contains radio links that are temporarily unavailable due to
formed Ll with elements equal number of paths that include fading, traffic distributed to those links would be lost.
link e. In order to overcome these two drawbacks, we propose
N N Nk (i , j )
traffic distribution optimization by varying balancing factor
Ll (e) = ∑∑ ∑ b(i, j , k ) ⋅ δ (i, j , k , e), e = 1,..,E (9) according to link load b(i,j,k) to gain either minimum value of
i =1 j =1 k =1
Lmax as optimization goal, which we denote as Opt-Lmax, or to
In special case, when Nk(i,j)=1, i=1,...,N, j=1,...,N, each gain minimum value of Lave as optimization goal, which we
traffic demand is served by only one path, Ll(e) is an integer denote as Opt-Lave. Due to nonlinearity of Opt-Lmax, for its
that represents number of traffic demands served by given implementation NMinimize function for constrained
link. If traffic load balancing is applied, Ll(e) is a rational numerical global optimization is used. NMinimize function
number that represents equivalent number of flows that serve implemented in software package Mathematica [13] and it is
given link. state of the art combination of several optimization methods
Since the link load increases when the number of flows like: Nelder-Mead, differential evolution, simulated annealing
that this link serves increases, and decreases when this link and random search. Due to very intensive calculation caused
capacity increases, by simple division of Ll(e) with link by a large number of optimization variables, this algorithm
capacity we obtain the vector of normalized link loads: does not have practical value, and it is used just to obtain
lower limit value of Lmax. Unlike the first one, the second Opt-
Ll (e) Lave is linear and could be treated as a linear programming
L(e) = , e = 1,...,E (10) problem.
c(e)

82
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

V. PERFORMANCE ANALYSIS OF TRAFFIC DISTRIBUTION TABLE I. NETWORK PERFORMANCE IN ABSENCE OF FADING

ALGORITHMS Perform. Balancing BUR type
criterion algorithm ECMP2 NOPHY
A. Network topology Equal 0.0050 0.0070
Lmax
Analysis and performance calculation of proposed [paths / Mbit/s]
Opt- Lmax 0.0040 0.0040
methods are performed on test network with topology given in Opt- Lave 0.0080 0.0080
Equal 0.0040 0.0053
Fig. 1 Network consists of N=7 nodes, connected with E=18 Lave
Opt- Lmax 0.0040 0.0040
unidirectional links (9 bidirectional links). It is assumed that [paths / Mbit/s]
Opt- Lave 0.0040 0.0040
every node communicates with all other nodes in the network,
therefore 42 traffic demands should be served. D. Network performance in presence of fading
For performance analysis in presence of fading, it is
assumed that fading cause link capacity reduction of ten fold
cR=10, that gives new link capacity values of 100Mbit/s. In
the case when one link operates with lower capacity, due to
network redundancy, performance is not noticeably degraded.
Therefore, performance analysis considers the case when two
links simultaneously have lower capacity. Other cases when
three or more links have degraded capacity, for properly
tailored individual link fading margins have very low
probability, and therefore are not considered in this analysis.
Fig. 1. Network topology and example of NOPHY routes
Two links in network are randomly chosen to have
degraded capacity. Results are represented in form of
B. Traffic routing cumulative distribution function of performance criteria in
Primary route that serves specific traffic demand in such cases.
network is chosen by Dijkstra - shortest path algorithm [14]
with link costs inversely proportional to maximum link E. Performance Analysis Results
capacity. It is assumed that full link capacity is C0=1Gbit/s for In Fig.2. and Fig.3. values of CDF for performance criteria
all links in the network. Lmax and Lave respectively are given in case of ECMP2 when
In addition to primary route one backup route is defined, two links have ten fold capacity reduction.
for certain number of traffic demands. For performance
100
analysis in this paper, backup routes are calculated using two 90
algorithms: ECMP2 and NOPHY. 80
70
For some traffic demands there are several shortest path
CDF[%]

60
routes. In this network 18 from 42 traffic demands have 50
40
multiple shortest path. For each of them two equal cost routes 30
are chosen and traffic is balanced between them. This 20
10
algorithm is denoted as ECMP2 - Equal Cost Multi Path with 0
two routes. 0 0.02 0.04 0.06 0.08
The second algorithm for backup routes calculation is L max [ paths / Mbit/s ]
denoted as NOPHY and it is described in [9], [10]. The key Equal Opt-Lmax Opt-Lave
idea of this algorithm is traffic protection in case of link
outage, so the backup route is chosen as a shortest path route Fig. 2. Lmax in case ECMP2 two links have ten fold capacity reduction
that has no common links with primary route. Example of
NOPHY backup route choice for one traffic demand is shown 100
90
in Fig. 1. Where multiple shortest path route exists, the one of 80
them is chosen to be primary route, and according to it 70
CDF[%]

NOPHY backup route is calculated. In this case each of 42 60

50
traffic demands is served by two routes. 40
30
C. Network performance in absence of fading 20
Values of network performance criteria Lmax and Lave in 10
0
absence of fading, when all links in network operate with 0 0.005 0.01 0.015 0.02
maximum capacity of 1Gbit/s are shown in Table 1. L ave [ paths / Mbit/s]
As expected results have shown that, in majority of cases,
Equal Opt-Lmax Opt-Lave
some gain in load balancing could be achieved by varying
traffic distribution coefficients. Also it is shown that Opt-Lmax Fig. 3. Lave in case ECMP2 two links have ten fold capacity reduction
give good values both for Lmax and Lave, while Opt-Lave in this
case gives result for Lmax even worse than equal distribution. Fig.2. shows that according to Lmax criterion Opt-Lmax has
in 42% of cases better results than Opt-Lave, while both

83
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

algorithms have considerably better results than equal traffic calculation of traffic distribution coefficients could be
distribution. Similarly, according to Lave criterion Opt-Lave has performed in each router. The straightforward implementation
the best results (Fig 3.). Note that differences between CDF of optimization algorithm, especially for Opt-Lmax which is
curves for Opt-Lmax and Opt-Lave algorithms are much higher nonlinear, could be difficult and time consuming, and
for Lmax criterion than for the Lave criterion. therefore it could be subject for future work.
For NOPHY backup route choice algorithm, compared to
ECMP2, the difference between Opt-Lmax and Opt-Lave CDF VII. CONCLUSION
curves is greater according to Lmax criterion (Fig.4.).
According to this criterion Opt-Lmax has from 25% to 40% Usage of unequal traffic balancing coefficients between
better performance according to Lmax criterion in 75% of main and backup route in IP radio-relay network with path
cases, while in 25% of cases the performance is the same. On diversity can serve both as traffic protection and load
the other hand, difference between Opt-Lmax and Opt-Lave balancing method. Simulation on test network confirmed that
CDF curves according to Lave criterion is only about 10% but unequal traffic distribution gives up to three times better
in all cases (Fig.5). For NOPHY algorithm both optimization results for link loads than equal traffic distribution. It is also
algorithms have about three times better results than equal showed that optimizing the maximum of link load, besides
traffic distribution (Fig.4., Fig.5.). from giving the best values of the maximum link load, also
100 gives good performance according average link load criterion.
90 The main drawback of minimizing maximal link load in
80
70 optimization process is its nonlinear character, which requires
CDF[%]

60
50
much more processing time for its calculation than
40 minimizing average link load.
30
20
10
ACKNOWLEDGEMENT
0
0 0.02 0.04 0.06 0.08 This work was partially supported by the Ministry of
L max [ paths / Mbit/s]
Education and Science of the Republic of Serbia under Grant
TR-32024.
Equal Opt-Lmax Opt-Lave

Fig. 4. Lmax in case NO PHY two links have ten fold capacity reduction REFERENCES
100
90 [1] Wells J., “Multigigabit Wireless Technology at 70 GHz, 80 GHz and 90
80 GHz”, RF Design, pp.50-58, May 2006.
70
[2] Sklar B. Digital Communications: Fundamentals and Applications,
CDF[%]

60
50 Prentice Hall, 2001.
40
[3] Xiong F. Digital Modulation Techniques, Artech House, 2000.
30
20 [4] ITU-R Rec. P.530-13, "Propagation Data and Prediction Methods
10 Required for Design of Terrestrial Line of Sight Systems", 2009.
0
[5] ITU-R F.1703, "Availability objectives for real digital fixed wireless
0 0.005 0.01 0.015 0.02
links used in 27 500 km hypothetical reference paths and connections",
L ave [ paths / Mbit/s] 2005.
Equal Opt-Lmax Opt-Lave [6] ITU R F.1668, "Error performance objectives for real digital fixed
wireless links used in 27 500 km hypothetical reference paths and
Fig. 5. Lave in case NO PHY two links have ten fold capacity reduction
connections", 2004.
[7] Perić D., Perić M., Petrović G., "Redundant Topology in Computer
VI. REMARKS FOR PRACTICAL IMPLEMENTATION Network Realized with Millimeter Wave Radio Links", 14th IST Mobile
and Wireless Summit, Dresden, June 2005.
[8] Srivastava V, Motani M., "Cross-Layer Design:A Survey and the Road
For practical implementation of traffic distribution Ahead", IEEE Communications Magazine, Vol.43, pp.113-119,
coefficient optimization, it is important to have information December 2005.
about current link capacity, which is adjusted to actual fading [9] Perić D., Perić M., Todorović B., "Traffic Protection Method in IP
condition. Typical multipath fading event, known also as fast Radio Networks above 70 GHz", IEEE Communications Letters, Vol.
fading has noticeable signal level changes at about 10ms, 14, Nо. 10, pp. 981-983, October 2010.
while reaction time of adaptive modulation techniques is [10] Perić D., Traffic Protection Method in IP Radio Network above 70 GHz,
PhD Thesis, Faculty of Techical Sceiences, University of Novi Sad,
about 40ms [1][3][4]. For rain attenuation fading [4] signal June 2011.
level changes are much slower about 10s. According to this, [11] Pioro M., Medhi D., Routing, Flow, Capacity Design in Communication
for frequency bands above about 18GHz, where rain and Computer Networks, Morgan Kaufmann, 2004.
attenuation is predominant effect, information about link [12] Retana A., Slice D., White R., Advanced IP network design (CCIE)
capacities, could be obtained from the link hardware by traffic professional development, CISCO Press, 1999.
monitoring protocol such as SNMP [15]. For lower frequency [13] Wolfram S., The Mathematica Book, 5th ed., Wolfram Media, 2003.
bands more agile technology should be used. [14] Dijkstra, E.W. "A Note on Two Problems in Connection with Graphs",
Numerische Mathematik 1, 269–271., 1959.
When information about link capacities are obtainable in
[15] RFC 1157, "Simple Network Management Protocol (SNMP)", IETF,
entire network, since routing tables are known to all routers, May 1990.

84
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Optical Receiver Sensitivity Evaluation in Presence of

Noise in Digital Communication System
Krasen Angelov1, Stanimir Sadinov2, Nataliya Varbanova3
Abstract – The performance of an optical receiver in a digital shows a simple block diagram of the front end of an optical
optical communication link is studied. In the design of an optical receiver module.
receiver, it is vital that the module is capable of converting and The received optical signal is first converted into
shaping the optical signal while meeting or surpassing the photocurrent and amplified by the TIA. The limiting amplifier
maximum BER. Ultimately, the noise influence on the signal will
(LA) acts as a “decision” circuit, where the sampled voltage
determine the system sensitivity. The challenge is to find a way to
determine the QoS of an optical transmission channel v(t) is compared with the decision threshold vTH. At this data
independent of data format and bit rate within a short time decision point, the signal is significantly degraded by the
frame. The Q-factor itself significantly reduces measurement accumulation of random noise and inter-symbol interference
time and is thus more cost-effective. The analysis is based, (ISI), resulting in erroneous decisions due to eye closure [3,4].
assuming an input signal with impairment from factors like
inter-symbol interference, jitter, and transmitter relative
intensity noise. II. CHARACTERISTICS OF PERFORMANCE
ANALYSIS
Keywords – optical receiver sensitivity, bit error rate, inter-
symbol interference, transimpedance and limiting amplifier.
A. Eye diagram

The eye diagram represents a superposition of all bits in the

I. INTRODUCTION signal on top of each other [8]. Fig. 2 shows the eye diagram
of a NRZ signal. There are two basic types of adverse effects
Most applications of optical signals in digital visible in the eye diagram: Fist, the effect of intersymbol
communications require the detection and subsequent interference (ISI) and, second, the effect of jitter. The ISI is
conversion of the light to an electrical signal. In this process, caused by overlap of individual modulation pulses and it leads
the useful signal will be corrupted by noise and the ultimate to the amplitude errors at the sampling instances. The jitter is
sensitivity and performance of the system is limited by the defined as short-time deviations of a digital signal from its
noise characteristics. Optical receiver adds noise; usually ideal position in time [8]. A larger “eye opening” signifies less
thermal noise and shot noise. In communication systems, noise or distortion and therefore a higher quality of signal.
where electrical, radio or optical signals are transmitted; noise
can be viewed as an impairment resulting in the degradation B. Bit Error Ratio (BER)
of the information contained in the signal [1,7]. Optical
amplifiers can be used to improve the effective receiver In digital communication systems, the decision when to
sensitivity in optical systems. The optical amplifier works on sample and whether the sampled value represents a binary 1
the principle of stimulated emission [5]. The optical amplifiers or 0 is affected by noise and signal distortion in the real
add noise to the amplified signal, and at some point, this noise system and there is nonzero probability of an erroneous
becomes the dominant noise source. The basic manifestation decision. Therefore, the received signal quality is directly
of noise in optical amplifier is in the form of amplified related to the bit error rate (BER), which is a major indicator
spontaneous emission (ASE). So, the bit error probability of the quality of the overall system [8]. Eq. (1) explains a
(BER) is also affected by the ASE noise added by the optical calculation bit error rate if the Q-value is known [4,6]:
amplifier [2,3,7].
A typical optical receiver is composed of an optical photo  − Q2 
exp 
detector, a transimpedance amplifier (TIA) [9], a limiting 1  Q   2 .
BER = erfc  ≈ (1)
amplifier (LA), and a clock-data recovery (CDR) block. Fig. 1 2  2 Q 2π
1
Krasen Angelov is with the Faculty of Electrical Engineering and C. Q-factor
Electronics, Technical University of Gabrovo, 4 H. Dimitar St., 5300
Gabrovo, Bulgaria, E-mail: kkangelov@tugab.bg
2
Stanimir Sadinov is with the Faculty of Electrical Engineering The Q-factor expresses the quality of an optical signal with
and Electronics, Technical University of Gabrovo, 4 H. Dimitar St., respect to its signal-to-noise ratio (SNR). It includes all
5300 Gabrovo, Bulgaria, E-mail: murry@tugab.bg physical impairments of the signal, such as noise, non-linear
3
Nataliya Varbanova is with the Faculty of Electrical Engineering effects, dispersion (chromatic and polarization). These impair-
and Electronics, Technical University of Gabrovo, 4 H. Dimitar St., ments degrade the signal and cause bit errors. Consequently, a
5300 Gabrovo, Bulgaria, E-mail: nataliavarbanova@abv.bg higher value of the Q-factor means a better SNR and therefore

85
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 1. Simplified block diagram of the optical receiver module

in SNR
OMAmin = . (4)
ρ
In communication systems noise can be viewed as an
impairment resulting in the degradation of the information
contained in the signal [7]. The optical receiver adds two
types of noise namely thermal noise and shot noise. Since
optical amplifiers are based on the principle of stimulated
emission, its main contribution to noise is ASE noise.

A. Thermal noise

Fig. 2. An example of an eye-diagram and its interpretation The thermal noise of a receiver arises from the fact that
electrons in a receiver circuit have some probability of
a lower BER. Eq. (2) gives the Q-factor of an optical signal generating a current even in the absence of an optical signal.
[4,6]: This noise (referred to as Johnson noise), can be represented
by the variance of thermal current per unit frequency [5]:
V1 − V0
Q= , (2)
σ1 +σ 0 σ th2 = 4kT / R , (5)

where V1 is the value of the binary 1, V0 is the value of the where T is the absolute temperature, k is Boltzmann's
binary 0, σ1 is the standard deviation of the binary 1 and σ0 is constant and R is the detector load resistance.
the standard deviation of the binary 0.
B. Shot noise
III. OPTICAL RECEIVER SENSITIVITY EVALUATION
The shot noise arises from the Poisson distribution of the
In optical communication systems, sensitivity is a measure electron-hole generation by the photon stream. The latter is a
of how weak an input signal can get before the bit-error ratio stochastic process having random arrival times. On average,
(BER) exceeds some specified number. the number of electron-hole pairs created will be proportional
Sensitivity can be expressed as average power (PAVG) in to the number of photons, with a given constant of
dBm or as optical modulation amplitude (OMA) in WP-P (peek- proportionality.
to-peek). Each gives a figure of merit for the receiver [3,4,6]. During a given time interval, with a certain number of
To achieve the best optical sensitivity, it is important to photons incident upon the detector, the number of electron–
maximize the signal Q-factor before data decision. The hole pairs generated will have fluctuations as determined by
equation for calculating sensitivity is as follows [4,6]: Poisson statistics [5]. A dc photocurrent of Ipd will generate a
shot noise power density of:
 i SNR (re + 1) 
PAVG = 10 log n 1000 , dBm , (3) σ sh2 = 2eI pd . (6)
 2 ρ (re
− 1) 
D. ASE Noise
where in is the noise of TIA; ρ – responsivity flux
(conversion efficiency) of the photodetector, in A/W; re – the
ASE is produced by spontaneous emission that has been
ratio of a logic-one power level (P1) relative to a logic-zero
optically amplified by the process of stimulated emission in
power level (P0) [4,7].
gain medium. Noise associated with ASE is the limiting factor
The process in estimating the minimum peak-to-peak swing
in determining the ultimate signal-to-noise ratio in any system
of the optical signal begins with the choice of the maximum
using optical amplifiers [3,4,5]. The output ASE power can be
BER. This determines the signal-to-noise ratio (SNR). Next,
calculated using classical derivation in:
the RMS input referred noise (in) of the TIA and the
responsivity (ρ) of the photodetector must be found from the PASE = nsp (G − 1)hvBo , (7)
vendor’s data sheets. These are related by:

86
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

where h is Plank's constant, v is the optical frequency of GPin

transition, Bo is the optical bandwidth and nsp is the inversion SNR = . (13)
4hvnsp (G − 1) Be
parameter, given by
Provided that G is reasonably high, the SNR is determined
σ e (λ ) N 2
nsp = , (8) only by the input power and the inversion parameter nsp. More
σ e (λ ) N 2 − σ a (λ ) N 1 specifically, the SNR is independent of the gain. This is an
important result that governs the system performance and
where σa(λ) and σe(λ) are the absorption and emission cross determining of optical receiver sensitivity using Eq. (3):
sections, respectively; N1 and N2 are the population density in
lower and upper states; G is the overall gain of the amplifier.
Eq. (7) gives the ASE power for one polarization mode. So, F. CDR jitter-tolerance penalty
for single mode fiber, the right hand side of Eq. (7) must be
multiplied by a factor of 2. As the signal goes through the receiver amplifier chain to
When an amplified optical signal and accompanying the limiting stage, the amplitude noise is converted into timing
spontaneous emission are detected in a photodetector, the jitter at the data midpoint crossing. Random and deterministic
noise is transformed into the electrical domain and appears jitter is generated due to the existence of random noise,
along with the induced photocurrent as a noise current. limited bandwidth, passband ripple, group-delay variation,
Photodetection is a nonlinear square-law process [5]. The AC-coupling, and nonsymmetrical rise/fall times. The
photocurrent is therefore composed of a number of beat combination of these jitter components decreases the eye
signals between the signal and noise optical fields ES and EN, opening available for error-free data recovery. Consequently,
respectively, in addition to the squares of the signal field and CDR jitter-tolerance capability is another critical factor for
spontaneous emission field. The photocurrent Ipd is found as: determining optical sensitivity. CDR jitter tolerance is a
measure of how much peak-to-peak jitter can be added to the
I pd ∝ ( Etot ) 2 = ( E S + E N ) 2 = E S2 + E N2 + 2 E S .E N . (9) incoming data before errors occur due to misalignment of the
data and recovered clock. For a PLL-based CDR design, a
In Eq. (9), one can identify the first term as pure signal, the minimum data-eye opening is required, which is determined
second term as pure noise and it is referred to as spontaneous- by the clock-to-data sampling position, the retiming flip-flop
spontaneous (sp-sp) beat noise, and the third term as mixing setup/hold times, and the phase detector characteristics.
component between signal and noise and it is referred to as Assuming that the random jitter is RJRMS, the total
signal-spontaneous (s-sp) beat noise [3,6]. deterministic jitter is DJP-P, and the CDR minimum required
The power spectrum of current corresponding to s-sp beat eye opening is TOPEN at a specified BER, then the timing Q-
noise is uniform in the frequency interval (–Bo/2) to (Bo/2) and factor is defined as:
has an equivalent one-sided power density of
Tb − TOPEN − DJ P − P
2
Q= , (14)
4e 2 RJ RMS
N s − sp = Pin nsp (G − 1)G . (10)
hv
RJ P − P = 2QBER RJ RMS . (15)
The power spectrum of current corresponding to
spontaneous-spontaneous beat noise extends from 0 to Bo with When the jitter frequency at the CDR input is higher than
a triangular shape and a single-sided power density near dc of the PLL bandwidth, the CDR jitter tolerance (noted as JTP-P)
is related to TOPEN as:
N sp − sp = 2nsp2 (G − 1) 2 e 2 Bo . (11)
JTP − P = Tb − TOPEN . (16)
E. SNR calculation
To avoid degrading the optical sensitivity, the CDR high
frequency jitter tolerance should satisfy:
The Q-factor can be also expressed in terms of the optical
signal-to-noise ratio (OSNR) as: JTP − P ≥ 2QBER RJ RMS + DJ P − P
. (17)
Bo 2OSNR At the optical receiver input, it is assumed that the TIA is
Q= , (12)
Be 4 OSNR + 1 + 1 linear before the limiting amplifier. Therefore, random jitter
can be expressed as a function of the peak-to-peak current to
with Bo and Be the optical and electrical bandwidths, the total RMS noise ratio at TIA input:
respectively. Eq. (12) seems nonlinear and can be used to
derive the OSNR needed to obtain a given BER, for an ideal tr
RJ RMS = , (18)
system with only amplifier noise and without nonlinearities or I P − P / N TOTAL 0,6
inter-symbol interference.
In case of amplifier operating with moderately large optical where tr is dependent on the overall receiver small-signal
input signals, the SNR at the amplifier output is dominated by bandwidth BWTOTAL. Assuming a first-order lowpass filter:
signal spontaneous beat noise. In this limit, the equivalent
t r ≈ 0,22 / BW TOTAL . (19)
electrical SNR at the amplifier output is given by [5]

87
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The CDR jitter-tolerance penalty on optical sensitivity can From Figs.3 and 4 it is evident that for OSNR lower than
be estimated by combining Eqs. 17 and 18, then solving for IP- 12,75dB a dominant factor is the amplifier noise when
P as: determining the optical receiver sensitivity. For higher OSNR
dominant is the jitter-tolerance penalty. In general, to achieve
2QBER .t r .N TOTAL a specified BER, the minimum TIA input current should
I P−P = . (20)
(JTP−P
− DJ P − P ).0,6 satisfy the QBER in both amplitude and timing.

Substituting the Eq. (20) in Eq. (3) the optical receiver

sensitivity in terms of PAVG can be obtained.
V. CONCLUSION

By applying the technique presented in this paper, it is easy

IV. RESULTS to estimate and predict more realistic optical receiver
sensitivity. It is necessary to consider error sources in both
Examples are given for optical receiver using MAXIM amplitude and timing. It has been shown how the amplitude
devices MAX3277 TIA and MAX3272 LA. The datasheet and timing-error sources separately affect the overall receiver
parameters are as follows: NTIA = 0,35µA, Rf = 3,3kΩ, NLA = BER. A general expression of Q-factor can be predicted when
0,22mV, the total receiver small-signal bandwidth is 7.0GHz, shot noise and thermal noise are also considered. Using these
and ISI = 0. Assuming re = 6,6 and ρ = 0,85A/W, the guidelines, optical receiver performance can be accurately
calculated optical sensitivity (PAVG) versus optical signal-to- predicted. In reality, the optical input is not an ideal signal,
noise ratio (OSNR) is shown in Fig. 3. The results on Fig. 3 because it suffers random noise from the transmitter as well as
are based on Eqs. (3), (13) and (20). It is shown the minimum ISI from fiber dispersion. The approach presented in this
required optical sensitivity PAVG for a given SNR. For article can be used for estimating the signal Q-factor and,
example, when the SNR = 14,1dB (which is equivalent to BER therefore, determining the BER.
= 10–12 or QBER = 7,1) the optical sensitivity is –21,78dBm in
ideal case, –14,98dB considering optical link with amplifier
noise, and –14,59dB considering the jitter-tolerance penalty.
ACKNOWLEDGEMENT
Another useful representation of minimum required optical
sensitivity PAVG is the dependence of BER needed. The results This paper has been sponsored under the auspices of the
are shown in Fig. 4. “Increasing Efficiency and Quality of Service in PBX (TU –
Gabrovo)” project – a part of the University Center for
-12
Research and Technology (UTzNIT) at the Technical
-14 University of Gabrovo, contract Е1102/2011.
Optical Receiver Sensitivity, dBm

-16
REFERENCES
-18

-20 [1] O. Panagiev, Adaptive Compensation of the Nonlinear

Distortions in Optical Transmitters Using Predistortion.
-22 Radioengineering, vol. 17, № 4, Dec. 2008, pp. 55-58.
without noise [2] J. Barry, E. Lee, “Performance of Coherent Optical Receivers”,
-24 with amplifier noise Proceedings of IEEE, Vol. 78, No. 8, August 1990.
with jitter-tolerance
[3] K. Angelov, K. Koitchev, S. Sadinov, An Investigation of Noise
-26
10 11 12 13 14 15 16 17 Influences in Optical Transmitters and Receivers in Cable TV
OSNR, dB
Networks, ICEST 2006, Proceedings of Papers, pp.102-105,
Fig. 3. Optical receiver sensitivity versus optical signal-to-noise ratio Sofia, Bulgaria, 2006.
10
-6 without noise [4] K. Koitchev, K. Angelov, S. Sadinov, Determining Bit Error
with amplifier noise Rate in Digital Optical Transmission Network Using the Q-
with jitter-tolerance
Factor, ICEST 2010, Proc. of Papers, Vol. 1, pp.53-56, ISBN:
-8
10 978-9989-786-57-0, Ohrid, Macedonia, 2010.
[5] P. Becker, N. Olsson, Erbium-Doped Fiber Amplifiers:
-10 Fundamentals and Technology, Acad. Press, New York, 1999.
10
[6] R. Freeman, Fiber-Optic Systems for Telecommunications, John
BER

Wiley & Sons, New York, 2002.

-12
10 [7] S. Derevyanko, S. Turitsyn, "Bit Error Probability for Direct
Detection of Optical RZ Signal Degraded by ASE Noise and
-14
Timing Jitter," IEEE Lightwave Technol., vol.25, pp.638-643,
10
2007.
[8] V. Tejkal, M. Filka, J. Šporik, P. Reichert, Possibilities of
10
-16 Increasing Power Budget in Optical Networks, ElektroRevue,
-12 -14 -16 -18 -20 -22 -24 -26
vol.1, №4, December 2010, ISSN 1213-1539.
PAVG, dBm
[9] B. Karapenev, Analogue Circuits and Systems. Methodological
Fig. 4. Bit Error Rate versus minimum required optical receiver Handbook for Course Design. Publishing house „M-PRES”,
sensitivity (PAVG) ISBN 978-954-8455-47-3, pp. 28-36, 2012.

88
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

New Teletraffic Loss System - Polya/G/n/0

Seferin Mirtchev1, Rossitza Goleva1, Georgi Balabanov1 and Velko Alexiev1
Abstract – In this paper we have proposed to use Polya and by two parameters – the mean value and the variance [2]. This
Pareto distribution to describe peaked arrival processes. With way we present a basic study of the influence of the offered
this arrival processes it is possible to analyze the real loss systems traffic’s peakedness on the blocking probability.
in telecommunication networks. We study a version of the The MMPP (Markov Modulated Poisson Process) traffic
“classical” Erlang's loss model M/G/n/0 called Polya/G/n/0. This
model is proposed in [3]. It accurately approximates the long
is a full accessible loss system with Polya input flow (negative
binomial distributed number of arrivals in a fixed time interval), range dependence characteristics of Internet traffic. The
general distributed service time and “n” servers. This model is hidden Markov models (D-BMAP/D/1/k) is presented in [4]
evaluated by simulation with Pareto distributed inter-arrival with an arrival process of frames and service process in the
time. An algorithm for calculation of the state probabilities and wireless channel. The method for analysis of the M/D/n is
blocking are presented. It is shown that the variance of the input given in [5]. A simple accurate model for a multi-server
stream changes significantly the characteristics of the loss central-queue system M/G/K when service requirements have
systems. heavy tails is considered in [6]. The bursty Internet traffic
stream is studied in [7,8] as an ON/OFF model.
Keywords – Polya distribution, Pareto distribution, loss
The main reason for studying the Polya/G/n/0 teletraffic
system, peaked flow.
system is that it can be used to analyze the real loss systems in
telecommunication networks. We consider that the network
analysis requires a technique that can represent any kind of
I. INTRODUCTION traffic, and especially peaked.

The most common choice for telecommunication network

design is based on the exponential assumption. Usual choice II. POLYA ARRIVAL PROCESS
is the Poisson arrival of the calls or sessions and exponential
service times. However, networks and applications of today The Polya arrival process is a pure birth process with two
generate a traffic that is bursty over a wide range of time parameters [9]. The probability Pi(t) of i arrivals in an interval
scales. A number of empirical studies have shown that the with duration of t seconds is given by
network traffic is self-similar or fractal in nature. 1
−
There is no single traffic model that can efficiently capture Po (t ) = (1 + βλt ) β

the traffic characteristics of all types of networks. The study i

(1)
1 / β + i − 1  βλt 
of traffic models for a specific environment has become a Pi (t ) =     Po (t ) ,
crucial and important task. Good traffic modelling is also a  i   1 + βλt 
basic requirement for accurate capacity planning.
In [1] a careful overview of some of the widely used where λ > 0 and β > 0.
network traffic models is made, highlighting the core features The Polya distribution is a variant of the negative binomial
of the model and traffic characteristics they capture. For distribution. Its mean value (the average number of arrivals in
heavy-tailed traffic it can be shown that the Poisson model an interval of length t) is
underestimates the traffic. In case of high-speed networks
M (t ) = ∑i=1 iPi (t ) = λt .
∞
with unexpected demand on packet transfers, Pareto (2)
distribution is a good candidate since the model takes into
consideration the long-term correlation of packet arrival This means that λ is an arrival rate.
times. The variance of the number of arrivals in an interval of
In this work, we propose to use the Polya and Pareto length t is
distribution to describe the input flows. We study a version of
V (t ) = ∑i=0 [i − M (t )]2 Pi (t ) = λt (1 + βλt ) .
∞
the “classical” Erlang's loss model M/G/n/0 called (3)
Polya/G/n/0. This is a full accessible loss system with Polya
input flow (negative binomial distributed number of arrivals The peakedness of the Polya input flow is
in a fixed time interval), general distributed service time and
“n” servers. V (t )
The Polya arrival process is peaked process. It is defined z (t ) = = 1 + βλt > 1 . (4)
M (t )
1
Seferin Mirtchev, Rossitza Goleva, Georgi Balabanov and Velko When β = 0, M(t) = V(t) = λt i.e. it is a regular Poisson
Alexiev are with the Faculty of Telecommunications at Technical process. When β = 1 the Polya distribution is a geometric
University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, Bulgaria, E- distribution.
mails: stm@tu-sofia.bg, rig@tu-sofia.bg, gbalabanov@ieee.org,
velko.alexiev@gmail.com

89
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

III. MODEL DESCRIPTION  1 / β + i − 1  β A 

i

   
Pi = 
i  1+ β A  (7)
Let us consider a multi servers loss system Polya/D/∞∞ with k
n
 1 / β + k − 1  β A 
a Polya input stream with arrival rate λ, constant service time
∑
k =0 

 k
 
 1+ β A 
τ and infinite number of servers. This queueing system is a  
non-Markovian model or renewal process (Fig. 1).
The name truncated is due to the fact that the solution may
be interpreted as a conditional Polya distribution Pi(| i≤n). The
Sources Servers Ao fact that we are allowed to truncate the Polya distribution
1
1 means that the relative ratios between the state probabilities
are unchanged. We prove this fact by simulation.
2
A=λτ 2 IV. TRAFFIC CHARACTERISTICS OF
M, V POLYA/D/N/0 SYSTEM

Knowing the state probabilities, we are able to find all

performance measures defined by state probabilities.
∞ n Time Congestion
The probability that all n channels are busy at a random
Fig. 1. Generalised queueing model with peaked input flow. point of time is obtained from system equations when i = n:
n
1 / β + n − 1  β A 
   
To study this system, we consider two epochs (points of  n   1 + β A 
Bt = k
(8)
time) t and t + τ at a distance of τ. Every customer being n
1 / β + k − 1  β A 
served at epoch t has left the server at epoch ∑ 
k
  
t + τ. The numbers of the customers arriving during the k =0  1+ β A 
interval (t; t+τ) are still in the system at epoch t+τ (being Carried traffic
served) and they are described whit Polya distribution as the
arrival process. By definition the carried traffic is:
For this model with Polya arrival process the offered n
traffic is equivalent to the average number of call attempts in An = ∑ i Pi , erl. (9)
service time: i =1

A = λ τ , erl. (5) Utilization

Traffic carried by one channel:
The offered traffic is equal to the carried traffic because
there are no losses and delays and it is called intended traffic.
The states probability of the system under the assumption ρ = An n , %. (10)
of statistical equilibrium can be calculated by means of Polya
distribution because we choose arbitrary epoch to observe the Traffic congestion
system: The ratio of traffic lost and intended traffic:
1
−
Po = (1 + β A) β A − An
Ba = (11)
 1 / β + i − 1  β A 
i
(6) A
Pi =     Po
 i  1+ β A  There are different time and traffic congestion because the
arrival process is peaked.
The number of busy channels at a random point of time is
Insensitivity
thus Polya distributed with both mean value equal to the
offered traffic A and different variance which depends from A system is insensitive to the holding time distribution if
the chosen parameter β. the state probabilities of the system only depend on the mean
We still assume the same arrival process. The number of value of the holding time.
channels is now limited so that n is finite. The number of It can be shown with Cox distribution that loss formula
states becomes n+1. In this case we get the truncated Polya (11), which above is derived under the assumption of constant
distribution: service time, is valid for arbitrary service time distributions.
The state probabilities for both the Polya distribution and the
truncated Polya distribution only depend on the service time

90
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

distribution through the mean value which is included in the 1

offered traffic A. We prove this property by simulation. Mp = (16)
λ
It can be shown that all classical loss systems with full
accessibility are insensitive to the holding time distribution The mean value is the average inter-arrival time for our
[5]. study. The parameter λ is the average call arrival intensity.
The variance of the Generalized Pareto Distribution is:
V. SIMULATION WITH PARETO 1 + 2β
DISTRIBUTION Dp = (17)
λ2
The fundamental relationship between the number and
With these substitutions for every positive value of λ and
interval representations is given by Feller-Jensen's identity:
β the mean end variance of he GOD will be finite.
P{N t < n} = P{Tn ≥ t}, n = 1,2,3,... (12) Random number generation
where: Nt is the random variable for the number of calls Many programming languages do not yet recognize the
arrived in time interval t Pareto distribution. In the field of telecommunications, the
Ti is the random variable for the time interval until Pareto distribution is widely used to estimate the inter-arrival
there has been n arrivals. and service times.
Simulations are the main tools for studying the One can easily generate a random sample from Pareto
performance of telecommunication networks and we will distribution by using inverse distribution function. Given a
evaluate the Polya/D/∞ and Polya/G/n/0 queue using random variable U with uniform distribution on the unit
simulation. interval (0;1), the random variable x is Pareto-distributed.
The family of Generalized Pareto Distributions (GPD) has
σ (U −ξ − 1)
three parameters: the location parameter µ, the scale x= (18)
parameter σ and the shape parameter ξ. If we choose the ξ
location parameter µ = 0 then the cumulative distribution We have developed a real time trace simulation algorithm
function of the GPD is: for evaluating the state probabilities of the Polya/D/∞ system
1
and make a comparison with M/D/∞. The simulation results
−
 ξ x ξ have shone that with Pareto distribution inter-arrival time the
F ( x) = 1 − 1 + , (13)
 σ  state probabilities of the Polya/D/∞ system have the same
mean e variance as the Polya distribution. When there are
where σ > 0 and ξ > 0. truncations (Polya/D/n/0) or we change the service time
The mean value of the generalized-Pareto distribution is: distribution (exponential and Pareto) the state distribution by
simulation is identical with by Polya’s formula (7) and (6).
σ
Mp = . (14)
1− ξ VI. NUMERICAL RESULT
When ξ < 1 the mean value is finite. In this section, we give numerical results obtained by a
The variance of the Generalized Pareto Distribution is: Pascal program on a personal computer. The described
methods are tested over a wide range of arguments.
σ2
Dp = . (15) Figure 2 illustrates the stationary probability distribution
(1 − ξ ) 2 (1 − 2ξ ) in a loss system Polya/G/n/0 with a Polya input stream with
different peakedness z, 15 erl offered traffic and 50 servers. It
When ξ < 0.5 the variance is finite. is seen that when the peakedness increases the probability
The probability density function of the GPD is: than all servers are busy increase vastly. Figure 3 shows the
− 1 −ξ
traffic congestion as a function of the intended traffic for
1  ξ x ξ
various values of the peakedness when the number of the
f (t ) = (1 + λ ) 1 +  (16)
σ  σ  servers is 30. The peaked input flow increases the congestion
with several orders. Figure 4 presents the utilisation of the
We choose these substitutions servers (the carried traffic for one server) as function of the
intended traffic intensity with different peakedness of the
1− β β Polya input flow when the number of the servers is 20. The
σ= ; ξ= . (17)
λ (1 + 2 β ) 1 + 2β peakedness of the arrival stream decreases the utilisation end
in the same time increases the congestion.
Therefore, we receive another form of the mean value of It is shown that the influence of the variance of the input
the generalized-Pareto distribution: stream over the performance measures is significant and can
be easily evaluate by Polya and Pareto distribution.

91
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

VII. CONCLUSION [9] Korn G. and T. Korn. “Mathematical Handbook for Scientists
and Engineering”, McGraw Hill, 1968, pp. 575.
In this paper the Polya distribution is used to describe the
peaked arrival processes in telecommunication networks. A 1E+00
basic generalised multiple loss teletraffic model Polya/G/n/0 A = 15 erl
1E-01
with peaked input flow, generalise service time and full n = 50
accessibility is investigated. All performance measures of 1E-02
interest are estimated. The idea is based on the Polya

State probabilities, Pi
distribution as an analytical continuation of the Poisson 1E-03
distribution and the classical M/G/n/0 system. M/G/n/0
1E-04 z = 1.02
The proposed approach provides a unified framework to
z = 1.5
model peaked teletraffic in real telecommunication systems. z = 2.0
1E-05
Numerical results and subsequent experience have shown that z = 3.0
this approach is accurate and useful in analyses of traffic 1E-06 z = 4.0
systems and especially in Quality of Service and performance z = 5.0
parameters estimation. 1E-07
The importance of this multiple loss teletraffic system with 0 5 10 15 20 25 30 35 40 45 50

peaked input stream comes from its ability to describe Number of the calls in the system

behaviour that is to be found in complex real time queueing

systems. It is a general traffic system which is important in Fig. 2. Stationary probability distribution
designing telecommunication networks.

1E+00
ACKNOWLEDGEMENTS
Polya/G/n/0
1E-01
n = 30
Traffic congestion probabilities, Ba

This paper is sponsored by project “DVU_10_0271 "Public

1E-02
and Private Multimedia Network Throughput Increase by
Creating Methods for Assessment, Control, and Traffic 1E-03
Optimization" with National Science Fund, Bulgarian M/G/n/0
Ministry of Education and Science, 2010 – 2013 leaded by 1E-04 z = 1.02
z = 1.5
Prof. Ph.D. Vladimir Poulkov.
1E-05 z = 2.0
z = 3.0
REFERENCES 1E-06 z = 4.0
z = 5.0
1E-07
[1] Chandrasekaran B. “Survey of Network Traffic Models”. This
0 3 6 9 12 15 18 21 24 27 30
report is available on-line at
Intended traffic A, erl
http://www.cse.wustl.edu/~jain/cse567-06/traffic_models3.htm
[2] Ramos H., D. Almorza and J. Garcia-Ramos. “On
Characterizing the Polya Distribution”. ESAIM: Probability and Fig. 3. Traffic congestion as a function of the intended traffic for
Statistics June 2002, Vol. 6, pp.105-112, URL: various values of the peakedness.
http://www.emath.fr/ps/
[3] Muscariello L., M.Mellia, M.Meo, M.Ajmone Marsan, R. Lo
Cigno, “An MMPP-Based Hierarchical Model of Internet 1.0
Traffic”, IEEE ICC 2004, Vol. 27, no. 1, June 2004, pp. 2143-
2147. Polya/G/n/0
0.8
[4] Moltchanov D., Y. Koucheryavy, J. Harju. “Non-preemptive n = 20
_iD-BMAPi/D/1/K queuing system modelling the frame
transmission process over wireless channels”, ITC 19, Beijing,
Utilisation, %

0.6
China, Vol. 6a, 2005, pp. 1335-1344.
M/G/n/0
[5] Iversen, V. B. “Teletraffic Engineering Handbook”, ITU-D &
z = 1.02
ITC. 312 pp. Edition spring 2004, 0.4
http://www.com.dtu.dk/education/34340/. z = 1.5

[6] Psounisa K., P. Molinero-Fernandez, B. Prabhakar, F. z = 2.0

Papadopoulosd. "Systems with Multiple Servers Under Heavy- 0.2 z = 3.0

Tailed Workloads". Performance Evaluation 62, 2005, pp. 456– z = 4.0
474. z = 5.0
[7] Yang X. "Designing Traffic Profiles for Bursty Internet Traffic". 0.0
0 4 8 12 16 20 24 28
Proceedings of IEEE Global Internet, 2002.
Intended traffic A, erl
[8] Siriwong K., L. Lipsky, R. Ammar, "Study of Bursty Internet
Traffic", Sixth IEEE International Symposium on Network
Computing and Applications (NCA 2007), 2007, pp.53-60. Fig. 4. Utilization of servers

92
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

An Evaluation of an UMTS/WLAN Interworking

Architecture using IEEE 802.21
Alexandru Vulpe 1 and Octavian Fratu 2
Abstract – This paper proposes an interworking architecture These access networks represent an optimal answer for
for mobility between UMTS/HSDPA and WLAN access covering the regular functionalities of the access
networks based on the IEEE 802.21 (Media Independent networks in a high mobility environment, especially for
Handover - MIH) standard. This architecture uses MIH interactive or conversational services, but not at very
signalling and blocks to provide information flows for a Resource low operational costs;
Manager which controls the handover. The architecture is
flexible and can be used for other types of access networks, but  The WLAN/WiFi-type of access networks, including
the focus is on UMTS and WLAN interworking because of the here the latest versions of this family of wireless access
common presence of these two technologies in today's wireless network standards, very simple for deployment and
networks. The architecture is evaluated using the EXata suitable in the case of stationary or nomadic wireless
Network Emulator, and results are presented as handover time access, with low costs both for investments and for
between UMTS and WLAN networks. operation;
Keywords – access network selection, vertical handover,  The WiMAX type of access networks, very convenient
heterogeneous wireless networks, IEEE 802.21. for deployment in rural areas or in emerging regions,
brings the benefit of reasonable quality in NLOS
propagation conditions in a low mobility environment,
but the quality of service (QoS) in high mobility
I. INTRODUCTION conditions is questionable;
 The digital broadcasting technologies (DVB-H, but also
The 3rd Generation of mobile communication systems is
DVB-T or DVB-T2) can bring a benefit for services or
not focused on a ``golden service'', responsible with the most
applications including a large broadcast or multicast
revenues. It considers a diversity of services, with different downlink distribution of the required information;
requirements and, consequently, with different solutions for
an optimal access technology. For this reason, a  The WPAN technologies can represent a very attractive
heterogeneous access network was a continuous dream both access solution for low or very low range, with a
for operators and for the network equipment providers and potential good management of the battery and of the
only the technical difficulties and the estimations of higher radio spectrum.
costs for such access networks made that the interest for such It is clear that the diversity of the available access network
solutions to be moderate. Fortunately, the technical progress technologies represents both a technical challenge and a
in this matter is now important (the approval of the IEEE potentiality in optimizing the access network performances.
802.21 standard [1] focused on Media Independent Handover Due to the popularity of the UMTS 3G networks and of the
– MIH is only an example) so the doubts on the technical WLAN technologies, we consider that both represent a must as
difficulties related to the implementation of such starting point in any discussion related to the performances and
heterogeneous access networks for 3G mobile the optimization of the implied heterogeneous network and, for
communications systems, as well on the corresponding this reason, these technologies will be discussed in the present
CAPEX and OPEX must be now reconsidered. paper.
There are different candidates for the access network The rest of the paper is organized as follows: Section II
technologies in order to be considered as suitable in a presents some related work in this field, Section III presents
heterogeneous environment for the 3G mobile communications the proposed algorithm and Section IV presents its
systems: implementation in the EXata Network Emulator. Simulation
 The regular 3G access network (the two well-known approach and results are presented in Section V while
FDD (WCDMA) and TDD (TC-CDMA) access conclusions are drawn and further work is outlined in Section
networks promoted by 3GPP for UMTS-like systems or VI.
the one proposed by 3GPP2 for CDMA2000 systems).
II. RELATED WORK
1
Alexandru Vulpe is with the Faculty of Electronics, Many papers have studied and are still studying ways to
Telecommunications and IT at POLITEHNICA University of achieve interworking between heterogeneous wireless
Bucharest, 1-3 Iuliu Maniu Blvd, Bucharest 061071, Romania, E-
networks.
mail: alex.vulpe@radio.pub.ro.
2
Octavian Fratu is with the Faculty of Electronics,
Telecommunications and IT at POLITEHNICA University of
Bucharest, 1-3 Iuliu Maniu Blvd, Bucharest 061071, Romania.

93
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

WLAN discovery schemes based on MIH Services are

presented in [4]. They are either channel information based,
location information based, or channel and location
information based. The obtained results are good, but the one
thing assumed and lacking at the moment is an implementation
of a Media Independent Information Server (MIIS).
The authors in [5] propose an interoperability mechanism
for WLAN and UMTS-HSDPA based on the output of a cost
function as handover metric. However the study is carried out
for HIPERLAN/2, which hasn't been widely implemented in
commercial networks.
Another paper [6] proposes a load balancing scheme using
Admission Control and Vertical Handoff in a cellular/WLAN
integrated network. It measures its performance by means of
data call throughput per voice call arrival rate, which is a
questionable evaluation metric.

III. FRAMEWORK DESCRIPTION

The flow of the proposed framework is described in Fig. 1.
A Resource Manager (RM) that is running on the mobile node
collects the Received Signal Strength (RSS) (or Received
Signal Code Power (RSCP) for UMTS) and Carrier-to-
Interference-plus-Noise-Ratio (CINR) (or Ec/No for UMTS)
values from periodic measurement reports done by the multi-
radio mobile node (MN). These are fed into a decision-
making engine, where it is possible to implement any
handover decision-making algorithm. It should be noted that
the RM could collect any other handover metrics (such as
available bandwidth, user preference or cost), and that these
are not currently reported by measurements done by
technology-specific link layers. In the end, the RM should
output the best point of attachment (PoA) to which the mobile
node should be connected. The PoA is termed “the best” in
respect to it fulfilling the handover decision criteria as defined
by an access network selection algorithm.
The RM then issues a primitive called “Report_Best_PoA”
which carries the PoA identifier and the radio access
technology that is enabled on that specific node. This is
broadcast to the lower layers of each radio access technology
protocol stack of the Mobile Node where it is stored. When
Fig. 1. Proposed framework flow the mobile node roams inside the coverage area of a new PoA,
two situations may occur, depending whether it is of the same
The current solutions considered in standardization bodies technology or not. If it is of the same technology, then, if it
are discussed in [2]. These include I-WLAN (Interworked- becomes the best PoA (determined by the RM), then normal
WLAN), GAN (Generic Access Network) and IEEE802.21. technology-specific (horizontal) handover procedures (cell
Of these, I-WLAN is currently adopted by 3GPP, the GAN selection/reselection - in UTMS or association and
model, while not being restricted to WLAN networks is more authentication - in WLAN) are carried out. Otherwise, nothing
commonly implemented as unlicensed mobile access (UMA)- is done, and the RM keeps receiving measurement reports. If
enabled terminals, and IEEE802.21, although offering a the newly discovered PoA is of another technology, then,
flexible framework to facilitate different handover scenarios, naturally, the radio interface of the corresponding technology
has not seen adoption in 3GPP, but only in IEEE 802 belonging to the mobile node will want to carry out an
standards. attachment procedure to the PoA (the terms used are PS
In [3], the authors propose a WLAN/HSDPA Interworking attach for UMTS and association for IEEE 802.11). However,
Architecture for Linux-based mobile terminals. It is this is not wanted if there is no need for a handover (the
implemented on a Linux laptop with commercially available current PoA has better SNR, or greater bandwidth etc., i.e. it
HSDPA modems and WLAN cards. However, the testing is is still the best one), so the attachment procedures will not be
done through a movement simulator, which could provide executed until that PoA becomes the best PoA.
inaccurate results.

94
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

TABLE I
802.21 PRIMITIVES IMPLEMENTED IN EXATA

MIH Service Primitive Description

Layer 2 connectivity
Event Link_Up
established
Event Link_Down Layer 2 connectivity loss
Link_Parameters_ Link parameters periodic
Event
Report reporting
Layer 2 connectivity loss
Event Link_Going_Down
imminent
Request parameters from
Command Link_Get_Parameters
link layer
Request an action on a link
Command Link_Action
layer connection

When the MN becomes attached to a PoA, it sends a

Link_Up.indication as per IEEE 802.21 specifications [1].
When the RM receives it, it will send a Link_Action.request
message to the Link Layer of the radio access technology
corresponding to the old PoA. It contains the command to
disconnect from the old PoA. Therefore, the control of the
handover procedure is partially taken over by the Resource
Manager.
Fig. 2. Scenario Topology
IV. IMPLEMENTATION
the implementation of the MIH User to be above the
The framework described in Section III has been MIHF, so, for ease of coding, it is implemented at the
implemented in the EXata Network Simulator/Emulator. It same layer as the MIH Function.
uses a layered architecture which is similar to the one used in Simulations regarding the accuracy and performance of the
the TCP/IP protocol stack. In order to implement new implementation of the protocol have been published in
protocols, or modify existing protocols, the source code previous papers [8-10]. However, there we had not
(written in C/C++) of EXata has to be modified [7]. The main implemented a UMTS link extension, but only dealt with
components of the IEEE 802.21 standard that have been
WLAN and WiMAX.
implemented are:
 The MIH Function (MIHF): It receives MIH
messages from the local link layers or a peer MIH node V. SIMULATIONS AND RESULTS
and forwards them to the MIH User. It is implemented
as a protocol included in the Network/IP layer of the The simulated network topology consists of a UMTS Radio
EXata protocol stack. Access Network (RAN) and a WLAN network connected to
each other via an IP network. The topology is depicted in Fig.
 The MIH Protocol: It provides a way for conveying 2. Node 2 is a mobile node with both UMTS and WLAN radio
MIH primitives between remote MIH nodes and interfaces and is connected to the UMTS RAN (Nodes 3-8)
includes an acknowledgement service in the case that and the WLAN Access Point (Node 12). Nodes 9, 10 and 11
the transport protocol used is unreliable
form an IP network backbone which connects the UMTS
 The Link SAP: It is a collection of primitives used for RAN to the WLAN network. The MN moves through the
interfacing between the media dependent link-layer and UMTS network, following the path outlined by the read
the MIH Function. Only a selected set of primitives has waypoints in Fig. 2, and comes into the coverage area of the
been implemented, and they are represented in Table I. WLAN Access Point. At this point, the MN would have the
The SAP is implemented as a separate library and tendency to become associated with the AP, but, instead, the
included in the EXata source code. procedure described in Section III takes place: only when the
 Link layer extensions: The media dependent link RSS of the WLAN access point becomes better than the one
layers have been modified in order to support the from the UMTS network, by a margin of 2 dB, the Resource
above-described algorithm. These are done by Manager, by processing measurement reports, determines that
modifying the source code of the MAC Layer of the the best PoA is the WLAN Access Point, and, therefore, the
EXata protocol stack associated with a specific link- normal procedure of association and authentication takes
layer technology (in this case, UMTS and WLAN). place. A Link_Up.indication primitive, originating from the
WLAN link layer of the multi-radio mobile node is sent to the
 MIH User: The MIH User, which is termed here RM via the MIH Function. The RM then sends a
“Resource Manager”, has been implemented as a Link_Action.request message to the Mobile Node's UMTS
simple RSS-based handover decision, in order to prove
the concept. The IEEE 802.21 standard does not restrict

95
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

the grant no. 106/2011 “Evolution, implementation and

transition methods of DVB radiobroadcasting using efficiently
the radio frequencies spectrum” and by the Sectorial
Operational Programme Human Resources Development
2007-2013 of the Romanian Ministry of Labour, Family and
Social Protection through the Financial Agreement
POSDRU/107/1.5/S/76903.

REFERENCES
[1] IEEE 802.21-2008, Standard for Local and Metropolitan Area
Networks - Part 21: Media Independent Handover Services,
IEEE Std., Jan. 2009.
[2] R. Ferrus, O. Sallent, and R. Agusti, “Interworking in
Fig. 3. UMTS to WLAN handover delay heterogeneous wireless networks: comprehensive framework
and future trends,” Wireless Communications, IEEE, vol. 17, no.
interface, requesting it to disconnect from the network. In this 2, pp. 22–31, 2010. [Online]. Available:
way, a make-before-break handover takes place. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber =5450657
Fig. 3 shows the handover delay measured for 30 [3] D.-H. Kim, J.-Y. Gwak, W.-T. Kim, S.-J. Kim, and C.-H. Lee,
simulations, done with different seed values. We can conclude “Design and Implementation of WLAN/HSDPA Interworking
Architecture,” 2008 Third International Conference on
that the handover delay from UMTS to WLAN varies around
Convergence and Hybrid Information Technology, no. Cm, pp.
a mean value of 313 ms, with the lowest value reaching about 741–744, Nov. 2008. [Online].
27 ms, and the highest value situated at 667 ms. Available:http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?
arnumber =4682116
[4] W. Lim, D. Kim, and Y. Suh, “Efficient WLAN Discovery
Schemes Based on IEEE 802.21 MIH Services in
VI. CONCLUSION AND FURTHER WORK Heterogeneous Wireless Networks,” Conference, 2008. IEEE,
2008. [Online]. Available:
This paper proposed an interworking architecture for http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=4698822
mobility between heterogeneous networks based on the IEEE [5] K. Peppas, F. Lazarakis, A. Alexandridis, K. Dangakis, and D.
802.21 (MIH) standard and applied to handover from UMTS Axiotis, “NGL03-4: An Interoperability Mechanism for
to WLAN networks. It defines a framework for providing Seamless Interworking between WLAN and UMTS-HSDPA
handover-related information flows to a Resource Manager in Networks,” in Global Telecommunications Conference, 2006.
GLOBECOM’06. IEEE, 2006, pp. 1–5. [Online]. Available:
charge of taking the handover decision.
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=4150888
Results have shown that the handover delay has relatively [6] W. Song, W. Zhuang, and Y. Cheng, “Load Balancing for
small but different values as they depend also on network Cellular/WLAN Integrated Networks,” IEEE Network, no.
conditions. Further studies have to be carried out. In February, pp. 27–33, 2007
particular, the handover delay between WLAN and UMTS [7] EXata 2.0.1 Programmer’s Guide, Scalable Network
networks needs to be measured and evaluated in different Technologies, Inc., Jun. 2009
network conditions (also the corresponding scenarios [8] A. Vulpe, S. Obreja, O.-E. Barbu, C.-D. Penciu, M. Buga, and
implementation in EXata can be improved) in order to B. Ciobotaru, Qualnet implementation for mobility management
evaluate it in a general manner. Also, different handover in a MIH enabled system,” in 8th International Conference on
Communications (COMM) 2010, Jun. 2010, pp. 523–526.
metrics can be introduced. The framework also needs to be
[9] A. Vulpe, S. Obreja, and O. Fratu, “A study of mobility
enhanced to support more technologies, especially LTE, as it management using IEEE 802.21,” in 9th International
is the leading emerging cellular technology. Symposium on Electronics and Telecommunications (ISETC)
2010, Nov. 2010, pp. 205–208.
[10] ——, “Interoperability procedures between access technologies
ACKNOWLEDGEMENT using IEEE 802.21,” in 2nd International Conference on
Wireless Communication, Vehicular Technology, Information
This research activity was supported by the Ministry of Theory and Aerospace Electronic Systems Technology
Communications and Information Society of Romania under (Wireless VITAE) 2011, Feb./Mar. 2011, pp. 1–5.

96
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Simulation of Rare Events in Teletraffic Systems with

Single Queue
Elena Ivanova 1, Rostislav Raev 2 and Dimitar Radev 3
Abstract – The paper is devoted on computer simulation of rare A basic reference model for rare event simulation is the
event probability of estimation Quality of Service (QoS) parameters. single server queuing system M/M/1/N- FIFO with a finite
Under consideration is a basic reference model with single server buffer size N. The arrival rate is  and the service rate is  .
queuing system M/M/1/N where service disciplines are FIFO and
We are interested in the probability that the buffer content
LIFO. Simulation results for the queuing systems [X]/M/1/N
M/[X]/1/N and [X]/[X]/1/N are shown. reaches a high level k during one busy period (i.e., the time
interval between two successive periods in which the buffer is
Keywords – Rare event, Teletraffic systems, Quality of Service, empty). The maximum occupancy is B=N+1, and the traffic
CLR, .net. load is . The discrete random variable is describe with
stationary complementary distribution function (c.d.f.)
G(x)=1-F(x), for the loss probability PB (1).
I. INTRODUCTION
A. [X]/M/1/N
In modern communications very important is to estimate
such parameters of Quality of Service (QoS) parameter as Cell The investigated reference model for rare event simulation
Loss Ratio and packet loss. The values of these parameters are is the single server queuing system G/M/1/N - FIFO with a
smaller then 1.10-9, and they belong to rare events. finite buffer size N.
Such problems can’t be solved with standard stochastic The inter-arrival times are independent and uniformly
simulation. The aim of this paper is to solve with rare event distributed with arbitrary integral and f A () differential
simulation queuing network model with single queues. distribution laws.
Under consideration is importance sampling for estimation
rare event probability as one of the two basic techniques of β1
rare event simulation - importance sampling and splitting. β0

The program realization is made on the .NET Framework,

which includes a large library and provides language 0 1 n-2 n-1 n n+1

interoperability across several programming languages. The β2

β3
.NET Framework's Base Class Library provides user pn,0 βn
interface, data access, database connectivity, cryptography,
web application development, numeric algorithms, and
Fig. 1. Queuing system G/M/1
network communications. The .NET Framework is intended
to be used by most new applications created for the Windows
platform. The service rate is exponential 1/. We are interested in the
probability that the buffer content reaches a high level k during one
II. REFERENCE QUEUING MODELS busy period (i.e., the time interval between two successive periods in
which the buffer is empty). The maximum occupancy is B=N+1, and
 (i, x) . The discrete random variable is
The investigated queuing models are different. The the traffic load is  1
type of investigated queueing systems are with different 
arrival and service distributions such as discrete and describe with stationary complementary distribution function (c.d.f.)
continues: Pareto, Geometrical, Poisson, Erlang. The basic G(x)=1-F(x) the loss probability PB (1) and the local correlation
queuing models for rare event simulation, investigated here coefficient Cori (2) for the interval i-1≤x<i, i=1,…,k.
are classified in three types [X]/M/1/N, M/[X]/1/N and
[X]/[X]/1/N. The program realization is made for discipline of B. M/[X]/1/N
service FIFO or LIFO with a finite buffer size N. These types
of single queue are necessary, because new The other investigated reference model for rare event
telecommunication networks could be investigated with small simulation is the single server queuing system M/[X]/1/N with
amount of resources. a finite buffer size N.
The service times are exponential distributed, and the
1
service rates are uniformly distributed with arbitrary integral
Elena Ivanova is with Department of Telecommunications, and differential distribution laws.
University of Rousse, Studentska Str 8, 7017, Bulgaria Queuing system М/Erl/1/k is the right opposite Erl/М/1/N,
2
Rostislav Raev is with Department of Telecommunications,
University of Rousse, Studentska Str 8, 7017, Bulgaria
the arrival rate is exponential distributed, and the service rate
3
Dimitar Radev is with Department of Telecommunications,
University of Rousse, Studentska Str 8, 7017, Bulgaria

97
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

is Erlang distributed. The new client arrives in the left side of running Windows. All this is made possible with the .NET
the system, then he is serviced in k-stages with k, the Framework.
function of the arrival rate is defined with (1).
f ( x)  e x x  0 (1)
The service rate is (2).
k (kx) k 1 e  rx
F ( x)  x0
(r  1)!
(2)
We consider the Erlang distribution like a part of
Geometrical distribution.

C. [X]/[X]/1/N

Teletraffic systems with single queue with different

distribution as arrival and service rates, represents client
behavior in different networks like fast Internet network,
optical backbones and multiplex, or even broadband
convergence networks. Here, the model of interest for rare
event simulation is the single server queuing system
Elr/D/1/N with a finite buffer size N, the correlation
(a)
coefficient is smaller than 1,  is non negative =t, the inter-
arrival distribution is ERL(,), with deterministic process of
service D. The arrival rate is determined with (3).
(kt )i
F (t )  1  e kt i0
k 1
(3)
i!
The service rate is defined with k stages of service, CorV
with (4).
1 1
 e kt , k  (4)
2
CorV k X CorV2

   

0 1 2 k

 
 (b)
Fig.3 Program realization: simulation of rare events in
Fig. 2 Queuing system Еrl/D/1/k teletraffic systems

The .NET Framework is a whole lot of Classes (called

Again, the assumptions about erlang and geometric Namespaces) and the technology to get those Classes to work.
distributions are made again. The main component is called the Common Language
Runtime.. With .NET, more than 15 different programming
languages can use the Common Language Runtime. One of
III. PROGRAM REALIZATION these languages is, of course Visual Basic .NET. Another is
C#. They can all use the Common Language Runtime because
The program for Simulation of Rare events in teletraffic of something called the Intermediate Language.
systems with single queue is made with the .NET Framework, One of the most important goal of the program is
because it is the way that programming will be done on a simulation of different disciplines of service.
Microsoft machine from now. And not just on a Microsoft The conceptual model for rare event simulation of
machine. The ADO.NET is used for creating web site, and for single queue system starts with Random Number Generator.
manipulating databases. This framework create applications The RNGs are produced different lend of distributions, such
for mobile phones and PDA's with .NET. There is even a as Uniform, Geometric, Paretto, Poisson and Exponential. The
project in the making that allow to write a programme on a distributions are used as arrival rate and service rate at single
Windows machine that will then work on a computer NOT queue.
The simulation experiments are realized with different
queues, with service discipline First In First Out (FIFO) and

98
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Last In First Out (LIFO). Parameters for blocking probability TABLE II

and packet loss are investigated. With the help of Iterations SIMULATION OF RARE EVENTS IN GEO/M/1/N SYSTEM
Generator are made simulations for each of queues using
different distributions. The received results are shown as RE, Discipline Arrival Service Rare
N Trials
diagrams. % of service rate rate events
The simulation of the three main types of queuing systems LIFO 0,1 5 100000 0
is made of the AMD Athlon(tm) II X2 250 Processor 3.00 5
FIFO 0,1 5 100000 3
GHz and 3 GB RAM, running 32-bits operation system is 20
LIFO 0,1 5 100000 2
Windows 7 Professional. 10
FIFO 0,1 5 100000 4
LIFO 0,1 5 100000 0
5
FIFO 0,1 5 100000 0
30
IV. SIMULATION AND RESULTS LIFO 0,1 5 100000 1
10
FIFO 0,1 5 100000 1
The simulation of the queuing systems were executed for LIFO 0,1 5 100000 0
5
single server queuing system M/M/1/ N, Geo/M/1/N, FIFO 0,1 5 100000 0
40
Pareto/M/1/ N and M/Geo/1/ N FIFO and LIFO as a part of LIFO 0,1 5 100000 1
10
simulation system., error 5% or 10%. The purpose of relative FIFO 0,1 5 100000 2
error is to investigate occurred rare event in different systems.
Consider system M/M/1/N with N=20, 30 and 40 place for TABLE III
service in the queue, discipline of service is FIFO and LIFO, SIMULATION OF RARE EVENTS IN M/GEO/1/N SYSTEM
where the arrival and service rate is exponential and Poisson
distributed. RE, Discipline Arrival Service Rare
N Trials
The simulation of the buffer content is made for 100 000 % of service rate rate events
observation. The increasing of arrival rate leads to LIFO 10 0,9 100000 6
insignificant increasing of rare events. We choose arrival and 5
FIFO 10 0,9 100000 8
service rate, to receive appropriate result. As is seen from 20
LIFO 10 0,9 100000 10
Table I, rare events in system with LIFO are less than the 10
FIFO 10 0,9 100000 16
same system with the same parameters FIFO. LIFO 10 0,9 100000 5
5
FIFO 10 0,9 100000 7
30
LIFO 10 0,9 100000 5
TABLE I 10
FIFO 10 0,9 100000 7
SIMULATION OF RARE EVENTS IN M/M/1/N SYSTEM
LIFO 10 0,9 100000 1
5
RE, Discipline Arrival Service Rare FIFO 10 0,9 100000 3
40
N Trials LIFO 10 0,9 100000 6
% of service rate rate events 10
FIFO 10 0,9 100000 9
LIFO 15 5 100000 2
5
FIFO 15 5 100000 5 TABLE IV
20
LIFO 15 5 100000 5 SIMULATION OF RARE EVENTS IN GEO/GEO/1/N SYSTEM
10
FIFO 15 5 100000 6
5
LIFO 15 5 100000 0 RE, Discipline Arrival Service Rare
N Trials
30
FIFO 15 5 100000 2 % of service rate rate events
LIFO 15 5 100000 3 LIFO 0,3 0,8 100000 0
10 5
FIFO 15 5 100000 4 FIFO 0,3 0,8 100000 1
LIFO 15 5 100000 1 20
5 LIFO 0,3 0,8 100000 1
FIFO 15 5 100000 2 10
40 FIFO 0,3 0,8 100000 1
LIFO 15 5 100000 3 LIFO 0,3 0,8 100000 1
10 5
FIFO 15 5 100000 3 FIFO 0,3 0,8 100000 1
30
LIFO 0,3 0,8 100000 0
The system Geo/M/1/N with p=0,1 is shown on Table II . 10
FIFO 0,3 0,8 100000 0
The simulation was used to receive representative and stable LIFO 0,3 0,8 100000 0
results with sequences of 100 000 observations. 5
FIFO 0,3 0,8 100000 0
40
LIFO 0,3 0,8 100000 1
10
FIFO 0,3 0,8 100000 1

The same simulation is made for M/Geo/1/N queuing

systems with exponential characteristics of arrival rate and
geometric service rate, the received results for buffer size 20,
30 and 40, are presented on Table III. The prescribed error is

99
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

in the range of 5% and 10%. The number of simulation splitting, IEEE Transactions on Automatic Control AC-43
samples is n=100 000. (12),1998, p.1666-1679.
The same simulation is made for Geo/Geo/1/N queuing [12] Görg , C. Schreiber, F, The RESTART/LRE method for rare
systems with geometric characteristics of arrival rate and event simulation, Proceedings of the 28th conference on Winter
simulation, p.390-397, December 08-11, 1996, Coronado,
geometric service rate, the received results for buffer size 20, California, United States.
30 and 40, are presented on Table IV. The prescribed error is [13] Görg, C. Schreiber, F. The RESTART/LRE method for rare
in the range of 5% and 10%. The number of simulation event simulation. Winter Simulation Conference, California,
samples is n=100 000. The main difference here is the absence USA, 1999, pp. 390-397.
of rare events for 100 000 trials, for some of the chosen
parameters.

V. CONCLUSIONS AND FUTURE WORK

Simulation of rare events in different teletraffic systems is

the target of program realization. The number of rare events
defines the simulation probability of blocking different
network systems.
The simulation results for the basic reference model, single
server queuing system M/M/1/N with FIFO and LIFOservice
disciplines, M/[X]/1/N, [X]/M/1/N, and [X]/[X]/1/N systems
with a finite buffer size N, are very good. The results for the
M/[X]/1/N, [X]/M/1/N and [X]/[X]/1/N systems shows that
the prescribed error is in the range of 5% and 10%, where the
number of simulation samples is n=100 000.

REFERENCES
[1] Awan I.U., D.D.Kouvatsos, Arbitrary queueing network models
with service priorities and blocking, Proc. Of 12th UK
Workshop of Perf. Eng. Of Computer and Telecommunication
Systems, ed. D.D. Kouvatsos, ilkley, UK, 1997.
[2] Boer. P. Analysis and Efficient Simulation of Queuing models
of Telecommunication Systems. PhD thesis, Univ. of Twente,
2000.
[3] Bucklew, J. An Introduction to Rare Event Simulation. Springer
Series in Statistics, XI. Springer-Verlag, Berlin, 2004.
[4] Cerou, F., Guyader, A. Adaptive multilevel splitting for rare
event analysis. Rapport de Recherché 5710, INRIA, October
2005, pp.32-40.
[5] L’Ecuyer P., Demers V., Tuffin B., Splitting for rare-event
simulation, Proceedings of the 37th conerence on Winter
simulation, Monterey, California, 2006.
[6] L’Ecuyer P., Demers V., Tuffin B., Rare events, splitting, and
quasi-Monte Carlo, ACM Transactions on Modeling and
Computer Simulation (TOMACS), v.17 n.2, p.9-es, 2007.
[7] Garvels M., D.P. Kroese. A comparison of RESTART
implementations. In Proceedings of the 1998 Winter Simulation
Conference, Washington, DC, 1998 p. 601–609.
[8] Garvels M. The splitting method in rare event simulation. PhD
thesis, University of Twente, The Netherlands, 2000.
[9] Garvels, M., D. Kroese, A Comparison of RESTART
Implementations, University of Twente, Centre for Telematics
and Information Technology, 2005.
[10] Glasserman P., Heidelberger P., Shahabuddin P., Zajic T.,
Splitting for rare event simulation: analysis of simple cases,
Proceedings of the 28th conference on Winter simulation,
Coronado, California, United States,1996, p.302-308.
[11] Glasserman P., Heidelberger P., Shahabuddin P., Zajic T., A
large deviations perspective on the efficiency of multilevel

100
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

VoIP over a Cognitive Network with Limited Availability

Yakim Mihov and Boris Tsankov
Abstract—This paper investigates the call-level performance of a Arrangement of “cognitive” channels over given spectrum
cognitive radio network for dynamic spectrum access providing has a certain cost due to realization of functions like spectrum
VoIP service to secondary users. The paper proposes a new hole detection; detection of PU call arrival over channel
cognitive network paradigm, where only a part of the licensed occupied by a SU; spectrum handover realization etc. In case
channels are available to the unlicensed users by cognitive
functionality. Numerical results are presented and some
the secondary traffic volume does not need all of the licensed
conclusions are drawn. channels to be served, it looks rationally the secondary traffic
to have access only to some part of the channels of the
Keywords – Call blocking probability, call dropping primary network. Therefore, the “cognitivity” will be arranged
probability, cognitive radio network, limited availability, VoIP. to these channels only. As a consequence, operations such as
scanning, detection of idle/busy channel conditions, spectrum
handover will take place faster as they are performed over a
I. INTRODUCTION limited number of channels. This paper investigates the voice
traffic performance under the circumstances described above.

The term Cognitive radio (CR) was first introduced by J.

Mitola [1]. One of the most popular applications of CR is in II. THE TELETRAFFIC SYSTEM
dynamic spectrum access (DSA) networks, as a means to
mitigate the artificially created scarcity of spectrum resources
caused by the traditional static approach for spectrum The corresponding teletraffic serving system is shown on
regulation. Hierarchical spectrum overlay is a promising Fig. 1. The offered PU traffic is denoted with Ap and the
method for DSA. It allows secondary (unlicensed or cognitive) offered SU traffic is denoted with As. The PU calls have access
users (SUs) to temporarily utilize spectrum resources assigned to all of the N channels of the primary network. The SU calls
to primary (licensed or incumbent) users (PUs) if these have access only to cognitive channels (limited
resources are not currently being used for PU transmission. availability). The channels are not subject to
SUs have to release the occupied resources as soon as PUs cognitive activity, i.e. the secondary CR network is allowed to
start reusing them, i.e. PUs have preemptive priority over SUs. utilize only the predetermined channels of the primary
The cognitive network utilizes opportunistically the available network. The bandwidth of a SU call is assumed to be equal to
unoccupied spectrum of the primary network on a non- the bandwidth of a PU call, i.e. one channel is occupied by one
interference basis. Spectrum handover is an essential function PU or SU call. Perfect spectrum sensing and spectrum
of CR since it enables and facilitates the Quality of Service handover procedures are assumed. The service of PU calls is
(QoS) provisioning of the SUs. independent of the service of SU calls.
A popular and often quoted overview of CR is presented in A slight system modification is proposed on Fig. 2 where
[2]. Due to the great interest in using CR networks for DSA, the primary traffic is first directed to the group of
there are numerous publications in the literature. Spectrum channels devoted to PU calls only. Calls rejected from that
sensing is studied in [3], [4], [5]. Multiuser spectrum selection group of channels represent an overflow traffic directed to the
schemes for spectrum sharing and resolving channel cognitive channels . The cognitive channels form a serving
contention are analyzed in [6]. Spectrum handover is system with PUs and cognitive SUs in accordance with the
investigated in [7], [8], [9]. Various QoS-related issues in CR hierarchical spectrum overlay approach for DSA. PUs have
networks are studied in [10], [11], [12], [13]. The resource preemptive priority over SUs. If a PU starts transmitting on a
allocation problem in a multiuser orthogonal frequency
division multiplexing (OFDM) based CR system concerning
the QoS provisioning for both real-time and non-real-time
applications is investigated in [14]. An overview of the general
methodology for cross-layer design and some cross-layer
optimization schemes and algorithms are presented in [15].
The voice traffic service is of a particular interest. Some
examples are [16], [17], [18], [19], [20], [21], [22].

1
Yakim Mihov is with the Faculty of Telecommun. at TU-Sofia, 8
Kl. Ohridski Blvd, Sofia 1000, Bulgaria, yakim_mihov@abv.bg.
2
Boris Tsankov is with the Faculty of Telecommun. at TU-Sofia, 8
Kl. Ohridski Blvd, Sofia 1000, Bulgaria. bpt@tu-sofia. bg
Fig. 1. Illustration of the teletraffic serving system.
This research was partly supported by the Bulgarian Ministry of
Education and Science under Grants DVU02/135/2008 and
DDVU02/13/17.12.2010.

101
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

simulation experiments and some insightful conclusions are

drawn.
A simulation model has been developed which takes into
account all the essential factors necessary for performance
evaluation of the described teletraffic system (Figs. 1 and 2),
such as the Poisson PU and SU call arrival flows, the random
service time of a PU or SU call with negative exponential
distribution, the preemptive priority of PU calls over SU calls,
and the application of limited channel availability for the CR
network. Moreover, the proposed system modification (see
Fig. 2) has also been implemented as an option in the
simulation model.
We first analyze the effect of the number of available
channels to the secondary CR network (i.e. the number of
cognitive channels) Nc on the cognitive traffic capacity when
some predefined level of SU QoS provisioning (in terms of SU
call blocking probability Bs and SU call dropping probability
Fig. 2. Illustration of the proposed modified teletraffic serving system. Bd) has to be guaranteed. As Nc decreases, the traffic capacity
of the CR network decreases as well (see Fig. 3). Therefore,
cognitive channel which is occupied by a SU call, the the limitation of the availability of the PU channels for DSA
cognitive channel has to be vacated immediately. In this case, comes at the price of reduced cognitive traffic capacity, which
the SU performs spectrum handover to another idle cognitive is undesirable if the offered SU traffic that has to be served by
channel in order to ensure successful call service completion. the CR network is relatively large. However, when the offered
If there are no idle cognitive channels, the SU call is dropped. SU traffic is relatively small, a reasonable decrease in the
capacity of the CR network due to limited channel availability
would not degrade the service of SU calls. The limitation of
III. PERFORMANCE ANALYSIS the cognitive channels, i.e. the use of a predefined subset of PU
channels for DSA, may be desirable since the procedures and
operations for supporting the “cognitivity” of the secondary
The offered PU and SU traffic is modeled by two Poisson network do not have to be performed on all of the channels of
random processes with arrival rates λp and λs, respectively. The the primary network, i.e. the cognitive processing load
PU and SU call durations follow a negative exponential (including procedures such as spectrum sensing, spectrum
distribution with mean 1 . Because of the limited availability analysis, spectrum handover, etc.) can be reduced significantly,
which is especially favorable in a resource-constrained
of channels to the secondary CR network, it is impossible to
cognitive environment.
apply the traditional method of building a 2-D Markov chain
[23] or to find exact and simple closed-form solution of the Next, we analyze the effect of the offered PU traffic Ap on
corresponding steady-state equations and derive important QoS the traffic capacity of the secondary CR network when given
characteristics, such as the SU call blocking probability and
the SU call dropping probability . Bs <=1.5%; Bd <=0.1%; Ap =12 Erl; N =35; (Bp =3.5122e-008)
10
In the more practical and efficient arrangement proposed
on Fig. 2, there is an overflowing traffic , which is not a 9
Poisson traffic at all. Because of the preemptive priority of the
8
PU calls over the SU calls, it is impossible to apply the well
Allowable SU traffic - A s (Erl)

known equivalent random theory [24] used for overflow 7

traffic.
6
There is not any difficulty to obtain the PU call blocking
probability as the service of PU calls is affected neither by 5
the secondary traffic, nor by the cognitive functionality.
4
However, the application of limited availability influences the
service of the secondary traffic. This is investigated in the 3
paper by simulations.
2 The serving system with limited availability (Fig. 1)
The modified serving system with limited availability (Fig. 2)
1
IV. SIMULATION RESULTS 5 10 15 20 25 30
Number of cognitive channels - Nc

In this section, the performance of the secondary CR Fig. 3. Cognitive traffic capacity versus the number of channels
network with limited channel availability is analyzed by available to the secondary CR network.

102
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Bs <=1.5%; Bd <=0.1%; N =35; Nc =20; (Bp <=6.8593e-004) errors is considerably reduced, since some channels of the
14 primary network are never occupied by SU calls.
The suggested modification of the investigated serving
12 system with limited availability leads to considerable
performance enhancement of the secondary CR network in
Allowable SU traffic - A s, Erl

10 terms of increased cognitive traffic capacity and reduced

probability for spectrum handover of ongoing SU calls.
8 For future research work, the authors plan to develop
algorithms for determining the optimal number of primary
channels available for DSA based on different design criteria.
6

4
The serving system with limited availability (Fig. 1)
The modified serving system with limited availability (Fig. 2)
2 REFERENCES
5 10 15 20
Offered PU traffic - Ap , Erl
[1] J. Mitola III and G. Q. Maguire, “Cognitive Radio: Making
Software Radios More Personal”, IEEE Personal
Fig. 4. Cognitive traffic capacity versus the offered PU traffic. Communications, vol. 6, no. 4, pp. 13-18, August 1999.
[2] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “A
Survey on Spectrum Management in Cognitive Radio
SU call blocking probability Bs and SU call dropping Networks”, IEEE Communications Magazine, vol. 46, no. 4, pp.
40-48, April 2008.
probability Bd have to be maintained. As Ap increases, the
[3] W.-Y. Lee and I. F. Akyildiz, “Optimal Spectrum Sensing
maximum allowable offered SU traffic As decreases (see
Framework for Cognitive Radio Networks”, IEEE Transactions
Fig. 4). Consequently, the traffic capacity of the CR network on Wireless Communications, vol. 7, no.10, pp. 3845-3857,
depends on the PU traffic load and the use of CR for DSA is 28 October 2008.
rational in primary networks with sufficiently underutilized [4] A. A. El-Saleh, M. Ismail and M. A. M. Ali, “Optimizing
transmission resources. Spectrum Sensing Parameters for Local and Cooperative
Finally, we analyze the effect of applying the proposed Cognitive Radios”, 11th International Conference on Advanced
modification of the serving system (see Fig. 2) on the Communication Technology (ICACT), pp. 1810-1815,
Gangwon-Do, South Korea, 15-18 Feb. 2009.
performance of the secondary CR network. Figs. 3 and 4 show
[5] A. Ghasemi and E. S. Sousa, “Collaborative Spectrum Sensing
that the modified serving system provides greater cognitive for Opportunistic Access in Fading Environments”, 1st IEEE
traffic capacity. Because of the suggested modification of the International Symposium on New Frontiers in Dynamic
serving system, the channels (that are not used Spectrum Access Networks (DySPAN), pp. 131-136,
for DSA) are more likely to be occupied by PU calls than the Baltimore, MD, USA, 8-11 Nov. 2005.
Nc cognitive channels, which leads to a reduction in the SU call [6] C.-W. Wang, L.-C. Wang and F. Adachi, “Modeling and
dropping probability Bd. Similarly, the probability for spectrum Analysis of Multi-User Spectrum Selection Schemes in
handover of ongoing SU calls decreases as well, which Cognitive Radio Networks”, The 20th IEEE International
facilitates the QoS provisioning in the CR network, since under Symposium on Personal, Indoor and Mobile Radio
certain circumstances spectrum handover could cause Communications (PIMRC), pp. 828-832, Tokyo, Japan,
intolerable transmission delay. Therefore, the application of the September 2009.
proposed slight modification to the serving system considered [7] L.-C. Wang and C. Anderson, “On the Performance of
herein improves considerably the performance of the Spectrum Handoff for Link Maintenance in Cognitive Radio”,
secondary CR network. The 3rd International Symposium on Wireless Pervasive
Computing (ISWPC), pp. 670-674, Santorini, Greece, May
2008.
[8] O. Jo, H. H. Choi and D.-H. Cho, “Seamless Spectrum
V. CONCLUSION AND FUTURE WORK Handover Improving Cell Outage in Cognitive Radio Systems”,
4th International Conference on Cognitive Radio Oriented
Wireless Networks and Communications (CROWNCOM),
In this paper, the call-level performance of a VoIP pp. 1-6, Hannover, Germany, June 2009.
secondary CR network operating over a VoIP primary network [9] L. Giupponi and A. I. Perez-Neira, “Fuzzy-based Spectrum
in accordance with the hierarchical spectrum overlay approach Handoff in Cognitive Radio Networks”, The 3rd International
for DSA is investigated in the specific case when DSA is Conference on Cognitive Radio Oriented Wireless Networks
performed only on a predefined subset of primary channels. and Communications (CROWNCOM), pp.1-6, Singapore, May
The simulation results presented in this paper prove the 2008.
feasibility of the proposed new paradigm for DSA with limited [10] L.-C. Wang and C.-W. Wang, “Spectrum Management
availability. The main advantage of DSA with limited Techniques with QoS Provisioning in Cognitive Radio
availability is that the cognitive processing load can be Networks”, The 5th IEEE International Symposium on Wireless
significantly reduced. Moreover, the risk of causing intolerable Pervasive Computing (ISWPC), pp. 116-121, Modena, Italy,
interference to the primary network due to spectrum sensing May 2010.

103
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

[11] Y. Mihov and B. Tsankov, “QoS Provisioning via Channel

Reservation in Cognitive Radio Networks”, in Proc. IEEE
COMCAS, pp. 1-5, Tel Aviv, Israel, November 2011.
[12] T. Jin, C. Chigan and Z. Tian, “Game-theoretic Distributed
Spectrum Sharing for Wireless Cognitive Networks with
Heterogeneous QoS”, IEEE Global Telecommunications
Conference (GLOBECOM ’06), pp. 1-6, San Francisco, CA,
USA, December 2006.
[13] B. Ishibashi, N. Bouabdallah and R. Boutaba, “QoS
Performance Analysis of Cognitive Radio-based Virtual
Wireless Networks”, IEEE INFOCOM 2008, The 27th
Conference on Computer Communications, pp. 2423-2427,
April 2008.
[14] Y. Zhang and C. Leung, “Cross-Layer Resource Allocation For
Mixed Services in Multiuser OFDM-based Cognitive Radio
Systems”, IEEE Transactions on Vehicular Technology, vol. 58,
no. 8, pp. 4605-4619, October 2009.
[15] Q. Xin and J. Xiang, “Joint QoS-aware Admission Control,
Channel Assignment, and Power Allocation for Cognitive Radio
Cellular Networks”, IEEE 6th International Conference on
Mobile Adhoc and Sensor Systems (MASS), pp. 294-303,
Macau, 12-15 Oct. 2009.
[16] H. Lee and D.-H. Cho, “VoIP Capacity Analysis in Cognitive
Radio System”, IEEE Communications Letters, vol. 13, no. 4,
pp. 393-395, June 2009.
[17] H. Lee and D.-H. Cho, “Capacity Improvement and Analysis of
VoIP Service in a Cognitive Radio System”, IEEE Transactions
on Vehicular Technology, vol. 59, no. 4, pp. 1646-1651., May
2010.
[18] P. Wang, D. Niyato and H. Jang, “Voice Service Support over
Cognitive Radio Networks,” Proc. IEEE ICC, pp. 1-5, Dresden,
Germany, June 2009.
[19] P. Wang, D. Niyato and H. Jang, “Voice-Service Capacity
Analysis in Cognitive Radio Networks”, IEEE Trans. on
Vehicular Techn., vol. 59, no. 4, pp. 1779-1790, May 2010.
[20] Gunawardena, S. and W. Zhuang, “Voice Capacity of Cognitive
Radio Networks,” IEEE ICC, pp. 1-5, Cape Town, South
Africa, May 2010.
[21] K. Ali and W. Zhuang, “Link-layer Resource Allocation for
Voice Users in Cognitive Radio Networks”, in Proc. IEEE ICC,
pp. 1-5, Kyoto, Japan, June 2011.
[22] Z. Wang, T. Jiang, L. Jiang, and X. He, “VoIP Capacity
Analysis in Cognitive Radio System with Single/Multiple
Channels”, 6th International Conference on Wireless
Communications, Networking and Mobile Computing
(WiCOM), pp. 1-4, Chengdu, China, September 2010.
[23] Y. Mihov and B. Tsankov, “Cognitive System with VoIP
Secondary Users over VoIP Primary Users”, COGNITIVE: The
Third International Conference on Advanced Cognitive
Technologies and Applications, pp.30-35, Rome, 2011.
[24] R. I. Wilkinson. “Theories for toll traffic engineering in the
USA”, Bell System Technical Journal, 1956, vol.35, no.2,
pp.421-514.

104
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

BEP Performance of DE-QPSK and DE-OQPSK over

composite fading channels in the presence of imperfect
signal extraction
Milica I. Petković1, Bojana Nikolić2, Bata V. Vasić3 and Goran T. Đorđević4
Abstract –In this paper we consider partially coherent dimensional signal constellations, ambiguities about the exact
detection of differentially encoded quadrature phase-shift keying phase orientation of the received signal set exist at the
(DE-QPSK) and differentially encoded offset quadrature phase- receiver. In PSK systems, this ambiguity is resolved using
shift keying (DE-OQPSK) transmitted over the gamma- differential encoding (DE). Differentially encoded QPSK
shadowed Nakagami-m fading channel. The imperfect carrier
(DE-QPSK) and differentially encoded OQPSK (DE-OQPSK)
signal recovery has been taken into account through the phase
error that occurs in the phase-locked loop. The phase error is were introduced by researches as a mean of resolving the
described by Tikhonov distribution. The expressions for bit- carrier phase ambiguity of the data before transmission. One
error probability (BEP) are analyzed and numerical results are of the methods for detecting DE signals is coherent detection
presented. [8]. The phase-locked loop (PLL) is used for carrier signal
recovery. The phase error is a difference between the
Keywords – DE-QPSK, DE-OQPSK, BEP, phase noise, phase incoming signal phase and the recovered carrier signal phase
locked loop. in PLL. It is a statistical process described by Tikhonov
distribution. When the receiver is not ideal, a certain phase
I. INTRODUCTION error appears.
The expressions for bit-error probability (BEP) of DE-
In wireless communication one of the main problem is QPSK and DE-OQPSK for the ideal carrier-synchronization
fading, which is described by several statistical models as over additive white Gaussian noise (AWGN) channel was
Rayleigh, Rician or Nakagami model. Nakagami model is shown and compared in [9]. After the ideal case, the
more general than Rayleigh and Rician. Because of that, it is conditional BEPs of both modulations are given and
very often used in observations. The basis in all these fading combined with the different statistics of the phase error in the
models is the assumption that the average signal power maximum a posteriori (MAP) estimation carrier-
measured is constant. However, the existence of multiple synchronization loops to obtain the expressions of average
scattering may lead to the case where the received average BEPs for both modulations [9].
power becomes random. This phenomenon is called In this paper, we consider DE-QPSK and DE-OQPSK
shadowing. At first the shadowing was modeled using signals transmission over the gamma-shadowed Nakagami-m
lognormal distribution. However, it was often inconvenient fading channel. The expressions for BEP of DE-QPSK and
for further analyses because the obtained composite DE-OQPSK, while the imperfect carrier signal recovery has
probability density function (PDF) is in integral form. Newly, been taken into account through the phase error, are analyzed
gamma distribution was proposed because it is mathematically and numerical results are presented.
more corresponding model. Since fading and shadowing occur
simultaneously in wireless systems, it is necessary to have II. CHANNEL MODEL
models that can describe the faded-shadowed channel [1]-[7].
In this paper, we considered the composite signal modeled by As mentioned, we consider transmission of the signal over
gamma-shadowing Nakagami-m fading distribution. the channel affected by gamma-shadowed Nakagami-m
Quadrature phase-shift keying (QPSK) and offset fading.
quadrature phase-shift keying (OQPSK) are widely used Let the received signal envelope r has Nakagami
modulation techniques. Because of the symmetry in two distribution given by [1]
m
1 − r2
Milica I. Petković is with the Faculty of Electronic Engineering 2m m r 2 m−1e Ω
at the University of Nis, Aleksandra Medvedeva 14, 18000 Nis, p r / Ω ( r / Ω) = , r>0, (1)
Γ ( m) Ω m
Serbia, E-mail: milicapetkovic86@gmail.com.
2
Bojana Nikolić is with the Faculty of Electronic Engineering at
the University of Nis, Aleksandra Medvedeva 14, 18000 Nis, Serbia, where m is the Nakagami parameter, Ω is the average power
E-mail: bojana.nikolic@elfak.ni.ac.rs. Ω = E[r2] with E denoting mathematical expectation and Γ(.)
3 is the gamma function. The m parameter refers to the fading
Bata V. Vasić is with the Faculty of Electronic Engineering at
the University of Nis, Aleksandra Medvedeva 14, 18000 Nis, Serbia, severity. With lower values of m, the fading is stronger. In the
E-mail: bata.vasic@silicon-studio.com. case m=1, we have Rayleigh fading, and m=∞ is the no-fading
4
Goran T. Đorđević is with the Faculty of Electronic Engineering case.
at the University of Nis, Aleksandra Medvedeva 14, 18000 Nis, In the case when the shadowing is present, Ω is random
Serbia, E-mail: goran@elfak.ni.ac.rs.
variable and has gamma distribution given by [1]

105
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

−
ms
Ω synchronization, in the channel under the influence of fading
ms ms Ω ms −1e Ωs is (similarly as in [9]):
pΩ (Ω) = , Ω>0, (2)
Γ ( m s ) Ω s ms
1  ρ 
Pb (φc ; ρ ) | DE − QPSK = erfc (cos φc − sin φc ) 
where Ωs = E[ x ] is the gamma shadow area mean power. The 2  2 
 
parameter ms refers to the shadowing severity. With lower
values of ms, the shadowing influence is stronger. In the case  1  ρ 
ms=∞, shadowing is not exist. × 1 − erfc (cos φc − sin φc ) 
 2  2 
The composite envelope r of the gamma-shadowed  
Nakagami-m faded signal is:
1  ρ 
∞ + erfc (cos φc + sin φc ) 
 2 
p(r ) = ∫p
0
r / Ω (r / Ω) pΩ (Ω)dΩ . (3) 2  

 1  ρ 
Substituting (1) and (2) in (3), we have: × 1 − erfc (cos φ c + sin φc )  (7)
 2  2 
 
m + ms
4  mms  2
p(r ) =   where ρ represents SNR per symbol.
Γ(m)Γ(ms )  Ω s  In order to evaluate the average BEP for DE-QPSK, the
PDF of phase error should be known. The conditional BEP
 mms  must be averaged over that PDF.
× r m + ms −1 K ms − m  2r (4) The appropriate PDF of loop’s phase error in the form of
 Ω s 
 the Tikhonov distribution for DE-QPSK is given by [9]:
where Kν(.) is the modified Bessel function of the second kind 4 exp( ρ eq ( ρ ) cos 4φc )
pφ c (φc ; ρ ) QPSK = ,
and order ν and Ω s = E[r 2 ] = r2
is the average power. 2πI 0 ( ρ eq ( ρ ))
The instantaneous SNR per symbol, ρ, and the average
SNR per symbol, ρ0, are related by: ρ PLL ( ρ ) S L (ρ )
QPSK π π
ρ eq ( ρ ) = , - ≤ φc ≤ , (8)
µ2 ρ 16 4 4
= , µ > 0, ρ > 0 . (5)
µ 2 ρ 0 where ρ eq represents an equivalent loop’s SNR, ρ PLL is
The distribution of the SNR in gamma-shadowed Nakagami- loop’s SNR that can be expressed as
m fading channel can be obtained using (4) and (5) by ρPLL(ρ) = P / N0 BL = (Eb / N0 ) /(BLTb ) = (ρ / 2) /(BLTb ) (BL denotes
applying standard technique of transforming random the one-sided loop bandwidth). Degradation term referred to
variables: as “squaring loss”, SL is given by [9]:
m + ms
2  mms  2 S L |QPSK ( ρ ) =
p( ρ ) =  
Γ(m)Γ(ms )  ρ 0  2
  ρ 
erf   − 2 ρ exp − ρ 
m + ms − 2  mms    2  2π  2  . (9)
×ρ 2 K 
ms − m  2 ρ. (6) =
  ρ  ρ 1
2
 ρ 
 ρ0 
1 + ρ − 2 erf  + exp − 
   2 
Remark that ρ is the instantaneous SNR per symbol, and ρ0  2  2  π
is the average SNR per symbol. The average SNR per bit is
ρ 0b = ρ 0 / log 2 M = ρ 0 / 2 in the case of quadrature To obtain average BEP, it is necessary to average
conditional BEP (7) over PDF of phase error (8). Under the
modulation formats. influence of fading, the instantaneous SNR is random
variable. It is also required to do averaging of (7) over PDF of
III. AVERAGE BEP PERFORMANCE IN THE PRESENCE OF instantaneous SNR per symbol (6). The BEP of DE-QPSK is:
CARRIER PHASE ERROR OVER THE GAMMA-SHADOWED
∞ π /4
NAKAGAMI FADING CHANNEL
Pb | DE − QPSK = ∫ ∫ P (φ ; ρ ) b
ρ = 0 φ c = −π / 4
c DE − QPSK

A. DE-QPSK

For DE-QPSK, the expression for conditional BEP in the × pφ c (φc ; ρ ) QPSK p ( ρ )dφc dρ . (10)
presence of a phase error φ c , due to imperfect carrier

106
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

B. DE-OQPSK ∞ π /2

For DE-QPSK, the expression for conditional BEP in the

Pb | DE − OQPSK = ∫ ∫ P (φ ; ρ ) b
ρ = 0 φ c = −π / 2
c DE − OQPSK

presence of a phase error φc , due to imperfect carrier

synchronization, in the channel under the influence of fading × pφ c (φc ; ρ ) OQPSK p( ρ )dφc dρ . (14)
is (similarly as in [9]):
IV. NUMERICAL RESULTS
1  ρ 
Pb (φc ; ρ ) | DE − OQPSK ==  erfc cos φc 
  Fig.1. shows DE-QPSK BEP dependence on SNR per bit
 2  2 
for different values of BLTb. The system has better
performance for lower value of BLTb. In Fig.1. the best
1  ρ 
+ erfc (cos φc − sin φc )  performance is for the case when BLTb has the lowest value
4  
 2  (BLTb=0.01) and the parameters m and ms are higher. It is
noticed that influence of BLTb is more expressed for higher
 ρ  values of m and ms, i.e. with the lower influence of fading
1
+ erfc (cos φc + sin φc )  and shadowing.
4  
 2 
0
10

 1  ρ 
× 1 − erfc cos φc 
 4  2  10
-1

 

 ρ 
-2
10
1
− erfc (cos φc − sin φc ) 
Pb

8  2 
  -3 m=1, ms=3, BLTb=0.1
10
m=1, ms=3, BLTb=0.05

1  ρ  m=1, ms=3, BLTb=0.01

− erfc (cos φc + sin φc )  . (11) m=3, ms=6, BLTb=0.1

 
-4
10 m=3, ms=6, BLTb=0.05
8  2  m=3, ms=6, BLTb=0.01
DE-QPSK
-5
The appropriate PDF of loop’s phase error in the form of 10
0 5 10 15 20 25 30
the Tikhonov distribution for DE-OQPSK is given by [9]:
ρ0b (dB)

2 exp( ρ eq ( ρ ) cos 2φc ) Fig. 1. DE-QPSK BEP dependence on SNR per bit for different
pφ c (φc ; ρ ) |OQPSK = , values of BLTb
2πI 0 ( ρ eq ( ρ ))
The same BEP dependence for DE-OQPSK is shown in
ρ PLL ( ρ ) S L OQPSK (ρ ) π π Fig.2. The conclusion is the same: System has better
ρ eq ( ρ ) = , - ≤ φc ≤ . (12) performance for lower value of BLTb.
4 2 2
m=1, m s=3, B LT b=0.1
“Squaring loss”, SL, for DE-OQPSK is given by [9]: 10
0

m=1, m s=3, B LT b=0.05

m=1, m s=3, B LT b=0.01
S L |OQPSK ( ρ ) = m=3, m s=6, B LT b=0.1
-1
2 10 m=3, m s=6, B LT b=0.05
  ρ 
erf   − ρ exp − ρ  m=3, m s=6, B LT b=0.01

  2  2π  2  . (13) DE-OQPSK

Pb

= 2
10
-2

ρ  ρ  ρ  1  ρ 
1+ −  erf + exp − 
2  2  2  π  2  -3
10

Similar to DE-QPSK, it is necessary to average conditional

BEP (11) over PDF of phase error (12) to obtain average BEP 10
-4

for DE-OQPSK. Under the influence of fading, the 0 5 10 15 20 25 30

instantaneous SNR is random variable, and it is also required ρ 0 b (dB)

to do averaging of (11) over PDF of instantaneous SNR per Fig. 2. DE-OQPSK BEP dependence on SNR per bit for different
symbol (6). The BEP of DE-QPSK is: values of BLTb

Fig.3. shows DE-QPSK BEP dependence on SNR per bit,

for different values of the shadowing parameter ms. When the

107
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA
0
10 value of fading parameter m, we have severe fading.
The DE-QPSK and DE-OQPSK BEP dependences on SNR
per bit are shown in Fig.5. We have greater difference
10
-1
between DE-QPSK and DE-OQPSK when the parameters are
higher, i.e. when the impacts of fading and shadowing are
ms=0.5
lower. When the impact of fading and shadowing (m=3 and
ms=1
ms=5) is lower, the DE-QPSK and DE-OQPSK BEPs have
Pb

-2
10
ms=2
same value when SNR per bit is 4.5dB. For lower values of
ms=3
ms=4
SNR, system has better performance for DE-QPSK. When the
10
-3
m=1
SNR per bit is higher than 4.5dB, the system has better
BLTb=0.05 performance for DE-OQPSK.
DE-QPSK

10
-4 V. CONCLUSION
0 5 10 15 20 25 30
ρ0b (dB) In this paper, we have derived the expressions for BEP of
DE-QPSK and DE-OQPSK when the phase error introduced
Fig. 3. DE-QPSK BEP dependence on SNR per bit for different by loop’s finite SNR is taken into account and the signal is
values of ms
transmitted over the gamma-shadowed Nakagami-m fading
values of the parameter ms is lower, the influence of channel. The effects of the parameter BLTb and the fading and
shadowing is bigger and we have worse system performance. shadowing parameters on the BEP have been noted.
Fig.4. shows DE-OQPSK BEP dependence on SNR per bit,
for different values of the parameter m. The performance of ACKNOWLEDGEMENT
the system is the worst when the m=0.5. With decreasing
0
This paper was supported in part by the Ministry of Science
10
of Republic of Serbia under grant TR-32028 and in part by the
Norwegian Ministry of Foreign Affairs within the project
Norwegian, Bosnian and Serbian cooperation platform for
-1
10 ministry and industry in ICT R&D.

m=0.5 REFERENCES
Pb

-2
10 m=1
m=2 [1] I. M. Kostic, “Analytical approach to performance analysis for
m=3
m=4
channel subject to shadowing and fading”, IEE Proceedings
10
-3
ms=3 Communications, vol. 152, no. 6, pp. 821-827, December 2005.
BLTb=0.05 [2] P. S. Bithas, N. C. Sagias, P. T. Mathiopoulos, G. K.
DE-OQPSK
Karagiannidis, A. A. Rontogiannis, “On the performance
-4
analysis of digital communications over generalized-K fading
10
0 5 10 15 20 25 30
channels”, IEEE Communications Letters, vol. 10, no. 5, pp.
353–355, May 2006
ρ0b (dB)
[3] P. M. Shankar, “Error Rates in Generalized Shadowed Fading
Fig. 4. DE-OQPSK BEP dependence on SNR per bit for different Channels”, Wireless Personal Communications, vol. 28, no.3,
values of m pp. 233-238, February 2004.
10
0 [4] V. Milenkovic, N. Sekulovic, M. Stefanovic, M. Petrovic,
“Effect of microdiversity and macrodiversity on average bit
error probability in gamma shadowed Rician fading channels”,
10
-1
ETRI Journal, vol. 32, no. 3, pp. 464-467, Jun. 2010.
[5] N. Sekulovic, M. Stefanovic, “Performance analysis of system
with micro- and macrodiversity reception in correlated gamma
-2
10 shadowed Rician fading channels”, Wireless Personal
Communications, accepted for publication (published online
Pb

-3
12. Feb. 2011.), DOI: 10.1007/s11277-011-0232-8.
10
m=1, m s=1 DE-QPSK [6] A. Goldsmith, Wireless Communications, New York,
m=1, m s=1 DE-OQPSK Cambridge University Press, 2005.
-4 m=3, m s=5 DE-QPSK [7] M. K. Simon, M. S. Alouini, Digital Communication over
10
m=3, m s=5 DE-OQPSK Fading Channels: A Unified Approach to Performance
B lT B=0.1 Analysis, New York, John Wiley & Sons, Inc., 2000.
10
-5 [8] J. G. Proakis, Digital Communications, 4th ed., New York:
0 5 10 15 20 25 30 McGraw-Hill, Inc., 2001.
ρ0 b (dB) [9] M. K. Simon, “On the Bit-Error Probability of Differentially
Encoded QPSK and Offset QPSK in the Presence of Carrier
Fig. 5. DE-OQPSK and DE-OQPSK BEP dependence on SNR per Synchronization”, IEEE Transaction on Communications, vol.
bit 54, no. 5, May 2006.

108
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Quality of Service (QoS) – main principles and

managing tools
Miroslav Slavov1 and Pencho Penchev2
Abstract – In this paper the basic concepts behind Quality of available before you can successfully implement QoS. The
Service (QoS), the need for it will be discussed, and several of the transmission quality of the network is determined by the
types of QoS mechanisms available will be introduced. Quality of following factors:
Service itself is not something that you configure on a Cisco
 Loss – a relative measure of the number of packets that
router, rather it is an overall term that refers to a wide variety of
mechanisms used to influence traffic patterns on a network. QoS were not received compared to the total number of packets
has already proven itself as the enabling technology for the transmitted. Loss is typically a function of availability.
convergence of voice, video and data networks. As business needs  Delay – the finite amount of time it takes a packet to
evolve, so do demands on QoS technologies. reach the receiving endpoint after being transmitted from the
sending endpoint.
Keywords – Quality of Service (QoS), Cisco QoS Toolset, QoS  Delay variation (Jitter) – the difference in the end-to-end
levels. delay between packets [1][4].

B. Why is QoS Important for Enterprise Networks?

I. INTRODUCTION
QoS technologies refer to the set of tools and techniques to
Quality of Service (QoS) is a set of capabilities that allows manage network resources and are considered the key
delivering differentiated services for network traffic, thereby enabling technology for network convergence.
providing better service for selected network traffic. QoS QoS tools are not only useful in protecting desirable traffic,
expedites the handling of mission-critical applications, while but also in providing deferential services to undesirable traffic
sharing network resources with noncritical applications. such as the exponential propagation of worms [4].
QoS also ensures the available bandwidth and minimum The WAN devices can limit the bandwidth available to the
delays required by time-sensitive multimedia and voice traffic, or give the traffic priority, or even change the
applications. This allows using expensive network classification of the traffic. In this way, you can provide end-
connections more efficiently, and to establish service level to-end QoS in your network.
agreements with customers of the network. Figure 1 shows an example of an enterprise network.
QoS features provide better and more predictable network Typically, you classify traffic in the LAN before sending it to
service by: the WAN. The devices on the WAN then use the classification
 Supporting dedicated bandwidth for critical users and to determine the service requirements for the traffic [3].
applications.
 Controlling jitter and latency (required by real-time
traffic).
 Avoiding and managing network congestion.
 Shaping network traffic to smooth the traffic flow.
 Setting traffic priorities across the network [2][3].

II. DEFINING QUALITY OF SERVICE

A. What is QoS?

QoS is the measure of transmission quality and service

availability of a network (or internetworks).
Service availability is a crucial foundation element of QoS. Fig. 1. Example of an Enterprise Network
The network infrastructure must be designed to be highly

C. Which Applications Need QoS?

1
Miroslav Slavov is with the Faculty of Electrical Engineering and
Electronics at Technical University of Gabrovo, 4 H. Dimitar str., Understanding the Characteristics of Applications
Gabrovo 5300, Bulgaria, e-mail: miroslav_slavov@mail.bg. It is important to understand the characteristics of the
2
Pencho Penchev is with the Faculty of Electrical Engineering and applications that need protection. Some applications tend to be
Electronics at Technical University of Gabrovo, 4 H. Dimitar str., sensitive to latency or packet loss, while others are considered
Gabrovo 5300, Bulgaria.

109
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

"aggressive" because they are bursty or consume a lot of  Layer 4 parameters— L4 protocol (TCP/UDP), source/
bandwidth. If the application is bursty, determine if there is a destination ports
constant burst or a small burst. Is the packet size of the  Layer 7 parameters— application signatures via Network
application large or small? Is the application TCP or UDP Based Application Recognition (NBAR)
based [5]? NBAR is a Cisco proprietary technology that identifies
application layer protocols by matching them against a
TABLE I Protocol Description Language Module (PDLM), which is
APPLICATIONS THAT REQUIRED QOS essentially an application signature. The NBAR deep-packet
classification engine examines the data payload of stateless
Characteristic Guideline
protocols against PDLMs. There are over 98 PDLMs
Application that is Do not use weighted random embedded into Cisco IOS software 12.3 code.
delay− or early detection (WRED), traffic Additionally, Cisco IOS software 12.3(4)T introduces the
loss−sensitive. shaping, fragmentation (FRF−12 ability to define custom PDLMs which examine user-defined
(Voice and Real Time (describes the method of strings within packet payloads.
Video) fragmenting Frame Relay frames PDLMs can be added to the system without requiring an IOS
into smaller frames)), or upgrade because they are modular. NBAR is dependent on Cisco
policing. For this kind of traffic, Express Forwarding (CEF) and performs deep-packet
you should implement Low classification only on the first packet of a flow. The remainder of
Latency Queuing (LLQ) and use the packets belonging to the flow is then CEF-switched.
a priority queue for the Within an enterprise, marking is done at either Layer 2 or
delay−sensitive traffic. Layer 3, using the following fields:
Application that is Use WRED, policing, traffic
 802.1Q/p Class of Service (CoS)—Ethernet frames can be
consistently bursty or shaping, or class−based weighted
marked at Layer 2 with their relative importance by setting the
is a bandwidth hog. fair queuing (CBWFQ) to
802.1p User Priority bits of the 802.1Q header. Only three bits
(FTP and HTTP) guarantee bandwidth.
are available for 802.1p marking. Therefore, only 8 classes of
Application that is Use WRED since lost packets service (0-7) can be marked on Layer 2 Ethernet frames.
TCP−based. cause TCP to back off and then
 IP Type of Service (ToS) byte—Layer 2 media often
ramp up again using the
changes as packets traverse from source to destination, so a more
slow−start algorithm. If the
ubiquitous classification occurs at Layer 3. The second byte in an
traffic is UDP−based and does
IPv4 packet is the ToS byte. The first three bits of the ToS byte
not change its behaviour when
are the IPP bits. These first three bits combined with the next
packets are dropped, do not use
three bits are known collectively as the DSCP bits.
WRED. Use Policing if you need
 DSCPs and Per-Hop Behaviors (PHBs)—DSCP values
to rate−limit the application;
can be expressed in numeric form or by special standards-
otherwise just let the packets
based names called Per-Hop Behaviors. There are four broad
tail−drop.
classes of DSCP PHB markings: Best Effort (BE or DSCP 0),
RFC 2474 Class Selectors (CS1–CS7, which are
III. CISCO QOS TOOLSET identical/backwards-compatible to IPP values 1–7), RFC 2597
Assured Forwarding PHBs (AFxy), and RFC 3268 Expedited
This section describes the main categories of the Cisco QoS Forwarding (EF).
toolset and includes the following topics (Figure 2): DSCP values can be expressed in decimal form or with their
 Classification and Marking tools PHB keywords. For example, DSCP EF is synonymous with
 Policing and Markdown tools DSCP 46, and DSCP AF31 is synonymous with DSCP 26.
 Scheduling tools  IP Explicit Congestion Notification (IP ECN)—IP ECN,
 Link-specific tools as defined in RFC 3168, makes use of the last two bits of the
 AutoQoS tools IP ToS byte, which are not used by the 6-bit DSCP markings,
as shown in Figure 3.
 Call Admission Control tools

B. Policing and Markdown Tools

A. Classification and Marking Tools
Policing tools (policers) determine whether packets are
Classification and marking tools set this trust boundary by
conforming to administratively-defined traffic rates and take
examining any of the following:
action accordingly. Such action could include marking,
 Layer 2 parameters—802.1Q Class of Service (CoS) bits,
remarking or dropping a packet.
Multiprotocol Label Switching Experimental Values (MPLS
A basic policer monitors a single rate: traffic equal to or
EXP).
below the defined rate is considered to conform to the rate,
 Layer 3 parameters—IP Precedence (IPP), Differentiated while traffic above the defined rate is considered to exceed the
Services Code Points (DSCP), IP Explicit Congestion rate. On the other hand, the algorithm of a dual-rate policer
Notification (ECN), source/destination IP address (such as described in RFC 2698) is analogous to a traffic light.

110
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 2. The Cisco QoS Toolset

Fig. 3. The IP ToS Byte (DSCP and IP ECN)

mechanisms work best with TCP-based applications because

C. Scheduling Tools selective dropping of packets causes the TCP windowing
mechanisms to “throttle-back” and adjust the rate of flows to
manageable rates.
Scheduling tools determine how a frame/packet exits a
device. Whenever packets enter a device faster than they can The principle IOS congestion avoidance mechanism is
WRED, which randomly drops packets as queues fill to
exit it, such as with speed mismatches, then a point of
congestion, or bottleneck, can occur. Devices have buffers capacity.
that allow for scheduling higher-priority packets to exit sooner
than lower priority ones, which is commonly called queuing. D. Link-Specific Tools
Queuing algorithms are activated only when a device is
experiencing congestion and are deactivated when the Link-specific tools include the following:
congestion clears. Figure 4 shows the Layer 3 and Layer 2  Shaping tools—A shaper typically delays excess traffic
queuing subsystems of the Cisco IOS (Internetwork Operating above an administratively-defined rate using a buffer to hold
System) software LLQ/CBWFQ algorithm. packets and shape the flow when the data rate of the source is
higher than expected.
 Link Fragmentation and Interleaving tools—With slow-
speed WAN circuits, large data packets take an excessively
long time to be placed onto the wire.
 Compression tools—Compression techniques, such as
compressed Real-Time Protocol (cRTP), minimize bandwidth
requirements and are highly useful on slow links. At 40 bytes
total, the header portion of a VoIP packet is relatively large
and can account for nearly two-thirds or the entire VoIP
packet (as in the case of G.729 VoIP).
 Transmit ring (Tx-Ring) tuning—The Tx-Ring is a final
interface First-In-First-Out (FIFO) queue that holds frames to
Fig. 4. The IP ToS Byte (DSCP and IP ECN) be immediately transmitted by the physical interface. The Tx-
Ring ensures that a frame is always available when the
Selective dropping of packets when the queues are filling is interface is ready to transmit traffic, so that link utilization is
referred to as congestion avoidance. Congestion avoidance driven to 100 % of capacity. The size of the Tx-Ring is

111
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

dependent on the hardware, software, Layer 2 media, and

queuing algorithm configured on the interface. E. AutoQoS
Tools
The richness of the Cisco QoS toolset inevitably increases
its deployment complexity. To address customer demand for
simplification of QoS deployment, Cisco has developed the
Automatic QoS (AutoQoS) features. AutoQoS is an intelligent
macro that allows an administrator to enter one or two simple
AutoQoS commands to enable all the appropriate features for
the recommended QoS settings for an application on a specific
interface.
For Campus Catalyst switches, AutoQoS automatically
performs the following tasks: Fig. 5. Cisco QoS Feature, Design Guide and AutoQoS Evolution
 Enforces a trust boundary at Cisco IP Phones. Figure 5 shows the relationship between Cisco QoS
 Enforces a trust boundary on Catalyst switch access ports features, Design Guides, and AutoQoS.
and uplinks/downlinks.
 Enables Catalyst strict priority queuing for voice and
weighted round robin queuing for data traffic. IV. CONCLUSION
 Modifies queue admission criteria (CoS-to-queue
mappings). The purpose of this paper is to describe the terms and the
 Modifies queue sizes as well as queue weights where main concept of Quality of Service (QoS). In the paper a set
required. of tools, used for maintaining the QoS are described. These
 Modifies CoS-to-DSCP and IP Precedence-to-DSCP are the basics, needed for future researches of QoS and
mappings. finding a way to improve it in different types of networks.
For Cisco IOS routers, AutoQoS is supported on Frame
Relay (FR), Asynchronous Transfer Mode (ATM), High- ACKNOWLEDGEMENT
Level Data Link Control (HDLC), Point-to-Point Protocol
(PPP), and FR-to-ATM links. This paper has been sponsored by E 1102 project of
The AutoQoS Enterprise feature consists of two Technical University of Gabrovo.
configuration phases, completed in the following order:
 Auto Discovery (data collection)—Uses NBAR-based
protocol discovery to detect the applications on the network REFERENCES
and performs statistical analysis on the network traffic.
 AutoQoS template generation and installation— [1] M. Flannagan, etc “Administering Cisco QoS in IP Networks”,
Generates templates from the data collected during the Auto Syngress, Rockland, 2001, ISBN: 1-928994-21-0.
[2] T. Szigeti, “End-to-End QoS Network Design”, Cisco Press,
Discovery phase and installs the templates on the interface.
Indianapolis, 2005 ISBN 1-58705-176-1.
These templates are then used as the basis for creating the [3] “User Guide for CiscoWorks QoS Policy Manager” Software
class maps and policy maps for your network. After the class Version 4.1.5, Cisco Systems Inc, 2010.
maps and policy maps are created, they are then installed on [4] “Enterprise QoS Solution Reference Network Design Guide”
the interface. Version 3.3, Cisco Systems Inc, 2008.
[5] “Implementing Quality of Service”, Document ID: 13747, Cisco
Systems Inc, 2008 – 2009.

112
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Pitch perception in complex sound

Marko Janković 1 and Dejan Ćirić 2
Abstract – Sound of an instrument represents a typical example
of a complex sound. Such a sound can be decomposed in a set of
simple (pure) sounds. In a number of cases, an instrument sound
II. SOUND AND PERCEPTION
can be considered to consist of a fundamental and harmonics
that give it a unique characteristic, depending on their number, The simplest possible sound is a pure sound representing a
distribution, amplitude and envelope. The ability of a human sinusoidal signal (sound). When a sound consists of two or
auditory system to perceive pitch in a complex sound is more pure sounds (partials), it represents a complex sound
associated with the harmonics. In this paper, we will show how that can be given by:
distribution of harmonics in complex sound affects the pitch
perception. For that purpose, spectral analysis is performed and s (t ) = ∑ An cos(2πf n t + θ n ) , (1)
listening tests are applied. Special attention is paid to the n≥1
phenomenon known as “missing fundamental”.
where An is the amplitude, fn is the frequency and θn is the
Keywords – complex sound, harmonics, auditory system, pitch phase. The sound with the lowest frequency is called
perception. fundamental, while other sounds whose frequencies are higher
are called overtones. An interesting example of sound is
I. INTRODUCTION obtained when all of the frequencies are integer multiples of a
given (fundamental) frequency f, fn=nf. The sound signal is
Ability that a human auditory system perceives pitch in a then periodic with period T=1/f. The fundamental (caused by
complex sound is of extreme importance for understanding the first mode of vibration) determines the pitch of the
music and speech, and for analysing sound images [1]. Many corresponding complex sound (tone). The frequency fn (sound
instruments generate specific sounds, whose main attribute to at that frequency) is called the n-th harmonic. Accordingly,
create a melody is pitch. In verbal communication, pitch the fundamental is the first harmonic. Perceptually, a sound
carries prosodic information in languages such as English and containing only few harmonics is perceived as “poor”, while a
semantic information in tonal languages such as Mandarin and sound containing a number of harmonics is perceived as
Chinese. On the basis of differences in fundamental “rich”.
frequencies, we can differentiate sounds and order them in Sounds of all music instruments consist of many partials.
appropriate music scales [1]. Some of them are harmonically related to fundamental and
From an acoustical point of view, music instruments some are not [2]. According to Hornbostel-Sachs system, all
represent sources that generate sounds of specific music instruments can be classified into five groups
characteristics including spectral content. Majority of depending on what actually makes the sound [3]. These
instrument sounds consist of a fundamental (the lowest groups are: ideophones, membranophones, chordophones,
partial) and upper partials that can be harmonically related to aerophones and electrophones. In both aerophones and
fundamental or not [2]. Factors affecting perception of such chordophones, there is something long and thin that basically
sounds are related to the number, distribution, amplitude and vibrates in one dimension: a vibrating string or column of air.
envelope of the partials. Very interesting feature that has This produces strong resonances, and sound tends to be pure
attracted attention during the previous decades is pitch with a specific pitch. In membranophones, the membrane can
perception, and especially a phenomenon known as “missing basically vibrates in two dimensions, while in idiophones, the
fundamental”. This has been a motive to investigate the pitch body of the instrument can vibrate in three dimensions. As the
perception in complex sounds and the mentioned phenomenon number of dimensions goes up, the resonances become more
in more detail here. complex and weaker, and therefore, the sounds become more
For the purpose of investigating pitch perception, various complex and with more diffuse pitch [3].
complex sounds are first generated and then analyzed. The Generally, pitch represents an auditory sensation defined in
pitch perception in such complex sounds is examined through various ways. The most common one given by American
the listening tests. The impact of harmonics distribution on the National Standards Institute (ANSI) in 1994 describes pitch as
pitch perception is observed. Emphasis is placed on studying an “attribute according to which the sounds can be arranged
the perceptual consequences of the phenomenon “missing in a musical scale, from lowest to highest” [1]. Pitch is closely
fundamental”. related to frequency, but there is no absolute equivalency
between those two since the frequency is an objective
concept, while pitch is a subjective.
1
Marko Janković is a PhD student at the Faculty of Electronic With modern methods for measuring brain activity (fMRI,
Engineering, University of Niš, Aleksandra Medvedeva 14, 18000 EEG, MEG), it becomes possible to locate the area of auditory
Niš, Serbia, E-mail: marestudio2004@gmail.com cortex that corresponds to pitch perception called Heschl’s
2
Dejan Ćirić is with the Faculty of Electronic Engineering, gyris, and it is positioned in the right cerebral hemisphere
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E- [1,4]. In the auditory cortex, there are two types of neurons:
mail: dejan.ciric@elfak.ni.ac.rs

113
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

multi-peak and single-peak neurons [1,4]. Multi-peak neuron fundamental is 440 Hz, while the frequencies of harmonics
can be excited with multiple frequencies that are harmonically are: 880 Hz, 1320 Hz, 1760 Hz and 2200 Hz. The spectrum of
related. A frequency required to have minimum energy to such a complex sound is given in Fig. 1. The task of the
excite a multi-peak neuron is called the central frequency and subjects was to repeat the pitch of the reproduced sound by
it is equal to frequency of the first harmonic (fundamental). singing neutral syllable La.
The other frequencies are required to have more energy to
make the same excitation [1,4]. Single-peak neuron is excited 100

Relative amplitude (dB)

only by a single frequency [1]. In the auditory cortex, about
20 % of neurons represent multi-peak neurons (30 Hz-5 kHz),
50
whereas 80 % are single-peak neurons (5 kHz-20 kHz) [1,4].
Important frequencies for understanding human speech are
ranged up to about 4 kHz [1]. On the other hand, the highest 0
frequency of a fundamental tone of symphony orchestra has a
piccolo flute, and it is about 4.5 kHz [1]. The tests have shown
that people best understand the melody up to 5 kHz; beyond -50
this frequency understanding of melody is difficult or 1k 10k 20k 100
impossible [1]. Frequency (Hz)
When the first harmonic (fundamental) is excluded from a Fig. 1 Spectrum of generated complex sound used in the listening
test 1
complex sound whose partials are harmonically related, we
still can perceive the pitch equal to the frequency of missing
In the listening test 2, the complex sound used in the
fundamental. This phenomenon is known as “missing
listening test 1, but without fundamental was presented to the
fundamental” [1].
subjects in the same manner as in the test 1. The task of the
subjects was also to repeat the pitch of the reproduced sound
III. METHODS OF INVESTIGATION by singing the neutral syllable La. The purpose of this test is
to study the phenomenon “missing fundamental”.
For the purpose of spectral analysis of various instrument Two complex sounds separated by silence of 1 second were
sounds, the samples from Steinberg’s sounds library called presented to the subjects in the listening test 3. The first sound
Hypersonic II were used. Sample rate of all samples was 44.1 is the same as in the test 1, while the second one is the same
kHz while 16 bits were used for quantization. as in the test 2. The spectrogram of complex sound applied in
Sounds of some instruments belonging to the groups the test 3 is given in Fig. 2. This time, the task of the subjects
chordophones (acoustic guitar, harp, piano and violin), was to answer a question: “Is the difference between two
aerophones (oboa and flute), membranophones (timpani, kick, reproduced sounds in pitch or in timbre”?
snare and tambourine) and ideophones (closed hi-hat, open hi- 4k
hat, crash cymbal, finger snap, rattle and effects of wind) are
spectrally analyzed. Instruments from the group electrophones
Frequency (Hz)

3k
are excluded from the analysis because they generate sounds
identical to those of instruments from other groups.
2k
Since pitch perception is a subjective attribute, it is
investigated here through the listening tests. They were
carried out in a typical listening situation (in a typical living 1k
room). In the listening tests, different complex sounds were
presented to the subjects over the headphones (Beyedynamic 0
2 3 4 1
DT-770 Pro), while the subjects’ response was recorded by Time (s)
the condenser microphone (AKG C444) and stored in the Fig. 2 Spectrogram of generated complex sound used in the listening
corresponding audio tracks using HP 110 Mini computer with test 3
Intel Atom N270 processor and integrated sound board. The
task of the subjects was to repeat pitch of the reproduced In the listening test 4, the complex sound of duration of 2
sound by singing neutral syllable La or to give the answers to seconds consisting of 5 harmonics with random distribution
particular questions. was presented to the subjects in the same manner as in other
Ten subjects participated in the listening tests of which 7 tests. The frequencies of harmonics are: 440 Hz, 475 Hz,
were male and 3 were female. The mean age of the subjects 483 Hz, 502 Hz, 543 Hz. The spectrum of this complex sound
was 26. Nine subjects have graduated from the Faculty of Arts is given in Fig. 3. The task of the subjects was the same as in
(Music Arts) and one subject is a student of the secondary the tests 1 and 2 - to repeat the pitch of the reproduced sound
music school. In that regard, they can be considered to be by singing the neutral syllable La.
expert listeners. Similarly as in the test 2, the first partial (with frequency of
The listening tests are divided in five groups – test 1 to test 440 Hz) of the signal from test 4 was removed, and such a
5. In the listening test 1, the complex sounds of duration of 2 complex signal was presented to the subjects in the listening
seconds consisting of fundamental and 4 harmonics were test 5. The task of the subjects was the same - to repeat the
presented to the subject by the headphones. The frequency of pitch of the reproduced sound on the neutral syllable La. The

114
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

purpose of this test is to compare the results with those from 100

Relative amplitude (dB)

the tests 1, 2 and 4 in studying the phenomenon “missing
fundamental”.

100 50
Relative amplitude (dB)

50

0
100 1k 10k 20k
0 (a) Frequency (Hz)
100

Relative amplitude (dB)

-50
200500 1k
Frequency (Hz)
Fig. 3 Spectrum of complex sound consisting of 5 partials in non-
harmonic distribution used in the listening test 4 50

IV. RESULTS
0
Performed spectral analysis shows that chordophones and 100 1k 10k 20k
(b) Frequency (Hz)
aerophones instruments have harmonically related partials,
while membranophones and ideophones instruments do not 100

Relative amplitude (dB)

have a harmonic distribution of partials. Representative
examples of instrument sound spectra are given in Fig. 4.
The spectrum of piano (Fig. 4(a)) has harmonically related
partials. The frequency of the first harmonic (fundamental) is 50
215 Hz and frequencies of overtones are: 430 Hz, 645 Hz, 860
Hz, 1075 Hz, 1290 Hz, etc. The spectrum of oboa also has
harmonically related partials (see Fig. 4(b)). The frequency of
the fundamental is 656 Hz and the frequencies of overtones 0
are: 1312 Hz, 1968 Hz, 2624 Hz, etc. On the other hand, snare 100 1k 10k 20k
and crash chinel do not generate harmonically related partials (c) Frequency (Hz)
(see Fig. 4(c) and (d)). 100
Relative amplitude (dB)

In the listening test 1, all 10 subjects were able to

successfully repeat the pitch of the reproduced complex sound
containing fundamental and 4 harmonics. A representative
example of a subject’s response is given in Fig. 5. From
presented spectrum and spectrogram, it can be seen that 50
frequency positions of fundamental and harmonics are close
to those from the test sound.
The results of the listening test 2 are similar to those of the
test 1. So, all 10 subjects were able to successfully repeat the 0
1k 100 10k 20k
pitch of the reproduced complex sound without fundamental (d) Frequency (Hz)
(but with harmonically related partials), that is, they were able
Fig. 4. Spectra of sounds of representative instruments: piano (a),
to reproduce the pitch of missing fundamental. A oboa (b), snare (c) and crash cymbal (d)
representative example of the response of a subject is given in
Fig. 6. Similarly to the results of the test 4, the perceived and
In the listening test 3, where the subjects compared two reproduced pitch of complex sound without first partial
complex sounds (with and without fundamental) separated by presented in the listening test 5 is not the same for all subjects.
silence of 1 second, 9 subjects responded that there was a The distribution of the fundamental frequency of the subjects’
difference in timbre, and 1 subject responded that there was a responses (sang neutral syllable La) is shown in Fig. 8.
difference in pitch of compared sounds.
The results of the listening test 4 are different than those
from the tests 1 and 2. Since the partials of the presented V. CONCLUSION
complex sound are not in harmonic relation, the subjects were
not able to perceive and reproduce (sing) the same pitch. The From the previously mentioned results, it can be concluded
distribution of answers, that is, distribution of the frequencies that the distribution of partials (harmonics) in complex sound
of fundamental of sang neutral syllable La is shown in Fig. 7. plays a key role in the pitch perception. In the case of

115
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

harmonically distributed partials in a complex sound, the most

3
energy excites the multi-peak neuron(s), so that the
impression of pitch is unambiguously clear to the listener. In
absence of fundamental frequency, other harmonics excite 2

multi-peak neuron whose central frequency corresponds to the Number of

subjects
fundamental, and the listener’s impression of pitch
1
corresponds to that of the fundamental frequency. This
principle can explain the phenomenon “missing fundamental”.
0
107 Hz 140 Hz 215 Hz 226 Hz 236 Hz 495 Hz
100
Relative amplitude (dB)

Fundamental frequency of subjects response

Fig. 7. Distribution of fundamental frequency of subjects’ responses

(sang neutral syllable La) from the test 4
50
3

2
0
100 1k 10k 20k Number of
(a) Frequency (Hz)
subjects
1

20k
Frequency (Hz)

0
139 Hz 150 Hz 204 Hz 226 Hz 236 Hz 247 Hz 495 Hz
15k
Fundamental frequency of subjects response

10k Fig. 9. Distribution of fundamental frequency of subjects’ responses

(sang neutral syllable La) from the test 5
5k

0 ACKNOWLEDGEMENT
1 1.5 0.5
2
(b) Time (s)
This research is partially supported by the Ministry of
Fig. 5. Representative response of a subject in the listening test 1:
Science and Technological Development of Serbia through
spectrum (a) and spectrogram (b) of the sang neutral syllable La
the project No. 44009.
100
Relative amplitude (dB)

REFERENCES
[1] Xiaoqin Wang, “The Harmonic Organization of Auditory
50 Cortex”, The Neurophysiological Bases of Auditory Perception,
pp. 211-222, 2010.
[2] William A. Sethares, “Tunnig, Timbre, Spectrum Scale”,
Second Edition, London, Springer-Verlag, 2005.
[3] Martin Russ, “Sound Synthesis and Sampling”, Third Edition,
0
100 1k 10k 20k Oxford, Focal Press, 2009.
(a) Frequency (Hz) [4] ”Jennifer K. Bizley, Kerry M. M. Walker, “Sensitivity and
Selectivity of Neurons in Auditory Cortex to the Pitch, Timbre,
20k and Location of Sounds”, The Neuroscientist, Vol. 16, No. 4,
pp.453-469, 2010.
Frequency (Hz)

15k [5] B. Kostek, A. Czyzewski, “Representing Musical Instrument

Sounds for their Automatic Classification”, J. Audio Eng. Soc.,
10k vol. 49, no. 9, pp. 768-785, 2001.
[6] S. Ando, K. Yamaguchi, “Statistical Study of Spectral
5k Parameters in Musical Instrument Tones”, J. Acoust. Soc. Am.,
vol. 94, no. 1, pp. 37-45, 1993.
[7] J. Meyer, “The Sound of the Orchestra”, J. Audio Eng. Soc.,
0
1 1.5 0.5 vol. 41, no. 4, pp. 203-213, 1993.
(b) Time (s) [8] J.C. Brown, “Musical Fundamental Frequency Tracking Using a
Fig. 6. Representative response of a subject in the listening test 2: Pattern Recognition Model”, J. Acoust. Soc. Am., vol. 92, no. 3,
spectrum (a) and spectrogram (b) of the sang neutral syllable La pp. 1394-1402, 1992.

116
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Digital Bandpass IIR Filers with High Selectivity

Peter Apostolov1
Abstract – This paper proposes an optimal third-degree squares norm, and L∞ - weighted Chebyshev’s norm. In the
polynomial, which approximates Kronecker’s delta function with
high precision. The polynomial is obtained by a new literature different polynomial approximation methods are
approximation method, called “method with compressed proposed. Fig.1 shows approximations with Hausdorff
cosines”. The method is based on Chebyshev’s optimality norm. (Chebyshev) polynomial [1] using L∞ norm, with sinc ( .)
The polynomial is used for narrow bandpass IIR filter design.
The filter’s selectivity depends on the parameter Q without
increasing the polynomial’s order. With the proposed method an
IIR filter with 5(6) multipliers, a very narrow passband and a
high stopband attenuation can be designed.

Keywords – IIR digital filters, Frequency response, Polynomial

approximation.

I. INTRODUCTION
The task of filter synthesis is a mathematical problem of
approximating ideal functions with rectangular shape. The
transfer function of the filter results from the approximation.
The aim is to obtain a mathematical relationship which has the
lowest computational complexity and approximation error. In
approximations with polynomials, this indicator is the degree
of the polynomial. Fig. 1. Polynomial approximations of the Kronecker’s delta
This paper will show a method for digital filter design
based on a polynomial approximation with “compressed function using L2 norm, method of Parks-McClellan [2] with
cosines”. trigonometric polynomial using L∞ norm.
It is seen that a suitable trade-off between the flatness in the
II. BACKGROUND stopband and the bandwidth must be done. In all the criteria,
the functions have the oscillations in the stopband. These
In some practical cases the passband filter is required to oscillations are undesirable. The goal is to obtain a rectangular
have a very narrow bandwidth. The ideal characteristic of a shape of the ideal function, that has maximally flat pass band
supernarrowband filter is Kronecker’s delta function and stop band, and narrowest possible bandwidth. In L2 case
the oscillations increase near the main lobe. This is due to the
1, x = 0.5
δ ( x) =  ; x ∈ [ 0,1] . (1) Gibbs’ phenomenon [3]. In the approximations using L∞
0, x ≠ 0.5 norm the oscillations are with equal amplitude. These
This is a transfer function of a filter that has a pass approximations are known as optimal and equiripple.
bandwidth equal to zero, stopband gain equal to zero and an The approximations with rational functions [4, 5], have
infinite steepness of its characteristic. It cannot be realized in better properties than the polynomials approximations. The
practice. Hence, Kronecker’s delta function needs to be most popular are Chebyshev, Butterworth and Cauer.
approximated by another one, which can be realized. The In [5, 6] a polynomial approximation method with
approximation is carried out with a preset accuracy ε > 0 . The compressed cosines is proposed. With this method a third
difference between the ideal function and the approximating degree polynomial with significantly better properties than the
polynomial defines the error function. The two most popular other polynomials approximations is derived. The
approximation accuracy is close to the approximations with
norms for the approximation are L2 - weighted integral least-
rational functions. The polynomial has the form
4

P3 = ∑ bk cos [( k − 1) ϕ ] , (2)
1
Peter Apostolov is with the Department of Wireless k =1
Communications and Broadcasting at the College of
Telecommunications and Posts, Sofia 1700, 1 Acad Stefan Mladenov with coefficients:
St, Bulgaria. E-mail: p_apostolov@abv.bg

117
 I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

b1 = 0.5 − ε ; b2 = b4 = 0 ; b3 = −0.5 . (3) A normalization of the frequencies is done as

The function ∆f n = 2 ∆f pass f d = 0.025 ; f 0 n = 2 f 0 f d = 0.2 . (7)

  2 f − 2 f0 fd  The optimal approximation error is determined as

ϕ = −π − 2arctg Q  −  , (4)
  fd 2 f − 2 f 0  1
ε= DS 20
= 0.0909 . (8)
is the phase response off the allpass lattice filter with quality 1 + 10
factor Q, f ∈ [ 0, f d 2 ] is frequency, f d is the sampling When the normalized bandwidth ∆f n is set, the Q-factor
can be determined approximately by the following five
equations:
1 δ
11
P3
Q = f 0 n ∑ ak ∆f n
11− k
1-ε ; ∆f n ∈ [ 0.1, 0.0127 ) ; (9)
0.8 k =1

0.6
a1=5.6075e14 a2=-3.348e14 a3=8.7823e13 a4=-1.3321e13
a5=1.2944e12 a6=8.4311e10 a7=3.7445e9 a8=-1.1304e8
0.4
a9=2.2671e6 a10=-2.8705e4 a11=1885.71
11

Q = f 0 n ∑ ak ∆f n
11− k
0.2 ; ∆f n ∈ [ 0.0127,1.272e − 3 ) ; (10)
ε k =1
ε
0
a1=1.2838e25 a2=-9.0581e23 a3=2.79655e22 a4=-4.9689e20
-ε -ε
a5=5.6239e18 a6=-4.2396e16 a7=2.1628e14 a8=-7.435e11
-0.2
0 0.25 f
s1
0.5 f
s2
0.75 1 a9=1.6817e9 a10=-2.3764e6 a11=206.1499
11
Fig. 2. Approximation of the Kronecker’s delta by a third-degree
Q = f 0 n ∑ ak ∆f n
11− k
optimal polynomial ; ∆f n ∈ [1.272e − 3, 3.14e − 4 ) ; (11)
k =1

frequency, f 0 is the middle frequency of the passband. a1=9.934e34 a2=-7.838e32 a3=2.7414e30 a4=-5.6059e27
Fig. 2 shows an optimal approximation of Kronecker’s delta a5=7.4403e24 a6=-6.7267e21 a7=4.2269e18 a8=-1.8471e15
function by an optimal 3rd degree polynomial . f s1 and f s 2 are a9=5,5072e11 a10=-1.0692e8 a11=1.2226e4
the two normed stopband frequencies. Their difference 11

Q = f 0 n ∑ ak ∆f n
11− k
defines the bandwidth ∆f stop . The passband is defined by ; ∆f n ∈ [ 3.14e − 4, 5.965e − 5 ) ; (12)
k =1
∆f pass - the bandwidth at level -3dB. The approximation error
a1=-1.6808e41 a2=2.5067e38 a3=-1.4578e35 a4=3.5739e31
ε determines the stopband attenuation DS, and the quality
a5=1.3647e27 a6=-3.3608e24 a7=1.016e21 a8=-1.6545e17
factor Q the bandwidth ∆f stop . The filter’s coefficients are a9=1.6554e13 a10=-1.0265e9 a11=36979.1
obtained by those of the polynomial:
11

Q = f 0 n ∑ ak ∆f n
11− k
hk = b4 2 , b3 2 , b2 2 , b1 , b2 2 , b3 2 , b4 2 . (5) ; ∆f n ∈ [ 5.965e − 5, 5.5e − 6 ) . (13)
k =1

The filter’s transfer function has the form a1=-4.7623e49 a2=1.2294e46 a3=-1.1141e42 a4=2.0403e37
a5=3.8601e33 a6=-3.5797e29 a7=1.5185e25 a8=-3.7635e20
H = −0.25 + ( 0.5 − ε ) exp ( − jϕ ) − 0.25 exp ( − j 2ϕ ) . (6)
a9=5.697e15 a10=-5.176e10 a11=263267.1
By substituting the defined in (7) ∆f n = 0.025 into (9),
III. DESIGN EXAMPLE Q = 5.0714 is obtained. This allows for defining the transfer
function of the allpass lattice filter. The coefficients of the
The realization will be demonstrated with an example of a
denominator of the transfer function are determined by those
bandpass digital IIR filter design with the following
of the denominator of the Butterworth bandpass filter of first
specification: middle frequency in the pass band f 0 = 800 Hz; order with bandwidth
∆f pass = 100 Hz; sampling rate f d = 8000 Hz; degree of the
∆f Butt = f 0 Q = 157.7469 Hz. (14)
polynomial m = 3 ; attenuation in the stopband DS ≥ 20 dB.

118
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The coefficients in the numerator are the same as in the accordance with the commutative law. This allows the input
denominator, but in reverse order. signal to pass twice through one allpass lattice filter. The
bandpass filter is implemented with the scheme in Fig.4
−1 −2
c1 + c2 z + z Fig. 5 shows the magnitude response in dB (10 lg x ) . Fig.
H AP ( z ) = −1 −2
; z = exp ( − jω ) ;
1 + c2 z + c1 z 6 shows an output response of a computer simulation of the
filter with 5 multipliers. The filter input is fed with a
c1 = 0.8832 ; c2 = −1.5265 . (15) discretized at 8000Hz linear chirp signal with frequency
sweep from 1 to 4000Hz, amplitude of ± 0.25V and duration
Then of 10 seconds. It is seen that the filter’s output response
corresponds to the input specification.
−1 −2
0.8832 − 1.5265 z + z This implementation requires frame signal processing with
exp ( − jϕ ) = H AP ( z ) = −1 −2
. (16)
1 − 1.5265 z + 0.8832 z
In Fig.3, the diagram of the designed filter is shown. A
0
criterion for comparing the selectivity of the digital filters is
the number of multipliers with which they are realized.

Magnitude (dB) (normalized to 0 dB)

The coefficients having the same value are realized with
-5

-10

-15

0 0.5 1 1.5 2 2.5 3 3.5

Frequency (kHz)

Fig. 5. Magnitude response

buffers. The concatenation of two neighbor fragments is

treated with "overlap" to remove the uncertainties, which
result from the filtration at the beginning of each fragment.

1
Fig. 3. Functional diagram of the filter
0.8
one multiplier to reduce the power consumption. As it is
0.6
known, the allpass lattice filters are realized with 4
multipliers. Therefore, the total number of multipliers in the 0.4

diagram is 11. This scheme can be realized with only six

Output Response

0.2
multiplier as both allpass filters are the same and
0
h1 = h3 = −0.25 . If the filter’s coefficients are multiplied by 4,
-0.2
then h1 = h3 = −1 . Then the filter will be realized with 5 -0.4

-0.6

-0.8

-1
0 400 800 1200 1600 2000 2400 2800 3200 3600 4000
f(Hz)

Fig. 6. Computer simulation – output response

Fig. 4. Functional diagram of the filter with 5 multipliers IV. DISCUSSION

multipliers, as in digital signal processing the change of the
sign with the operation x = − x is performed. In this case the An advantage of the method of compressed cosines is that
the approximation is carried out with third-degree polynomial.
filter will amplify the signal four times (12dB).
The polynomial’s coefficients are calculated easily. To obtain
The scheme shows that all signals are summed. Therefore
a high selectivity, it is not necessary to increase the degree of
in the design the sequence of the operations is irrelevant, in

119
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

the polynomial, as in other polynomial approximations, but to 2. Butterworth, Chebysev and Cauer filters of second order
use an allpass lattice filter with high Q-factor. The bandwidth are implemented by 7-9 multipliers. Fig.9 shows a similar
of the stopband ∆f stop is a result of the approximation. It can comparison. In this case the magnitude response of the filter
using compressed cosines with 5 multipliers has a lower
not be defined in the input specification. With a third-degree selectivity.

Magnitude Response
0
0 Compr.cos
Butterworth
-20
-10 Chebyshev
20lglAl

-40 Cauer

-60 -20

-80
0 500 1000 1500 2000 2500 3000 3500 4000

20lglHl
f(Hz) -30
Pass Band Ripple Stopband
0
0
-40
-1 -20
20lglAl

20lglAl

-40 -50
-2
-60

-3 -80 -60
799 799.5 800 800.5 801 760 780 800 820 840 0 500 1000 1500 2000 2500 3000 3500 4000
f(Hz) f(Hz) Frequency(Hz)

Fig.7. Bandpass filter - magnitude response Fig. 9. Comparison of magnitude responses

polynomial, a filter with an arbitrary bandwidth ∆f pass and a

stopband attenuation DS can be realized. For example, Fig. 7 V. CONCLUSION
shows the magnitude response of a filter with a passband of
1Hz and a stopband attenuation of 60dB. The obtained results show that the selectivity of the filters
The most commonly used IIR digital filters are those of with “compressed cosines” is determined by the steepness of
Butterworth, Chebyshev and Cauer. A criterion for comparing the S-curve of the allpass filter’s phase response (Q-factor).
the selectivity is the number of multipliers. From (4) it is seen that it is the function arctg(.). To obtain a
1. The scheme of the digital filter of Fig. 4 is realized high selectivity it is necessary to use a function with a greater
always with 5 multipliers, regardless of the filter’s gradient, e.g. tanh(.). Unfortunately, an allpass filter with such
specification. This comes at the expense of using a larger a phase response has not been realized until now.
volume of the memory, which is not a serious disadvantage. The proposed method may be a good alternative in several
The filters of Butterworth, Chebysev and Cauer of first order applications in IIR bandpass filter design.
are implemented by 4 multipliers. Due to the low order, their
magnitude responses are identical. Fig.8 compares magnitude
REFERENCES
responses of bandpass filters with equal banwidth ∆f pass . The
magnitude response of the filter using compressed cosines has [1] Sendov, B. Hausdorff Approximations. Kluwer Academic
better selectivity. Publishers London 1990, ISBN: 0792309014.
[2] Parks, T. W. and J. H. McClellan. A Program for the Design of
Linear Phase FIR Digital Filters. IEEE Trans. on Audio and
0 Electoacoustics, Vol. AU – 20, №3, pp. 196-199, August 1972.
Compr.cos [3] B. Porat, A Course in Digital Signal Processing. New York:
Butterworth Wiley, 1997.
-10 Chebyshev
Cauer
[4] Daniels, R., Approximation Methods for Electronic Filter
Design. McGraw Hill, 1974.
-20 [5] Schaumann, R., M.E. van Valkenburg. Design of Analog
Filters, Oxford University Press 2001.
[6] Apostolov, P. S. Linear Equidistant Antenna Array with
20lglHl

-30
Improved Selectivity, IEEE Transaction on Antennas and
propagation, Vol.59, Issue10, pp.3940-3943, Aug. 2011.
-40 [7] Apostolov, P. S. Methof for FIR filter design with compressed
cosine using Chebyshev’s norm. Signal Processing Elsevier,
Vol. 91, Issue 11, pp. 2589-2594, Nov.2011.
-50

-60
0 500 1000 1500 2000 2500 3000 3500 4000
Frequency(Hz)

Fig. 8. Comparison of magnitude responses

120
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Features of time-frequency analysis visualization of large

dynamic range signals
Tihomir Trifonov1, Ivan Simeonov2 and Rosen Dzhakov3
Abstract – In this paper practical time-frequency analysis of and biometrics. The detection and classification of the
large dynamic range signals is made. The acoustic signals from transmitted object becomes easily. This way the methods of
unique Bulgarian bells recorded by PULSE 12 data acquisition biometric iris recognition can be used for process automation.
hardware are analyzed.
The mechanical and acoustical properties of these objects are
discussed and their features are determined. Some specific II. METHOD OF “SOUNDPRINT” VISUALIZATION
techniques of visualization are proposed.
A. Wavelet transform applications
Keywords – time-frequency analysis, acoustic signal
visualization, conformal map
Every signal (or function of time) (t), can be described by

I. INTRODUCTION
At work on the project about research of the valuable
Bulgarian bells the database was created, in which almost all
characteristics of the bells were included [1]. Database
includes acoustical records, obtained via unique measurement
set. The used measurement and processing methods can be
implemented to the other purposes.
The bell is a complicated sound source with a very wide
frequency range and an unique dynamic range of the
transmitted signal. Its spectrum contains infrasound, sound
and ultrasound partials. The dynamic range is very large too
and it cannot be detected entirely by human ear whose
dynamic range of perception is about 120dB. The best all over
the world measurement set with a corresponding measurement
Fig. 1 Localization characteristics of 
microphone at this moment was used because of this [2]. For
example, this set is able to process the signal without interval in the time axis and interval , in frequency
distortions with dynamic range up to 160dB. which are including 90% of his energy, concentrated around
The most modern processing methods and integrated center of mass of functions  and ². The
system of computer mathematics MatLab are used [3]. The modulation on this function is translation of the rectangle
features of the source and the raw records require this way of across axis ω, while the scaling of function (her contraction or
measurement [4].
In this paper we propose a new method of presentation of
some transformations (Fourier Transform and Wavelet
Transform for example). Wavelet Transform gives the
improvement for analysis and reception via conversion of 2D
signal into pseudo 3D signal. Conform transformation
improves these possibilities in addition. We introduce “sound
print” as analog of the “fingerprint”, used in the criminology

1
Tihomir Trifonov is with the Dep. of Algebra and Geometry at
Veliko Tarnovo University "St. St. Cyril and Methodius", 3 Arch.
Georgi Kozarov Str., V. Turnovo BG-5000, Bulgaria, e-mail:
tihomirtrifonov@ieee.org
2
Ivan Simeonov is with the Dep. of CIS at National Military
University Vasil Levski of Veliko Turnovo, 76 Bulgaria Blvd, V.
Turnovo BG-5006, Bulgaria, e-mail: ivanov_ivan@nvu.bg.
3
Rosen Dzhakov is with the Faculty of Artillery, Air-Defense and
CIS Faculty of NMU, 1 Karel Shkorpil Str., Shumen BG-9713
Shumen
Fig. 2. The disposition of unique bells, denoted as
Мelnik1 -1270AD and Melnik2-1220 AD, in the National
Historical Museum in Sofia.

121
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

stretching) changes the rectangle proportions. Theese objects denoted as Melnik2-1220 AD. and Мelnik1 -
In this case, the function (t) can be represented as
rectangle on the plane , as shown in fig. 1.
Unlike to Fourier transform - FT and Short time fourier
transform – STFT, wavelet transform will alters the rectangle
type for analysis according to the frequency, area of rectangle
will stay constant. An illustration of local properties of
wavelets in frequency area is shown on figure 3. This is a kind
of analyze where, the relation is constant or the
quality factor Q is equal. The time-frequency window area
stay for a different scales, where и are the
second central moments on the functions and .
More precisely, suppose that , or (a,b)
determine one point in right-half plane, then the continuous
wavelet transform (CWT) of a continuous, square-integrable Fig. 4. Data Acquisition Unit 3560В Brüel & Kjær
function is expressed by:

1270AD are shown in Fig.2 [2]. The measuring microphone

, (1) 4193 Brüel&Kjær and Data Acquisition Unit 3560В
Brüel&Kjær [2] is illustrated in Fig.4.
where < , > denotes the inner product. Experimental set-up includes:
• Pressure-field Microphone Type 4193
The wavelet transform of a one-dimensional signal is a two- Brüel&Kjær,[2] available in Transducer Electronic
dimensional time-scale joint representation, [7]. So the Data Sheet (TEDS) combinations with the classical
resolution of identity must be satisfied, that is expressed as Preamplifier Type 2669 with an individual
calibration; Dynamic Range: 19 … 162 dB,
, (2) Sensitivity: 12.5mV/Pa.
• Vibration Transducer TRV-01 SPM Instrument;
where, • Compact Data Acquisition Unit 3560В Brüel &
Kjær, [2] for outdoor use that consist: Dyn-X input
is basis that satisfy the conditions modules with a analysis range exceeding 160 dB and
of admissibility (the mean value equal to zero), regularity (has automatic detection of front-end hardware and
exponential decay, so that its first low order moments are transducers − supports IEEE 1451.4-capable
equal to zero), and orthogonality, see fig. 3, transducers with TEDS (Transducer Electronic Data
Sheet); output TCP/IP protocol communication - RJ
45 connector complying with IEEE−802.3100baseX;
Multiframe Control option;
• Base software PULSE 12 for CPB (Constant
Percentage Band) analysis 2 channels; 5-channel
Time Capture; PULSE Bridge to MATLAB®
• MathWorks Software - MatLab&Simulink, toolboxes
for FFT and Wavelet analysis.[3]
It can be seen, that the hardware equipment and the
software manufacturers are known for theirs high quality all
over the world. A part of equipments are shown in Fig.4.

Fig. 3 Wavelet basis functions , and time-

C. Conformal map soundprint visualization
frequency plane image, Continuous Wavelet Transform – CWT
Wavelet analysis as a tool allows a deeper analysis of sound
- the admissibility constant, frequencies. The using of scalogram plots was providing new
scale parameter, , pictures for complex sounds structure.
– time shift parameter. In this section we analyze the Bell sounds structures. The
preprocessed signals named “melnik1-1.mat” from Мelnik1 -
B. Experimental set-up 1270AD Bell and „melnik2.mat“ Melnik2-1220 AD are
obtained by Brüel & Kjær’s Data Acquisition Unit 3560В,[6].
An experimental set-up was realized to record the sound of The Continuous Wavelet Transform signals calculations
unique bells in the National Historical Museum, Sofia, [4,5]. were produced in MatLab, Continuous Wavelet 1-D tool [3].

122
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig.7 Conformal map visualization of the signal

Melnik1_10

Fig. 5 An conformal mapping example, “NearBreaks I” test

signal.

If we make known conformal mapping - logarithm function

ln(z), the rectangular graph will be transformed to a circular
graph. An example is given in Fig 5, where the signal is “Near
Breaks I” test signal.
Fragments of bell strikes Мelnik1 and Melnik2 in the time
scale are shown in Fig.6 and Fig.8 respectively. The images
that result from continuous wavelet transformations are
illustrated to Fig.6 and Fig.8 in the bottom.

These scalogram coefficients are calculaated by Daubechies

wavelets order 3.
On the figures 7 and 9 are illustrated the conformal map Fig. 8 Fragment of the eight strike tail and its respective
visualizations of the same bell strikes (Мelnik1_10 and scalogram (Melnik2_8) - CWT coefficient. Daubechies wavelets
Melnik2_8) for various number of coefficients order 3.

Fig. 9 Conformal map visualization of the signal Melnik2_8

Фиг. 6 Fragment of the signal Melnik1_10 - the strike tail, as well as

its respective scalogram - CWT coefficients.

123
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

In the figures 10 was shown fragments of the strikes (in

tails) in left and in the right was shown theirs conformal maps.
REFERENCES
[1] BELL: Research and Identification of Valuable Bells of the
Historic and Culture Heritage of Bulgaria and Development of
Audio and Video Archive with Advanced Technologies), KIN-
1009, Institute of Mathematics and Informatics, Bulgarian
Academy of Sciences , (http://www.math.bas.bg/bells)
[2] Brüe&Kjar, www.bksv.com
[3] MathWorks, www.mathworks.com
[4] Tihomir Trifonov, Tsvetanka Georgieva, Web based approach
to managing audio and video archive for unique Bulgarian bells,
Proceedings of the Tenth International Conference on System
Analysis and Information Technologies, Kyiv, Ukraine, 2008,
page 325
[5] G. Dimkov, Al. Aleksiev, I. Simeonov, T. Trifonov, K.
Simeonov, Acoustical researches on historical valuable
Bulgarian bells, In Proceedings of the National Scientific
Conference on Acoustics, Sofia, Bulgaria, 2008, pp. 115-124 (in
Bulgarian).
[6] Tihomir Trifonov, Georgi Dimkov, Rosen Dzhakov, lvan
Simeonov, Research and identification of valuable bells of the
national historic and cultural heritage of Bulgaria, In
Proceedings of the XXII Conference with International
Participation "Noise and Vibration", Nish, Serbia, 2010, pages
103-107.
[7] Alexander D. Poularikas, The Transforms and Applications
Fig. 10 Visualization of the signals Melnik1_10( top) vs Handbook, Second Edition, CRC Press, 1999
Melnik2_8(bottom).

APPENDIX
III. ADDITIONAL REMARKS
Bell’s donation inscriptions XIII century:
The possibility for analyze and perception is improved by
transformation on two-dimensional acoustic signal image to Bell Melnik2, 1211-1216 year
pseudo three-dimensional (scalogram). Material: bronze, Place: Tower-belfry on the metropolitan
Additional improvement in perception is achieved by using church of St. Nicholas, Melnik.
a conformal mapping of the obtained scalogram, because there "† The bell /is/ forged out of copper, a gift from despot Alexii,
are well-known iris recognition techniques that can be † pious Slav to St. Nicholas, he who is from Mira.“
applied.

IV. CONCLUSION Bell Melnik1, 1270

Material: bronze, Place: Belfry of the monastery of St.
In the presented paper we propose a new method of Charalambius - Saints Achangels, Melnik.
presentation of some transformations (Fourier Transform and
Wavelet Transform for example). Wavelet Transform, except
its well known advantages, mentioned above, gives the "Lord, help your servant Theodosii monk who for the first
improvement for analysis and reception via conversion of 2D time has created (sanctified) a bell for strategus Mihail,
signal into pseudo 3D signal (scalogram). Conform the one that is in Melnik, this one that has been fixed
transformation improves these possibilities in addition. We under the reign of Michael Paleologus, the new
introduce “sound print” as analog of the “fingerprint”, used in Konstantine. In March, indiction 12 year 6778 (=1270)."
the criminology and biometrics. The detection and
classification of the transmitted object becomes easily. This
way the methods of biometric iris recognition can be used for
process automation.
Our future work will be pointed to obtain more reasons of
practical implementation of proposed method.

124
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Accuracy Improvement of Allpass-based Digital Hilbert

Transformers
Kamelia Nikolova1 and Georgi Stoyanov1
Abstract – A design procedure to reduce the deviation of the special measures have to be taken to prevent that. Higher
phase from 90º of allpass-based digital Hilbert transformers is accuracy could be achieved by designing the HTs with higher
proposed. This is achieved by introducing the phase sensitivity TF order, but the portability of the equipment is imposing
minimization of each individual allpass section in the cascade another constraint – the power supply limitation. The main
realizations of the two branches of the structure used. The
aim of this work is to try to improve the accuracy of the
effectiveness of the proposed design is experimentally proven.
allpass-based HTs throughout minimization of their sensitivi-
Keywords – digital filters, allpass filters, Hilbert transformers, ties. It will reduce the computational load and will permit
sensitivity minimizations. shorter word-length and lower power consumption for given
accuracy. The design procedures should be straightforward,
without iterative and complicated optimization steps, in order
to be easily used by practicing engineers and the structures
I. INTRODUCTION have to be with the lowest possible TF order and complexity.

Hilbert transformers (HT) are very important building

blocks in both, analog and digital signal processing. They are II. DESIGN PROCEDURE
used in telecommunications for generation of analytic and
single-sideband signals [1] [2] and in many other modulation An ideal Hilbert transformer (also known as a 90-degree
and demodulation schemes (mainly for splitting the narrow- phase shifter) is described in frequency domain as [5]
 j, 0    
H HT (e j )  
band signals to two (I and Q) components), in complex signals
. (1)
processing, in audio and video signal processing, and even in  j,      0
fields like mechanical vibration signal processing. Many A way to synthesize an IIR Hilbert transformer (called also
approaches and methods of design of digital HTs have been a complex half-band filter) is to start with an odd-order half-
developed in the last 50 years and most of them have been band filter with specifications Fp, Fs, δp and δs, interconnected
well systematized in [3]. The FIR based HTs are providing by the relations [3]
easily a linear phase response and unconditional stability but
at the price of a very high transfer function (TF) order (say,  s  sin(max / 2);  p  1  1   s2 ; Fp  0.5  Fs ; (2)
several hundred), producing quite a high total delay and
and with a TF G(z ) that may be represented as a sum of two
requiring higher power consumption. These disadvantages are
eliminated in the IIR realizations, most often based on the allpass TFs [3] [5]
usage of allpass structures. The theory of the allpass-based G( z )  0.5[ A1 ( z 2 )  z 1 A2 ( z 2 )] . (3)
HTs is quite mature and several design methods using real or
complex allpass structures have been summarized in [3]. An "even-odd" decomposition (Fig. 1) and the substitution
Many new optimization-based methods for design of half- H ( z)  2G( jz) (4)
band filters and HTs have been proposed since then (including
even frequency response masking technique [4]), but no must be applied in order to obtain the real allpass TFs. Thus
specific methods for accuracy improvement have been repor-
ted. Meanwhile the practical importance of the HTs grew
considerably with the extension of the frequency ranges and
the growth of the proportion of the narrow-band signals,
described as analytic, in telecommunications. The problem
with the accuracy of the realization of the HTs is of
paramount importance in many of these telecommunication
applications, like in the maintenance of I and Q channels
balance in a wide frequency range. When the HTs are realized
using a fixed-point arithmetic (what is often the case in the
portable and mobile communication equipment), the limited
word-length may reduce considerably that accuracy and Fig. 1. "Even-odd" decomposition of the TF poles.

H HT ( z )  2G( jz)  [ A1 ( z 2 )  jz 1 A2 ( z 2 )] (5)

1
The authors are with the Faculty of Telecommunications at
represents the HT as a complex sum of two real allpass
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000,
Bulgaria; e-mails: ksi@tu-sofia.bg, stoyanov@ieee.org. functions, whose realization (for real input signal x(n)) is
given in Fig. 2. Details about the design are given in [3] [5].

125
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

comparing the results with our previous investigations in [8]

[9], it can be noted that the WS sensitivity behavior of the
special second order sections is very similar to that of the
corresponding first order sections but with the symmetry
around the frequency f = 0.5. It is clearly seen that there exists
a proper selection of the sections for every given TF pole
Fig. 2. HT realization. position because of the significant difference between the
maximal values of the sensitivities (in some cases it can reach
more than 100 times especially for the poles near 1 ).
III. ALLPASS SECTIONS REALIZATIONS
The allpass TFs in Eq. (3) are having all their poles on the
imaginary axes, while those in Eq. (5) are all on the real axes.
In order to obtain higher accuracy in the 90º phase shifting in
case of a limited word-length environment, the allpass TFs in
Fig. 2 could be realized as cascades of special second-order
allpass sections. It follows from Fig. 1 that if a cascade
realization would be used, as the possible real pole positions
are scattered all around the real axes, the allpass sections with
low sensitivities for all these positions will be needed.
We have studied [6] all known (about 20) first order
sections and it was found that several low-sensitivity sections
for every real pole position could be found. We select to use
the most typical four of them, namely the ST1 section,
providing low-sensitivity for poles near z=1, MH1 and SC,
having low sensitivity for poles near z=0 and SV section for (a) bMH1-2 = 0.18654
poles near z=-1. The special sections are obtained from these
real first order sections by changing the signs of the
coefficients of the allpass TFs in Eq. (3) and by replacing z-1
by z-2 as it is shown in Fig. 3. We denote these new second
order allpass sections as MH1-2, ST1-2, SV-2 and SC-2.
z -2 z -2
In a In

Out b

(a) ST1-2 (b) MH1-2

z -2
In c

Out

(c) SC-2 (d) SV-2 (b) bMH1-2 = 0.94167

Fig. 3. Different special second-order allpass sections. Fig. 4. Worst-case phase-sensitivities of second-order allpass
sections (Fig. 3) for two different TF poles positions.
Their TFs are:
 1  a   z 2  b  z 2
H ST 1 2 z   ; H MH12 z   ; (6) V. OVERALL SENSITIVITY INVESTIGATIONS
1  1  a z  2 1  bz 2
 b  z 2 1  c  z 2 In order to estimate how the proper choice of the special
H SС  2 z   ; H SV  2 z   . (7) sections will affect the behavior of the HT realization in a
1  bz  2 1  1  c z  2 limited word-length environment, we have designed and
investigated a ninth order HT having the TF poles positions
given in Fig. 1 (the initial elliptic half-band filter specifica-
IV. ALLPASS SECTIONS SENSITIVITY tions are: passband frequency Fp = 0.24 and stopband atte-
INVESTIGATIONS nuation δs = 0.01 (Rs = 40 dB), producing Δφmax = 1.15º).
Then, we have designed 4 different HT realizations (Fig. 2).
In Fig. 4 a, b the worst-case (WS) phase-response sensitivi- The first one was realized using the standard way (using only
ties of the above mentioned four special sections are given for MH1-2 sections) and it is marked in the figures as "4MH1-2".
realizations with two different TF pole positions. The sensiti- The allpass sections selection for the other realizations is
vities are obtained by using the package PANDA [7]. By based on the sensitivity minimization of the individual sec-

126
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

tions depending on their poles positions. Thus, in the second These additional deviations should be kept as lower as pos-
HT realization (denoted with "4 ST1-2") four ST1-2 sections sible mainly by reducing the influence of the parasitic effects
were used. In the third and fourth implementations, two MH1- (by minimizing the sensitivities to the variations of the multi-
2 and two ST1-2 sections have been selected. In the first case, plier coefficients values). It will appear from what follows,
we have a special section of each type in every branch of the that it might not be an easy straightforward procedure.
realization, while in the second case – two MH1-2 sections are The accuracies of the HT realizations (the phase difference
used in the upper branch (the real output) and two ST1-2 sec- between the two branches) in a limited word-length environ-
tions – for the imaginary output. The results for the overall ment are compared in Fig. 6. Based on the results shown in
sensitivity of the two branches are shown in Fig. 5. Fig. 5, it is natural to have a high sensitivity (to small changes
in the two branches) of the phase difference between the two
outputs in Fig. 2 for 4 MH1-2 HT realization, but the results
shown in Fig. 6a are quite surprising, compared to these in
Fig. 6b,c,d (with minimized sensitivity). We suppose that this
might be an effect due to some internal compensation between
the parasitic effects in the branches, explained with the
different signs of the sensitivities. The worst-case sensitivity
WS, used in our investigations, is not able to reveal these
mutual compensations, because it is eliminating the signs of
the individual sensitivities.
The highest accuracy, as it is shown in Fig. 6, is achieved
when we have two MH1-2 and two ST1-2 sections each in
every branch (I case) of the HT. In this case, the selection of
the sections and their placement in the branches are made
under the above mentioned observations.

(a)

(a) using 4 MH1-2 sections

(b)
Fig. 5. Worst-case phase-sensitivities of the HT (Fig. 2) realized
with different sets of allpass sections (for a 9-th order HT).
It appeared that the best configuration is with two MH1-2
and two ST1-2 sections, each in every branch (I case),
providing the lowest overall sensitivity in both paths.

VI. INVESTIGATION OF THE INFLUENCE OF THE

SECTIONS COMBINATIONS IN THE BRANCHES

The phase difference between the two outputs in Fig. 2 will

not be exactly 90º. Over some frequency range (narrower than
half of the sampling frequency) it will alternate around this
value with amplitudes Δφmax depending ideally only on the
selected value of δs Eq. (2), but in reality – also on the design (b) using 4 ST1-2 sections
accuracy and on the parasitic effects of the digital realization.

127
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

2. Decompose the TFs A1(z-2) and A2(z-2) to special second-

order allpass TFs and find where their poles are situated.
3. Select (from Fig. 3) or develop new allpass sections
realizing each couple of poles with the lowest sensitivity
and verify this by sensitivity studies as these in Fig. 4.
4. Investigate the overall sensitivities in the two branches
of Fig. 2 for all possible combinations of the selected
allpass sections realizations in order to select the best
set.
5. In case of a very high accuracy design, verify the selec-
tion by simulating the structure in a limited word-length
environment (as in Fig. 6).
We have applied this procedure for different sets of speci-
fications and it was always possible to find an implementation
clearly outperforming all the others as the case in Fig. 6 c.

(c) using 2 MH1-2 and 2 ST1-2 sections - I case VIII. CONCLUSION

A new approach to improve the accuracy of the allpass
based Hilbert transformers (realized as two parallel branches)
through sensitivity minimizations of each individual special
second-order allpass section in the cascade realizations of the
two branches was proposed in this paper. The design
procedure is simple and straightforward, without iterative and
complicated optimization steps and is achieving accuracy of
realizations close to the ideal case (nonquantized coefficients).
The low sensitivities so attained permit also a very short
coefficients word-length, a higher processing speed and lower
power consumption.

REFERENCES

(d) using 2 MH1-2 and 2 ST1-2 sections - II case [1] C. S. Turner, “An efficient analytic signal generator”, Signal
Proc. Magazine, vol. 23, pp. 91-94, July 2009.
Fig. 6. Word-length dependence of the accuracy of the HT phase [2] C. B. Rorabaugh, “Notes on Digital signal processing: practical
difference for realizations with different allpass sections. recipes for design, analysis, and implementation”, Notes 58-66,
As it can be seen after quantization to 2 bit (in CSD code) Prentice Hall, 2011.
not only the fluctuations of the phase difference in Fig. 6a,b,d [3] P. A. Regalia, “Special Filter Designs”, Chapter 13 in the book
are growing very much above the ideal, but the range of Handbook for Digital Signal Processing (Editors S. K. Mitra
and J. F. Kaiser), John Wiley & Sons, NY, pp. 909-931, 1993.
frequencies over which this difference is approximately
[4] L. Milic, J. Certic, M. Lutovac, “A class of FRM-based allpass
constant, is sharply reduced, while in Fig. 6c these parameters
digital filters with applications in half-band filters and Hilbert
are practically unchanged. transformers”, Proc. Intern. Conf. on Green Circuits and Syst.
The main conclusion of these investigations is that besides (ICGCS)’2010, Shanghai, China, pp. 273-278, June 2010.
the sensitivity minimization, an additional step, consisting of a [5] S. Mitra, Digital signal processing: A computer based
study of all possible combinations of the selected allpass approach, McGraw-Hill, 2006.
sections within the branches, has to be introduced. A more [6] G. Stoyanov and H. Clausert, “A comparative study of first
general solution of this problem will be a derivation of a order digital all-pass structures”, Frequenz, vol. 48, No 9/10, pp.
formula about the sensitivity of the phase quadrature to the 221-226, Sept./Oct. 1994.
changes of the multipliers’ values, but it may appear to be a [7] H. Sugino and A. Nishihara, “Frequency-domain simulator of
very difficult task. digital networks from the structural description”, Trans. of the
IEICE of Japan, vol. E73, No.11, pp. 1804-1806, Nov. 1990.
[8] G. Stoyanov, Z. Nikolova, K. Ivanova, V. Anzova, “Design and
VII. LOW-SENSITIVITY DESIGN PROCEDURE realization of efficient IIR digital filter structures based on sen-
sitivity minimizations”, Proc. 8th IEEE Conf. TELSIKS'2007,
Taking into account all results so obtained, we propose the Nish, Serbia, vol. 1, pp. 299–308, Sept. 2007.
following design procedure: [9] G. Stoyanov, K. Nikolova, and M. Kawamata, "Low-sensitivity
design of allpass based fractional delay digital filters", Chapter
1. Obtain H HT (z ) Eq. (5) by applying the standard design
7 in the book Digital filters, F. P. Márquez (Ed.), Intech Publ.
procedure from Sect. 2. House, pp. 155–178, 2011.

128
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Acoustic Standing Waves in Closed Cylindrical

Enclosures
Ekaterinoslav Sirakov 1 and Hristo Zhivomirov2
Abstract – The work presented in this paper provides a The solution of the wave equation is [6, 7]:
theoretical analysis of acoustic standing waves inside cylindrical
cos ( j ⋅ θ )
enclosures with rigid walls. Mathematical relationships are given
for the calculation and researching of modal frequencies and   n ⋅π ⋅ x   λ j ,k ⋅ r 
standing sound waves in a cylindrical box. The results from the p = or ⋅ cos  z  ⋅ J j ⋅  (2)
calculating and measuring of modal frequencies and the box sin j ⋅ θ  lz   lr 
response are shown graphically and in a table.  ( )
nZ , j , k = 0,1, 2,...
Keywords – Acoustic standing waves, Closed cylindrical
enclosures.
where: J j - Bessel function.
I. ACOUSTIC STANDING WAVES
In the paper are discussed the acoustic processes in a closed
cylindrical volume (Fig. 1). As a result of multiple reflections
of the sound waves from the walls of the volume three-
dimensional sound field arises [1], an example of which is
given in Fig. 2. Depending on the shape, dimensions and their
ratios in the enclosed volume fluctuations occur with a
different set of natural frequencies [2, 3].
The acoustic processes in a closed cylindrical volume can be
represented by the wave equation in cylindrical coordinate
system ( r , φ , z ) [4]:

 1 ∂  ∂p  1 ∂ 2 p ∂ 2 p  Fig. 2. Distribution of the magnitude of sound pressure in a

∇2 p = c ⋅  r + 2 + 2 (1)
 r ∂r  ∂r  r ∂φ
2
∂z  cylindrical box: j axial sound wave, f=3.538 kHz,
j = 3, k = 0, λ j , k = 4.201189 , cos( j ⋅ θ ) , mode (0, 3, 0)

where: φ - azimuth angle of the source.

Тhe natural frequencies for the corresponding values of nz,
j and k can be found by [5]:

2 2
c  nz   λ j , k 
f = ⋅   +  , Hz (3)
2  lz   π ⋅ lr 
φ
nZ , j , k = 0,1, 2,...
lZ (r , φ , z )
r The natural frequency of the cylindrical box, calculated in
accordance with mathematical dependence (3) is presented in
r Fig. 3 and Table I.
φ

2 ⋅ lr 1

Fig. 1. Cylindrical acoustic volume [5] 0.8

1 0.6
Ekaterinoslav Sirakov is with the Department of Communication
Engineering and Technologies, Faculty of Electronics, Technical 0.4
University-Varna, Studentska Street 1, Varna 9010, Bulgaria, E-mail:
katiosirakov@abv.bg 0.2
2
Hristo Zhivomirov is a Ph.D. student with the Department of
Communication Engineering and Technologies, Faculty of 0
0 1 2 3 4 5 6
Electronics, Technical University-Varna, Studentska Street 1, Varna frequency, kHz
9010, Bulgaria, E-mail: hristo_car@abv.bg. Fig. 3. Plot of mode distribution

129
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

TABLE I THE EIGHTEEN LOWEST NORMAL MODES AND THEIR NATURAL

FREQUENCIES FOR A CILINDRICAL BOX WITH RIGID WALLS.

№ mode nz, j, k frequency, kHz

1 j axial 0, 1, 0 1.551
2 j axial 0, 2, 0 2.573
3 z axial 1, 0, 0 2.774
4 z, j tangential 1, 1, 0 3.178
5 k axial 0, 0, 1 3.227
6 j axial 0, 3, 0 3.538
7 z, j tangential 1, 2, 0 3.784
8 z, k tangential 1, 0, 1 4.256
9 j axial 0, 4, 0 4.479
10 j, k tangential 0, 1, 1 4.491
11 z, j tangential 1, 3, 0 4.497
12 z, j tangential 1, 4, 0 5.268
13 z, j, k oblique 1, 1, 1 5.278
14 j axial 0, 5, 0 5.404
15 z axial 2, 0, 0 5.548
16 j, k tangential 0, 2, 1 5.648
17 z, j tangential 2, 1, 0 5.761
18 k axial 0, 0, 2 5.909

II. ENCLOSURE RESPONSE MEASUREMENTS IN

MODEL CLOSED CYLINDRICAL BOX Fig. 4. The measured sound pressure in the closed cylindrical
enclosure (dimensions: height 6.2 cm, diameter 13 cm and wall
thickness 0.1 cm)
The measured characteristics of the sound pressure in
cylindrical enclosure with the application software Realtime
Analyzer [8] are presented in graphical form in Fig. 4. REFERENCES
Measurements were made in a cylindrical box with
dimensions: height 6.2 cm, diameter 13 cm and wall thickness [1] Ekaterinoslav Sirakov, Hristo Zhivomirov, Boris Nikolov,
0.1 cm. The program allows the data from the measured “Green’s Function and Acoustic Standing Waves in Rectangular
values of sound pressure in dB to be stored in tabular and text Loudspeaker Enclosures”, ICEST 2011, Proceedings of Papers,
pp. 721-724, Nis, Serbia, 2011.
format for further analysis.
[2] Екатеринослав С. Сираков, “Собствени резонансни честоти
To examine the modal structure of the enclosure box на правоъгълно озвучително тяло”, Национална
response at the center of the volume was measured. конференция с международно участие „Акустика 1”, Варна,
2009, Списание „Акустика”, год. XI, бр.11, 2009 г., стр. 110-
118.
III. CONCLUSION [3] Екатеринослав Сираков, Борис Николов, Любомир
Камбуров, “Влиянието на съотношението на размерите на
In the cylindrical acoustic volumes as with rectangular [2] правоъгълен акустичен обем върху разпределението на
"axial" and "tangential" natural frequencies can be defined. собствените честоти”, Национална конференция с
The z -axial natural frequency for j = k = 0 is given in Table международно участие „Акустика 1”, Варна, 2010,
Списание „Акустика”, год. XII, бр.12, 2010 г., стр. 94-103.
I, № 3 and № 15. When n z = 0 by analogy with the [4] Philip M. Morse, Vibration and Sound, New York, McGraw-
rectangular speaker enclosure z , φ -tangential natural Hill, 1936.
[5] Екатеринослав Сираков, “Акустичен модел на цилиндрични
frequencies can be defined – Table I, № 10 and № 16. If затворени обеми”, Национална конференция с
n z = 0 and j = 0 the sound is focused along the axis of the международно участие „Акустика 1”, Варна, 2011.
cylinder, the sound wave propagates radially and the natural [6] Frank Fahy, Foundations of Engineering Acoustics, Academic
frequencies are r-axial – Table I, № 5 and № 18. When Press, San Diego, 2005.
[7] István L. Vér, Leo L. Beranek, Noise and Vibration Control
n z = 0 and k = 0 the natural frequencies can be called φ - Engineering: Principles and Applications, Wiley, 2006.
axial [4] (perpendicular to z and r ) as shown in Table I, № [8] http://www.ymec.com/products/dssf3e/
1, 6, 9 and № 14.

130
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Control of Radiation Directivity Applying Independent

Element Dodecahedral Loudspeaker
Marko Jelenković1, Dejan Ćirić2, Jelena Zdravković3 and Stefan Tomić4
Abstract – Control and synthesis of sound source directivity in dodecahedron form is used for radiation directivity control.
have a significant position in modern acoustics. For this purpose, Method for analyzing the radiation directivity is partially
special sound sources that consist of independent element developed in earlier researches [4], [5], but it is modified here
loudspeaker array are used. Sound source analyzed in this paper in order to enable using all twelve channels independently.
is a loudspeaker array in the form of dodecahedron. To generate
different directivities, the loudspeaker array is fed independently
Special attention is paid on establishment of unique, easy to
with a few characteristic sets of signals. Special setup was use, measurement system, and synthesis of basic radiation
established for radiation directivity measurements. For directivity patterns.
analyzing and presenting the results, a MATLAB application is
developed. The measured directivities obtained by feeding of
some of the loudspeakers with particular signals show interesting II. MEASUREMENT SYSTEM
patterns of radiation.
The computer-based measurement system was established
Keywords – Sound source, loudspeaker array, dodecahedron, especially for the purpose of radiation directivity
radiation directivity, swept sine technique. measurement. It consists of a computer (laptop), M-Audio
ProFire 610 audio interface (sound card), M-Audio Fast Track
Pro audio interface, twelve-channel audio amplifier, spherical
I. INTRODUCTION sound source with independent elements in the form of
dodecahedron (Fig. 1), condenser measuring microphone with
Radiation directivity pattern is one of the most important preamplifier, and power supply.
properties of a real sound source [1], [2]. It reflects source’s
interaction with the environment and therefore it is a
significant parameter in various acoustics researches. A sound
source can generally be characterized by three properties:
timbre related to spectral and temporal attributes, intensity,
and directivity providing spatial information on the sound
radiated from the source [3]. Today’s loudspeaker systems can
faithfully reproduce the sound tone (timbre) and intensity, but
they have typical directivities that are significantly different
from the directivities of natural sound sources [4]. Therefore,
there is a need for simulating (synthesizing) directivity
characteristics of the real sound sources [5]. This can be
realized using the three-dimensional sound sources, such as
dodecahedral loudspeaker [6-9].
Another problem with modern sound sources comes from
the requirement for omni-directivity. Sources are typically not
omni-directional in the whole frequency range, but only at
lower frequencies [4], [5]. That gives importance to
researches dealing with radiation directivity control. Fig. 1. Sound source with independent elements (loudspeakers) in
the form of a dodecahedron
In this paper, a sound source based on a loudspeaker array

1
Since the dodecahedral loudspeaker array consists of twelve
Marko Jelenković is a student at the Faculty of Electronic loudspeakers, independent signals were fed via two audio
Engineering at University of Niš, Aleksandra Medvedeva 14, Niš interfaces to every of these twelve channels. Eight channels of
18000, Serbia, E-mail: virtus@elfak.rs.
2 M-Audio ProFire and four of M-Audio Fast Track Pro provide
Dejan Ćirić is with the Faculty of Electronic Engineering at
University of Niš, Aleksandra Medvedeva 14, Niš 18000, Serbia, E- the needed twelve channels, which could be controlled
mail: dejan.ciric@elfak.ni.ac,rs. independently. The diagram of the measurement system is
3
Jelena Zdravković is a student at the Faculty of Electronic shown in Fig. 2. Due to the different characteristics of the
Engineering at University of Niš, Aleksandra Medvedeva 14, Niš interfaces, setting up of the system was more complicated, and
18000, Serbia, E-mail: jelena.z@elfak.rs. a latency problem was inevitable. The problems were resolved
4
Stefan Tomić is a student at the Faculty of Electronic Engineering by applying special calibration method that includes
at University of Niš, Aleksandra Medvedeva 14, Niš 18000, Serbia, measuring of impulse responses for both sound cards
E-mail: stefan@elfak.rs. independently and determination of latency difference

131
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

between two impulse responses. Based on this information, clock applied for the test signal reproduction and recording of
silence of appropriate duration is automatically added before the response during the measurement is not required [12]. In
the measurement signal with smaller latency. In this way, the present research, an excitation swept sine signal of
signals from both sound cards have the same latency. duration of about 24 seconds with frequency range from 10
Hz to 22 kHz sampled at 44.1 kHz is used.
Three excitation combinations are utilized. For the first one,
six loudspeakers from one half of the sphere are fed with the
same sweep signal, and the other six with opposite phase of
the same signal. For the second and the third one, two
opposite loudspeakers on the sphere are fed with
corresponding signals. In one of the configurations,
loudspeakers are fed with the same signals in phase and in the
other one with same signals, but in opposite phases.
The radiation directivity is measured only in horizontal
plane with resolution of 10o, so there are 36 impulse responses
for each excitation combination. Measurements were
performed in a way that the microphone was fixed, and sound
source was rotated around its horizontal axis. Processing of
the extracted part of the impulse response (free of reflections)
is performed in the frequency domain, which is based on FFT
and determination of impulse response spectrum. After
determination of spectrum, next step is determination of
direction of maximal radiation (reference axis), which is used
for normalization and presentation of directivity patterns.
Using this kind of processing, radiation pattern can be
obtained for many different frequencies. Actually, the
maximum number of frequencies (the frequency resolution)
depends on the length of the extracted impulse response
containing only direct sound. Summary of impulse response
processing is given in Fig. 3.

Fig. 2. Measurement system

Also, the measurement system has basic signal control

options implemented in every channel, such as amplitude
regulation and phase inversion. This provides a lot of
combinations that could be used for achieving the wanted
characteristics of radiation.

III. RADIATION DIRECTIVITY MEASUREMENT

The measurements were carried out in pseudo anechoic
conditions. Distance to the first obstacles was greater than 1.5
meters, so difference between direct sound and first reflection
was about 10 ms, which allows us to identify and cut off all
reflections, and analyze only direct sound. Microphone and
sound source were placed on the same height and the distance
between them was 1 meter.
Measurement process is based on swept sine technique
[10], [11]. In our earlier research, it has been shown that this
technique is more reliable than some other tehnuiques (like
MLS technique) [12], [13]. The excitation signal used in this Fig. 3. Impulse response processing and determination of directivity
technique is so called swept sine or sweep. It represents a sine radiation pattern
signal with frequency varying in time, e.g. linearly (linear
sweep) or exponentially (logarithmic sweep). Swept sine It should be kept in mind that processing of extracted parts
technique is considered to be immune to some disturbances of impulse responses, in order to obtain radiation directivity
such as time variance or non-linearities [10], [11]. Also, one patterns, may also be performed in the time domain, as an
of its advantages is that precise synchronization between the alternative to processing in the frequency domain.

132
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

IV. RESULTS measurement configuration, where elements (loudspeakers)

on opposite sides of dodecahedron are fed with opposite phase
For aforementioned configurations, the radiation signals, the radiation patterns at both lower and higher
directivities are obtained using the described procedures of frequencies have the expected bi-directional shape (figure-of-
measurement and processing in the frequency domain. eight), see Fig. 6. From the presented results, it can be seen
Configuration of six independent elements from one half of that the pattern is somewhat wider at lower frequencies (Fig.
dodecahedral loudspeaker array fed with same signal in phase, 6(a)), while it is somewhat narrower at higher frequencies
and the other six with opposite phase signal leads to the (Fig. 6(b)).
expected radiation patterns at lower frequencies, which tend to It should be noted that for some directions, curves
be close to bi-directional (Fig. 4(a)). By increasing frequency, presenting radiation directivity come out of the range of 0 dB.
the patterns begin to lose primary form, and in this case, at This occurs because chosen reference axis is not the axis that
higher frequencies they get a flower shape (Fig. 4(b)). has absolute maximum of radiation at all frequencies. Hence,
regardless the criterion of referent axis choice, there is no
0o0 dB 134 Hz uniform solution that yields the direction of absolute
-30o
30o 402 Hz maximum at all frequencies. Due to that, the described
804 Hz disturbance in presenting the directivity radiation is inevitable.
-10 In all presented figures, the reference axis is positioned in the
-60o 60o direction of 0°.
-20
0o0 dB 201 Hz
o
-30 30o 403 Hz
o o 805 Hz
-90 90
-10

-20 -60o 60o

-20
-120o -10
120o

-90o 90o
o o
-150 0 dB 150
(a) 180o -20

-120o -10
120o
0o0 dB 6166 Hz

-30o 30o 8713 Hz

14745 Hz
-150o 0 dB 150o
-10
(a) 180 o
-60o 60o
-20
0o0 dB 6242 Hz
o o 8659 Hz
-30 30
o o 14700 Hz
-90 90
-10

-20 -60o 60o

-20
-120o -10
120o

-90o 90o
o o
-150 0 dB 150
(b) 180o -20

-120o -10
120o
Fig. 4. Radiation directivity of dodecahedral loudspeaker array with
six elements in phase and six elements in opposite phase at lower (a)
and higher (b) frequencies
-150o 0 dB 150o
(b) 180o
In the second configuration case, with two opposite
elements fed with the same signal in phase, the obtained
radiation characteristics are omni-directional at lower Fig. 5. Radiation directivity of dodecahedral loudspeaker array with
frequencies, while they get shape of narrowed figure-of-eight two opposite elements in phase at lower (a) and higher (b)
pattern at higher frequencies (Fig. 5.). For the last frequencies

133
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

approximately bi-directional patterns. At higher frequencies,

0o0 dB 236 Hz their shape is changed and converges to the particular form.
-30o 30o 472 Hz The developed system for control of radiation directivity
943 Hz together with the system for measurement of this directivity
-10 will be involved in further research related to synthesis of
-60o 60o some specific radiation directivity patterns similar to those of
-20 real sound sources.

ACKNOWLEDGEMENT
-90o 90o

-20
Presented results are obtained within the scope of the
project no. 36026 financed by Ministry of Science and
Technological Development of Republic of Serbia.
-120o -10
120o

REFERENCES
-150o 0 dB 150o
(a) 180o [1] J. H Rindel, F. Otondo, “The interaction between room and
musical instruments studied by multi-channel auralization,” in
Proc. of Forum Acusticum, Budapest, Hungary, 2005.
0o0 dB 6132 Hz
[2] K. Brian FG, D’A. Christophe, “Directivity measurements of
o o 8726 Hz
-30 30 singing voice,” in Proc. 19th International Congress on
14621 Hz
Acoustic, Madrid, Spain, 2007.
-10 [3] O. Warusfel and N. Misdariis, “Directivity synthesis with a 3D
-60o 60o array of loudspeakers application for stage performance,” in
Proc. Conf. Digital Audio Effects (DAFX-01), Limerick,
-20
Ireland, 2001.
[4] B. Stojić, D. Ćirić, M. Marković, “Radiation directivity of
dodecahedral sound source with independent elements”, (in
-90o 90o Serbian), in Proc. LIV ETRAN, paper AK3.5, Donji Milanovac,
Serbia, 2010.
-20 [5] D. Ćirić, “Control of radiation directivity of multi-loudspeaker
sound source”, (in Serbian), in Proc. LIII ETRAN, paper
-120o -10
120o AK3.3, Vrnjačka Banja, Serbia, 2009.
[6] B. Stojić, D. Ćirić, M. Marković, “Radiation directivity of
spherical sound source with independent element dodecahedral
-150o 0 dB 150o loudspeaker array”, in Proc. 1st EAA-EuroRegion, paper S20-
(b) o 11, Ljubljana, Slovenia, 2010.
180
[7] P. Kassakian, D. Wessel, “Caracterization of spherical
loudspeaker arrays,” presented at 11 th AES Convention, San
Fig. 6. Radiation directivity of dodecahedral loudspeaker array with Francisco, USA, 2004.
two opposite elements, one in phase and one in opposite phase on [8] N. Misdariis, O. Warusfel, R. Causse, “Radiation control on a
lower (a) and higher (b) frequencies multi-loudspeaker device,” in Proc. International Symposium on
Musical Acoustic 2001, Perugia, Italy, 2001.
[9] N. Misdariis, F. Nicolas, O. Warusfel, R. Causse, “Radiation
V. CONCLUSION control on multi-loudspeaker device: La Timée,” in Proc.
International Computer Music Conference, Havana, Cuba, 2001.
[10] A. Farina, “Simultaneous measurement of impulse response and
In this paper, radiation directivity of spherical sound source
distortion with a swept-sine technique,” presented at 108th
with independent elements in the form of dodecahedron is Convention Audio Eng. Soc., abstract in J. Audio Eng. Soc.,
analyzed. The presented results and up to now experience vol. 48, no. 4, p. 350, 2000.
show that using develop procedure of measurement, radiation [11] S. Müller and P. Massarani, “Transfer-Function Measurement
directivity characteristics of sound source could be obtained in with Sweeps,” J. Audio Eng. Soc., vol. 49, no. 6, pp. 443–471,
an adequate way. Measurement system along with the 2001.
developed software support and measuring and processing [12] M. Ličanin, A. Đorđević, M. Jelenković, Mentor: D. Ćirić,
procedures are well established and provides sufficient “Sensitivity of impulse response measurements with maximum
flexibility and control. length sequences and sweeps”, in Proc XLVI ICEST, pp. 567-
570, Niš, Serbia, 2011.
For the observed configuration cases, applying various
[13] M. Jelenković, D. Ćirić, M. Ličanin, A. Đorđević,
combinations of excitations and disposition of elements “Repeatability of room impulse response measurements”, (in
(loudspeakers), some specific radiation directivity patterns can Serbian), in Proc. LV ETRAN, paper AK2.7, Banja Vrućica,
be achieved. Most of these patterns have the expected forms at Bosnia and Herzegovina, 2011.
lower frequencies - one omni-directional, and two

134
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Modulated bandpass Farrow Decimators and Interpolators

Djordje Babic1 and Vesa Lehtinen 2
Abstract – The Farrow structure provides a flexible way for without increasing the polynomial degree of the lowpass
adjustable filtering, including adjustable fractional-delay filters prototype Farrow filter. The modulating function is
and sample rate conversion (SRC) by arbitrary, non-integer, constructed as a low-order polynomial in order to avoid costly
factors. Recent publications have pointed out its suitability to generation of trigonometric functions.
bandpass SRC. When the passband centre is located above the
input sample rate, the interpolator complexity becomes
approximately proportional to the centre frequency. We propose II. POLYNOMIAL MODULATION
a new construct, the modulated Farrow structure, which allows
arbitrarily high centre frequencies without increasing the
polynomial degree of the prototype Farrow filter. The
Bandpass filters can be implemented with lowpass filters as
modulating function is constructed as a low-order polynomial in building blocks by applying frequency translation to the signal
order to avoid costly generation of trigonometric functions. before and after filtering which is equivalent to multiplying
the impulse response by cosine whose frequency is equal to
Keywords – Digital Filters, Farrow Structure, Decimators, the center frequency of the bandpass filter [4]. While this
Interpolators, Band-pass Filtering. permits arbitrary and variable centre frequencies, generation
of costly cosine function is needed, this increases the
implementation complexity. The same principle can be
applied to the case of polynomial-based filter, as well.
I. INTRODUCTION In order to minimize the implementation complexity, it may
be beneficial to approximate the modulating function with a
Discrete-time filter structures are used in many applications piecewise polynomial. The modulating polynomial is chosen
to interpolate new sample values at arbitrary points between first, and the modulation is taken into account in the optimized
the discrete-time input samples [1], [2]. In those cases, it is subfilter coefficients. The non-idealities of the modulating
beneficial to use polynomial-based interpolation filters. The function can be compensated by the filter. This eliminates the
advantage of the above system lies in the fact that the actual need for costly (co)sine generation.
implementation can be efficiently performed by using the Using Taylor polynomials a function can be approximated
Farrow structure [1] or its modifications [2], [3]. In [2], [3], as closely as desired by a polynomial provided that the
several design methods for polynomial-based interpolation function posses sufficient number of derivatives [7], [8].
filters have been developed in the time and in frequency Based on the definition, the modulating cosine function
domain. Using the frequency domain approach, it is possible cos(ω0t) can be approximated with following polynomial:
to design the polynomial-based filter realized in the form of
Farrow structure with an arbitrary baseband (zero center j=Ja  (ω t ) 2 j (ω t ) 2 ( j +1) 
frequency) frequency response. cos(ω0t ) = ∑ (−1) j cos(ω0 a ) 0 − sin(ω0 a ) 0  . (1)
However, there are some applications where there is a need
j =0
 (2 j )! (2 j + 1)! 
for a bandpass interpolation. Some example applications are:
bandpass sampling, bandpass sampling rate conversion, The above equation can be expressed in more suitable way
rational filter banks, etc [4], [5]. It has been shown in [6], that for practical realization:
if the center frequency of the bandpass interpolation filter is Ma

smaller than the half of the sample rate, for the given
passband and stop-band requirements, the filter order does not
cos(ω0t ) = ∑g t
m =0
m
m
, (2)

increase compared to the baseband filter with the same where

requirements. The design methods presented in [2] and [3] can  (ω0 ) m
be effectively used for the design of bandpass polynomial  ( −1) m/2
cos( ω 0
a ) for m even, and
(m)!
interpolation filters with minor modifications [6]. g m=  (3)
− (−1) ( m −1) / 2 sin(ω0 a) (ω0 ) for m odd,
m

However, when the center frequency is (far) higher than the

input sample rate, the polynomial degree would become  (m)!
approximately proportional to the centre frequency.
Therefore, we propose a new construct, the modulated Farrow Ma=2*Ja+1 is a polynomial degree which is related to the
structure, which allows arbitrarily high centre frequencies approximation error, and t∈[0,T). Here, T is the input
sampling interval Tin in the interpolation case, and output
sampling interval Tout in decimation case. Thus, we have
1
Djordje Babic is with the School of Computing at Union obtained a polynomial that approximates the cosine function
University, Belgrade, Serbia, E-mail: djbabic@raf.edu.rs. over one polynomial segment. It is possible to generalize the
2
Vesa Lehitnen is with the Department of Communications approximation of the cosine over a longer interval by making
Engineering at Tampere University of Technology, Tampere, it a piecewise having N polynomial segments, and in each
Finland, E-mail: vesa.lehtinen.tut.fi. segment the order of the polynomial is Ma. The piecewise

135
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

polynomial approximation of the cosine can be written as: Here ηl(n) is a modulating term. There are altogether N
Ma
modulating terms ηl(n), one for each polynomial segment.
∑∑ b (n)t
N

cos(ω0t ) = m
, (4) Equations (12) and (13) can be used to define a bandpass
m
n =0 m =0 modulated modified Farrow structure, which is shown in Fig.
where 1. The structure can be used for different passband central
frequencies, with a single prototype filter.
 (ω0 ) m
(−1) cos(ω 0 a (n)) (m)! for m even, and
m/2

bm =  (5)
 − (−1) ( m −1) / 2 sin(ω0 a(n)) (ω 0 ) for m odd.
m III. SPECIAL CASES OF MODULATED FARROW
 (m)! STRUCTURE
Here, a(n) determines the initial phase of the approximated
cosine function in each polynomial segment. The value of There are several special cases of polynomial modulation,
a(n) can be selected in such a manner that overall in which the process of building the bandpass Farrow
approximated cosine function is symmetric around N/2, structure can be further simplified. These special cases are
a(n)=n+0.5, for n=0, 1…N-1. In this way, the impulse derived from mutual relation between the desired central
response of modulated lowpass prototype filter is also frequency f0= ω0/2π and the sample frequency of the bandpass
symmetric. In the actual implementation, the symmetry can be filter Fs=1/T. If the central frequency is an integral multiple of
exploited in a similar way as for polynomial-based filters, the the sample frequency, integral multiple of the half or quarter
polynomial in t can be transformed to polynomial in (2t-1) of the sample frequency then the modulating term ηl(n) has
thus we obtain special values as shown in a sequel.
The most straightforward case of modulated Farrow
Ma Ma
structures is that with the modulating frequency of kFs. All
∑∑ b (n)(2t − 1) = ∑∑ b (n) f
N N

cos(ω 0t ) = m
m
m m
(n, T , t ) . (6) that is needed is to multiply the output of the lowpass Farrow
n =0 m=0 n=0 m =0
structure with a cosine at kFs, i.e., when the modulation
The modulated impulse response of the polynomial based frequency is an integral multiple of Fs, the cosine generation
filter is obtained by multiplying the impulse response of the is simplified by:
lowpass prototype filter by cosine:
 (ω ) m
(−1) m / 2 (−1) k 0 for m even, and
hm (t ) = ha (t ) cos(ω0t ), (7) bm =  (m)! (13)
yielding to: 0 for m odd.
M Ma

∑∑ ∑
N

hm (t ) =  c m
( n ) f m
( n, T , t ) bm1 (n) f m1 (n, T , t )] . (8) Therefore, bm depends only on polynomial order m, and
n=0  m =0 m1 =0 integral multiple k, reducing (12) to
Ma
The modulated impulse response is a piecewise polynomial η l = ∑ bm (2 µ l − 1) m . (14)
with the same number of polynomial segments as the lowpass m =0

prototype filter and polynomial order of M+Ma in each

segment. The polynomial order increases as the central Fig. 2. depicts the modulated Farrow structure for modulation
frequency of the passband increases. Furthermore, the frequencies kFs. Impulse response modulation at kFs is
obtained bandpass filter is not configurable, as the new set of attained simply by implementing modulation of Farrow
coefficients is calculated for different value of passband structure by ηl.
central frequency. When the modulation frequency is an odd multiple of Fs/2,
If we calculate the output sample according to hybrid the modulating function has the same shape in every
analogue/digital model, the lth output sample can be polynomial segment, but its sign alternates between adjacent
expressed, after some manipulations, in the following form segments. The calculation of the modulating term reduces to
[2], [3]:
x(n l + N/2) η l(0)
N −1
M N Ma
 cM(0) c1(0) c0(0)
y(l ) = ∑  ∑ x(nl − n + )cm (n)(2µl − 1) m ∑ bm1 ( n)(2µl − 1) m1  (9) Fin
n =0  m =0 2 m1=0  η l(1)
−1 −1 −1
Z Z Z
cM(1) c1 (1) c0(1)
where
nl = lTout Tin  and µl = lTout Tin − nl . (10) −1 −1 −1
Z Z Z

Alternatively, the lth output sample can be expressed as:

η l(N-1)

M N cM(N − 1) c1(N − 1) c0(N − 1)

y (l ) = ∑ ∑ η l (n)x(nl − n + N / 2)c m (n)(2µ l − 1) m , (11)
m = 0 n =0
vM (nl) v1(n l) v0(nl)
Fout
with
Ma 2µ l − 1 y(l)
η l (n) = ∑ bm (n)(2 µ l − 1) m . (12)
m =0
Fig. 1. The modulated modified Farrow structure

136
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

x(n) Lowpass y(l) x(nl + N/2) (-1)k

Farrow
interpolator cM(0) c1(0) c0(0)
ηl Fin
(-1)k+1
Fig. 2. The modulated Farrow structure for center frequencies kFs. −1
Z
−1
Z
−1
Z
cM(1) c1 (1) c0(1)

0 for m even, and

 (ω ) m
−1 −1 −1

b m ( n) = 
Z Z Z
(15)
− (−1) ( m−1) / 2 (−1) n (−1) k 0 for m odd
 (m)! (-1) k+N-1

cM(N − 1) c1(N − 1) c0(N − 1)

and, thus
Ma
ηl
η l (n) = (−1) n ∑ bm (2µ l − 1) m . (16) vM (nl) v1(n l) v0(nl)
Fout
m =0

2µl − 1 y(l)
Fig. 3. illustrates the modulated Farrow structure for
modulation frequencies (2k+1) Fs/2. As above, modulation at Fig. 3. The modulated Farrow structure for center frequencies
(2k+1)Fs/2.
(2k+1)Fs/2 is attained simply by multiplying by ηl(n). and by
alternating the sign of coefficients cm(n) for odd n.
 ( 2 k +1) n −1 k −1
π (ω ) m
(−1) (−1) (−1) 2 cos( ) 0
m/2
Let us consider modulating the impulse response by a 2
for m even
4 (m)! (23)
cosine at the frequency (2k+1)Fs/4. In each polynomial b m ( n) = 
− (−1) ( m−1) / 2 (−1) 2 (−1) 2 sin( π ) (ω0 )
( 2 k +1) n −1 k −1 m

segment, the modulating function has a similar shape:(i) for m odd.

between the nth and (n+2)th segment, the shape is the same  4 (m)!
but the signs are opposite; (ii) adjacent segments are time-
It is possible to see from (20)-(23), that the values of
domain mirror images of each other with the same or opposite
coefficients bm(n) in this case depend only on the polynomial
signs. The modulating functions can be written as
order m, and sign is determined by n, and k. The main
 1 π (ω0 ) m consequence is that all bm(n)s are the same for given m, while
 (−1) cos((2k + 1)(n + 2 ) 2 ) (m)! ,
m/2
for m even, the sign alternates according to (20)-(23). In order to reduce
b m ( n) =  (17) the number of operations needed for modulation, it is
− ( −1) ( m −1) / 2 sin((2k + 1)( n + 1 ) π ) (ω0 ) , for m odd.
m


beneficial to decompose the mth-degree subfilter into two
2 2 ( m)!
polyphase branches:
After applying series of trigonometric identities, the
coefficients bm(n) are reduced to: c m 0 ( n) = c m ( 2n)
(24)
cm1 (n) = cm (2n + 1).
 m ( 2 k +1) n−1
π (ω ) m

 ( −1) (−1) sin(( 2k + 1) ) 0 , for m even,

2 2

4 (m)! In this way, two interleaved Farrow structures are obtained:

b m (n) =  (18)
( −1) 2 (−1) 2 cos(( 2k + 1) π ) (ω0 ) , for m odd,
m +1 ( 2 k +1) n−1 m

 4 ( m)! C m ( z ) = C m 0 ( z ) + z −1C m1 ( z ). (25)

for odd n, and In each interleaved Farrow structure the sign of coefficients is
 m ( 2 k +1) n
π (ω ) m determined by n and k, and the output is multiplied by
(−1) (−1) cos((2k + 1) ) 0
2 2
for m even,
b m ( n) = 
4 (m)! (19) corresponding value of ηl0 or ηl1. Fig. 4. shows the modulated
m+1
(−1) (−1)
( 2 k +1) n
π (ω ) m Farrow structure for modulation frequencies (2k+1)Fs/4.
2 2
sin((2k + 1) ) 0 for m odd.
 4 (m)! It is possible to see from (20)-(23), that the values of
coefficients bm(n) in this case depend only on the polynomial
for even n. These coefficients can be further simplified based order m, and sign is determined by n, and k. The main
on the value of modulating integer k as shown in a sequel: for consequence is that all bm(n)s are the same for given m, while
n even, k even in (20); for n even, k odd in (21); for n odd, k the sign alternates according to (20)-(23). In order to reduce
even in (22); for n odd, k odd in (23); the number of operations needed for modulation, it is
 ( 2 k +1) n k
π (ω0 ) m beneficial to decompose the mth-degree subfilter into two
(−1) (−1) (−1) cos( )
m/ 2 2 2
for m even,
4 (m)! (20) polyphase branches:
b m (n ) = 
− (−1)( m−1) / 2 (−1) 2 (−1) 2 sin( π ) (ω0 )
( 2 k +1) n k m

for m odd. c m 0 ( n) = c m ( 2n)

 4 (m)! . (26)
cm1 (n) = cm (2n + 1).
 ( 2 k +1) n k −1
π (ω ) m
(−1) (−1) (−1) 2 sin( ) 0
m/2 2
for m even,
4 (m)! (21) In this way, two interleaved Farrow structures are obtained:
b m ( n) = 
− (−1)
( 2 k +1) n k −1
π (ω ) m
( m −1) / 2
(−1) 2
(−1) cos( ) 0
2
for m odd.
 4 (m)! C m ( z ) = C m 0 ( z ) + z −1C m1 ( z ). . (27)
 ( 2 k +1) n −1
( −1) π (ω )
k m

(−1) (−1) (−1) sin( ) 0

m/2 2 2
for m even, In each interleaved Farrow structure the sign of coefficients is
4 (m)! (22)
b m ( n) =  determined by n and k, and the output is multiplied by
− (−1)
( 2 k +1) n −1 k
π (ω ) m
( m −1) / 2
(−1) 2 (−1) 2 cos( ) 0 for m odd. corresponding value of ηl0 or ηl1. Fig. 4. shows the modulated
 4 (m)! Farrow structure for modulation frequencies (2k+1)Fs/4.

137
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Arrange sign
x(nl) based on n, and k

CM(z2) C2(z2) C1(z2) C0(z2)

z-1
0

vM(nl) v2(nl) v1(nl) v0(nl) y(l)

2µl-1
ηl 0
Arrange sign
based on n, and k
x(nl)

CM(z2) C2(z2) C1(z2) C0(z2)

1
Fig. 5. The frequency response of the lowpass prototype.
vM(nl) v2(nl) v1(nl) v0(nl)

2µl-1
ηl 1

Fig. 4. The modulated Farrow structure for center frequencies

(2k+1)Fs/4.

IV. DESIGN EXAMPLES

In order to illustrate the effectiveness of the proposed Fig. 6. The frequency response of the filter modulated by 3Fs/4.
modulation method for the design of bandpass Farrow
structure we use several illustrative examples. The first step is
the design of the lowpass prototype filter with passband edge
fp=0.3Fs, stopband edge fs=0.5Fs, required stopband
attenuation As=60dB, and passband tolerance δp=0.1. The
filter, with performance shown in Fig 5., is designed using
minimax design of [2], having N=14 polynomial segments,
and polynomial degree M=4.
The next step is to build the bandpass filter by modulating Fig. 7. The frequency response of the filter modulated by 2Fs.
the obtained lowpass prototype. We use two example cases,
with central frequencies of 3Fs/4 and 2Fs. The corresponding Optimization of Energy-efficient Computer and
polynomial-based modulation function that approximates Communications Systems.”
cosine has the same length N=14, and polynomial order Ma=7
for 3Fs/4, and Ma=19 for 2Fs. According to Section 3, there REFERENCES
are 10 additional multiplications to produce modulation by
frequency 3Fs/4, and also 10 to produce modulation by [1] C.W. Farrow, “A Continuously Variable Digital Delay
frequency 2Fs. Frequency domain performances of modulated Element,” IEEE Int. Symp. Circ. Syst., pp. 2641–2645, Espoo,
bandpass filters are shown in Figs. 5 and 7 respectively. Finland, 1988.
Though there is a slight degradation of performance in [2] J. Vesma and T. Saramäki, “Polynomial-based interpolation
stopband, the filtering requirements can be met by over Filters - Part I: Filter synthesis,” Circuits, Systems, and Signal
designing the lowpass prototype filter. Processing, vol. 26, no. 2, pp. 115-146, March/April 2007.
[3] D. Babic, T. Saramäki, M. Renfors, “Conversion between
arbitrary sampling rates using polynomial-based interpolation
V. CONCLUSIONS filters,” in Proc. 2nd Int. TICSP Workshop on Spectral Methods
and Multirate Signal Processing SMMSP’02, pp. 57-64,
We have presented a modulated Farrow structure, which Toulouse, France, 2002.
allows band-pass realization with arbitrarily high centre [4] R. Zukunft, S. Haar, T. Magesacher, “Digital interpolation in
the passband domain,” Proc. IEEE Int. Conf. Acoust. Speech
frequencies without increasing the polynomial degree of the
Signal Proc. (ICASSP), pp. 1545–1548, Orlando, USA, 2002.
lowpass prototype Farrow filter. The modulating function is [5] H. Johansson, “Farrow-structure-based reconfigurable bandpass
constructed as a low-order polynomial in order to avoid costly linear-phase FIR filters for integer sampling rate conversion,”
generation of trigonometric functions. The non-idealities of IEEE Trans. on Circuits and Systems II: Express Briefs, (58), 1,
the modulating function are then mitigated by taking them pp. 46-50, 2011.
into account when optimizing the filter coefficients. [6] D. Babic, “Polynomial-based filters in bandpass interpolation
and sampling rate conversion,” Proc. Int. Workshop on Spectral
Methods and Multirate Signal Processing (SMMSP), pp. 31–37,
ACKNOWLEDGEMENTS Florence, Italy, 2006.
[7] Greenberg, Michael, Advanced Engineering Mathematics (2nd
This work was supported by the Serbian Ministry of ed.), Prentice Hall, ISBN 0-13-321431-1, 1998.
Science under technology development projects: TR32028 – [8] D. Babic, “Piecewise Polynomial Approximation Based on
“Advanced Techniques for Efficient Use of Spectrum in Taylor Series with Efficient Realization using Farrow
Structure,” the 9th Int. Conference TELSIKS 2009, pp. 241-
Wireless Systems” and TR32023 – “Performance
244, Niš, Serbia.

138
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Simulation of Codec for Adaptive Linear Prediction

Rumen P. Mironov1

Abstract – The software model of Adaptive Linear Prediction average component and correlation function Rx ( r ) for:
Codec for one dimensional signals is developed. The presented
codec is simulated on Simulink for Matlab 6.5 environment and r = 0 , n − 1 . Then the basic equation of the linear prediction
the obtained results for real sound signals are evaluated by the will be presented by the following way:
calculation of MSE and SNR for the decoded signals. n−1
Keywords – Digital Signal Processing, Linear Prediction,
Adaptive Data Compression, Matlab Simulation.
x̂( i ) = a1 x( i − 1 ) + ........ + an−1 x( i − n + 1 ) = ∑ a x( i − k ) ,
k =1
k (1)

where x̂( i ) is the value of the predicted element from the

I. INTRODUCTION input signal x( i ) . The prediction error is described by the
equation:
The basic methods for digital signal compression are e( i ) = x( i ) − x̂( i ) , (2)
lossless and lossy compression. Due to the low coefficient of and the quantization error - by the equation:
information reduction, the first group of methods is used most
eq ( i ) = Q[e( i )] . (3)
frequently in archiving systems and didn’t find widespread
use in the information transmission systems. The methods for According to the LMS algorithm of Widrow [6], the weight
lossy compression are divided into three groups: statistical, coefficients ak of the prediction filter can be determined
psychoacoustic and transforms methods ([1], [2], [3] and [4]). recursively:
The psychoacoustic compression methods are based on the ∂e2( i )
shortcomings of the human hearing system and are suitable ak ( i ) = ak ( i − 1 ) − µ .∇ k = ak ( i − 1 ) − µ , (4)
∂ ak
only when the recipient of the record signals is a man. The
transformation methods provide a high degree of information where: ∇ k is the gradient of the squared error by the
reduction and work mainly on the basis of unitary prediction and µ is the adaptation step for the i-th step of
transformations ([3], [4]), for which after decompression the adaptation.
specific block distortions are received. Statistical compression The derivative of the prediction error is determined by the
methods are based on the reduction of the information Eqs. (1), (2), (3), and (4). The sequence for the components of
redundancy of the transmitted signals and proceeds in two weights is given by:
stages: decorrelation of communication signals and reduction
of binary digits, necessary for transmition of signals ([1], [2]). ak ( i ) = ak ( i − 1 ) + 2µ e( i )x( i − k ) . (5)
One of the simplest and most convenient for practical
implementation methods for decorrelation is the method for The beginning values of the weighted coefficients can be
linear prediction ([2], [5]). calculated after the minimization of mean square error of the
An adaptive method for coding of one dimensional digital prediction. From Eqs. (1) and (2) follows:
 2

[x( i ) − x̂( i )] = E  x( i ) − ak x( i − k )  , (6)
signals, based on the linear prediction and Least Mean Square 1 N −1 n −1

(LMS) weight coefficients adaptation [6] of the prediction e2( i ) =

N i =0
∑ ∑
filter is presented. From the developed mathematical  k =1  
equations a general block scheme of Line Prediction Codec where E is the averaging operator.
(LPC) is synthesized and experimental results from the
The partial derivative of e 2 ( i ) with respect to any weight
simulation by Simulink for Matlab 6.5 environment for test
signals are given. coefficient al can be expressed as:
{
∂ e 2 ( i ) ∂E [x( i ) − (a1 x( i − 1 ) + .....an−1 x( i − n + 1 ))]
=
2
}
II. MATHEMATICAL DESCRIPTION ∂ ak ∂ al .
= −2 E{[x( i ) − (a1 x( i − 1 ) + .....an−1 x( i − n + 1 ))]x( i − l )}
We will assume that the correlation covers n neighborhood
Once equated to zero and transformed, from the upper
elements of the input digital signal, represented by the expression is obtained:
stationary series {x( i )} with N > n values, which have zero n −1
E{x( i )x( i − l )} = ∑ ak E{x( i − k )x( i − l )} . (7)
k =1

1 The autocorrelation function of the digital signal, presented

Rumen P. Mironov is with the Faculty of Telecommunications,
Technical University of Sofia, Boul. Kl. Ohridsky 8, Sofia 1000, with the series {x( i )} is:
Bulgaria, E-mail: rmironov@tu-sofia.bg

139
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

1 N −1 for autocorrelation functions Rx ( 0 ), Rx ( 1 ), Rx ( 2 ) , calculated

∑ x( i )x( i − r ) = E{x( i )x( i − r )} . (8)
Rx ( r ) =
N i =0 from the Eq.(8). The starting weight vector
r
From the Eqs. (7) and (8) the following expression is A( 0 ) = [a1( 0 ), a2 ( 0 )]t is calculated from the equation:
obtained:
[Rx ].A( 0 ) = 
Rx ( 1 ) 
n −1
,
Rx ( l ) = ∑ ak Rx ( k − l ) . (9)  Rx ( 2 )
k =1
∆1 ∆
For l = 1, n − 1 equation (9) represent a linear system with by the Kramer’s formulas: а1( 0 ) = , а2 ( 0 ) = 2 .
∆ ∆
n − 1 unknowns. Using Eq.(5) the values of the weight vector for the next
The following equations are used for decoding of signals: iterations are calculated by the following:
x′( i ) = e′( i ) + ˆx′( i ) , (10)  а1( i )   а1( i − 1 )   x( i − 1 ) 
 =  + µ .e( i ). ,
a2 ( i ) a2 ( i − 1 )  x( i − 2 )
[ ]
e′( i ) = Q −1 eq ( i ) . (11) where the errors are calculated from the Eq.(2) and the
The synthesized by the Eqs. from (1) to (11) adaptive line predicted values are calculated from:
prediction codec (ADPCM) from 2nd order ( k = 2 and x̂( i ) = a1 ( i − 1 ).x( i − 1 ) + a2 ( i − 1 ).x( i − 2 ).
provided that there is no quantization unit) is shown on Fig. 1. The adaptation step depends on the inequality:
On Fig. 2 and Fig.3 the synthesized coder and decoder blocks 1
0<µ < and the value µ=0.01 is chosen.
for ADPCM are shown respectively. The presented schemes Rx ( 0 )
are developed through the Simulink package for Matlab 6.5 The mean squared error is calculated by the equation:
environment and included the following units:
1 N
ε 2 = .∑ e 2 ( i ).
– input unit; N i =1
– output unit;

- input unit from wave file;

- output unit to wave file;

- input unit for buffered reading from file;

- multiplication unit;

- delay unit for one cycle;

- addition/subtraction unit;

- oscilloscope (GUI output); Fig. 4. Input signal x( i )

- multiplexer ( 2 ,3,...n inputs, vector output);

- S-function (user defined function in

Simulink).

III. Experimental Results

The developed ADPCM codec is used for simulation on
Matlab 6.5 environment of real audio signals (WAV file
format, 1 channel (mono), 16 bits, 44.1 KHz sampling rate).
The starting values of the weights are obtained from the
correlation matrix:

[Rx ] =  x
R ( 0 ) Rx ( 1 ) 

 Rx ( 1 ) Rx ( 0 )
Fig. 5. Output predicted signal x̂( i ) .

140
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The input test signal and output predicted signal are shown On Fig.8 the input signal and the error signal are visualized
on Fig.4 and Fig.5 respectively. On Fig.6 the input and output together.
signals are shown together (visualized from the oscilloscope
Scope 4). On Fig.7 the same signals are visualized by the
zooming in horizontal direction (from 0.031s to 0.0324s).
IV. CONCLUSION
An adaptive method for coding of one-dimensional digital
signals, based on the linear prediction and Least Mean Square
(LMS) weight coefficients adaptation of the prediction filter is
presented. From the developed mathematical equations an
algorithm and a general block scheme of adaptive linear
prediction codec is synthesized and experimental results from
the simulation by Simulink for Matlab 6.5 environment for
test signals in WAV format are given
The developed ADPCM codec provides minimum
processing error and lied to increase of PSNR with about 0.3
dB in comparison with 2 coefficients non-adaptive prediction
codec.
The presented simulation model can be used in digital
signal processing for spectral analysis, coding and
transmission of one-dimensional signals and in distance
Fig. 6. Input and output predicted signal MUX{ x( i ) , x̂( i ) }. learning by the using a Matlab Web Server.
The developed image codec is used in laboratory work on
the disciplines: "Image and Signal Processing" and "Audio
and Video Communication on Internet" and in the
experimental work in laboratory "Electronic System for
Visual Information" in Technical University of Sofia.

V. ACKNOWLEDGEMENT
The authors thank the National Fund for Scientific
Research of the Bulgarian Ministry of Education and Science
for the financial support by the contract VU-I-305/2007.

VI. REFERENCES
[1] D. Stranneby, W. Walker, Digital signal processing and
applications, Elsevier, 2004.
Fig. 7. Input and output signal MUX{ x( i ) , x̂( i ) } (zoomed). [2] S. Mitra, Digital signal processing. A computer-based
approach, Mc Graw Hill, 2006.
[3] V. Madisetti, D. Williams, Eds. The Digital Signal Processing
Handbook, CRC Press, 1998.
[4] A. D. Poularikas. The Transforms and Applications Handbook,
Second Ed., CRC Press, 2000.
[5] G. Blanchet, M. Charbit, Digital Signal and Image Processing
Using MATLAB, ISTE, 2006.
[6] B. Widrow, S. D. Stearns. Adaptive Signal Processing.
Engelewood Cliffs, New York, Prentice-Hall, Inc., 1985.

Fig. 8. Input and error signal MUX{ x( i ) , e( i ) }.

141
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 1. ADPCM Codec.

Fig. 2. ADPCM Coder.

Fig. 3. ADPCM Decoder.

142
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Image Compression with Inverse Pyramid

Decomposition over Wavelet Spectrum
Teodora G. Sechkova1 and Ivo R. Draganov2
Abstract – In this paper a novel approach is presented for of the image based on partial information transmitted in terms
compression of digital images. It consists of finding the wavelet of bit-planes. Despite all the novelties in this approach at the
spectrum of an image into certain number of sub-bands after level of visual objects inside the image some real multi-level
given levels of regular or irregular decomposition. Then each scheme is thought to be proper for further enhancement.
sub-band is decomposed with the inverse pyramid algorithm
Recently a new approach has been proposed for digital
using some linear orthogonal transform such as DCT. Depending
on whether a lossless or lossy compression is desired all or only images decomposition denoted as inverse pyramid [6]. It
some of the spectral coefficients from the inverse pyramid are introduces decomposing levels for a single image by dividing
preserved and then entropy coding is applied. Higher it in smaller blocks over which linear orthogonal transform of
compression ratios are achieved at high image quality levels any kind could be applied. Then restoring of the blocks in the
compared to some popular algorithms from the practice. next level is done by using only some of the spectrum
coefficients while the difference between the original and the
Keywords – Image Compression, Wavelet, IPD, DCT. approximated blocks is being preserved and passed to the next
level. This particular decomposition is thought to be very
I. INTRODUCTION efficient and flexible for digital images compression as well as
in the field of pattern recognition for simplified object
During the last three decades a large number of algorithms representation and speeding up algorithms. It is in the base of
were designed for digital images compression. Many methods the proposed here algorithm along with the advantages given
rely on spectral decorrelation combined with entropy coding by the wavelet transforms.
[1]. Historically one of the most popular approaches that had In the next section detailed description of the new algorithm
become a standard and is still in wide use is the JPEG is presented, then in part three some experimental data is
compression algorithm [2]. It can achieve extremely high given and then conclusion is made in part four.
compression ratios (over 100 times) at relatively high peak
signal-to-noise ratios (PSNR, in some cases above 30 dB). II. ALGORITHM DESCRIPTION
Nevertheless some serious disadvantages could be pointed out
for it such as the block effect due to the localizing property of The input image I is grayscale consisting of N by N pixels
the discrete cosine transform (DCT) and the lack of any and intensity range from 0 to 255. It is transformed by using
scaling ability apart from the group change of the quantization the Haar wavelet according to:
matrix coefficients in some more flexible fashion, e.g. some
multilevel or multistage processing. Extension like that could
provide even more opportunities for scalable progressive B2 x 2 = TI 2 x 2T T , (1)
transmission of image data over narrowband communication
channels such as far distance satellite links for planetary where I2x2 is a block of the image with size 2×2 and T – the
observation. Haar transform matrix of the same size given by:
Another large group of algorithms for image compression is
based on the spatial-scale decomposition dividing the 1 1 1 
information for large and small objects from the image into T=  . (2)
separate bands. The wavelet transforms [3, 4] are the essence 2 1 − 1
of such approaches and naturally led to the well-known JPEG
2000 standard [5] in which the block effect is absent. Here Going through all 2×2 blocks of the image and rearranging
some more advanced techniques were introduced at the level the resulting spectral coefficients from the resulting spectrum
of information redundancy reduction such as the Embedded B into 4 sub-bands the complete 1-level Haar decomposition
Zero-Tree Wavelet (EZW) coding and the Embedded Block is achieved.
Coding with Optimal Truncation (EBCOT). Another Then each sub-band of the wavelet spectrum BR (R = 1÷4)
advantage is the ability for scalable transmission (restoration) is divided to blocks with dimensions 2n×2n and each of them
is presented with Inverse Pyramidal Decomposition (IPD) [6]:
1
Teodora G. Sechkova is with the Faculty of Telecommunications,
~
8 Kliment Ohridski Blvd., 1000 Sofia, Bulgaria, E-mail: [BR ( 2 n )] = [B0 R ( 2 n )] +
teodora.sechkova@gmail.com
2 2
Ivo R. Draganov is with the Faculty of Telecommunications, 8 ~
Kliment Ohridski Blvd., 1000 Sofia, Bulgaria, E-mail: + ∑ [E p −1,R( 2 n )] + [E 2 ,R( 2 n )], (3)
p =1
idraganov@tu-sofia.bg

143
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

where 3 levels are used and [E 2 ,R( 2 n )] is the matrix of the k

Here s pRp (u,v) are elements of the transformed block
residual from the decomposition. In (3) each matrix is with
[S pRp ( 2 n− p )] which is obtained also by the 2D-DCT:
k

dimensions 2n×2n. The first component [B~0 R ( 2 n )] for the

level p = 0 is a rough approximation of the block [BR(2n)]. It is
[S pRp ( 2 n− p )] = [T p( 2 n − p )][E p −p 1,R( 2 n − p )][T p( 2 n − p )],
k k
obtained by applying inverse 2D-DCT over the transformed (9)
block in correspondence with the expression:
~ where [T p( 2 n − p )] is a matrix with dimensions 2n-p×2n-p for
~
[B0 R( 2 n )] = [T0( 2 n )] −1 [S 0 R( 2 n )][T0( 2 n )] −1 , (4) level р = 0 by which DCT is applied.
It is possible to represent each group of four neighbouring
where [T0( 2 n )] −1 is a matrix with dimensions 2n×2n for the k
elements ~
s pRp (u,v) for one and the same u and v in the way
inverse 2D-DCT.
defined by (10) which allows to gain even higher correlation
The matrix is the transform block of the cut 2D-DCT over
between the spectral coefficients since the last three ones for
[BR(2n)]. Here m0(u,v) are the elements of the binary matrix-
positions (0,1), (1,0) and (1,1) form differences two by two
mask [M0(2n)] with the help of which the preserved
and these differences often are zero valued because
coefficients are being determined [S~0 R ( 2 n )] in accordance to
neighboring blocks contain almost identical content.:
the equation:
d~ k p ( 0 ,0 ) ~ k p 
1, if s0 R(u,v) is retained coefficient,  pR   1 1 1   s pR ( 0 ,0 )
1
m0( u,v )=  d~ k p ( 0 ,1 )  1  0 4 0 − 4  ~s p ( 0 ,1 ) 
k
 0 - otherwise, (5)  ~pR =   pR  (10)
( 1,0 )  4  − 4 0  ~
kp k
 d pR 0 4 s pRp ( 1,0 ) 
 ~k   
for u,v = 0,1,.., 2n-1.  d pRp ( 1,1 )   0 0 − 4 4   ~ k p 

   s pR ( 1 ,
1 ) 
The values of the elements m0(u,v) are chosen by the
condition the retained coefficients to correspond to those with
the highest average energy into the transformed blocks for all The inverse transform which leads to full restoration of
the blocks to which the image has been divided. The ~ k
s pRp (u,v) is given by:
transformed block from [BR(2n)] is found by the 2D-DCT:

~s p ( 0 ,0 )
k  ~k p 
[S0 R( 2 n )] = [T0( 2 n )][BR( 2 n )][T0( 2 n )], (6)  pR  4 − 1 − 3 − 2 d pR ( 0,0 )
s p ( 0 ,1 )  1 4 3 2   d pR ( 0 ,1 ) 
~ k ~k p
 1
 pR  =   . (11)
where [T0( 2 n )] is a matrix with dimensions 2n×2n for level р ~ ~k
s pRp ( 1,0 )  4 4 − 1 1 − 2  d pRp ( 1,0 ) 
k
 k   
= 0 which is used for implementing the DCT.
~ p
 4 − 1 1 2   ~ k p 

The rest components in decomposition (3) are the  s pR ( 1,
1 )   d pR ( 1,
1 ) 
approximation matrices for p = 1, 2. Each of them consists of
sub-matrices with dimensions 2n-p×2n-p for kp=1,2,.,4p The difference matrix [E p −1,R( 2 n − p )] for level р containing
obtained by its quad-tree split. On the other hand each sub-
[E p −p1,R( 2 n − p )]
k
matrix is calculated by: the sub-matrices is determined by the

~k ~k following equation:
[E p −p1,R( 2 n − p )] = [T p( 2 n − p )] −1[S pRp ( 2 n − p )][Tp( 2 n − p )] −1
(7)
[E p −1,R ( 2 n − p )] =
for kp=1,2,.,4p. Here 4p is the number of the branches of the ~
[B R ( 2 n )]-[B 0 R ( 2 n )] − for p = 1;
quad-tree in level p of the decomposition; [T p( 2 n− p )] −1 - = (12)
n− p ~
matrix for inverse 2D-DCT; [S~pR  [E p − 2 ,R ( 2 )] − [E p −2 ,R ( 2 )] − for p = 2.
n-p
( 2 n− p )] - the transformed
kp

block of the cut 2D-DCT of the difference matrix

kp
~ p Over the coefficients ~s pR (u,v) for all the levels of the
[E p −p 1,R( 2 n − p )]. The elements s pR (u,v) = m p (u,v). s pR (u,v)
k k p k

~ k p n− p pyramid for each sub-band a lossless entropy coding is

of the matrix [S pR ( 2 )] depend on the elements mр(u,v) of applied which includes run-length encoding, Huffman coding
the binary mask [Mр(2n-р)] for u,v = 0,1,.., 2n-p-1 according: and arithmetic coding. The resulting values could be stored in
a file which volume determines the compressed image size.
1, if s pRp (u,v) - retained coefficient,
k A decoder for restoring the compressed images with the
m p( u,v )= (8) proposed approach should consist of all the opposite
 0 - otherwise. operations to those described by (1)-(12).

144
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

III. EXPERIMENTAL RESULTS classical JPEG approach. Here the block effect is much more
suppressed.
Two test images were used with the proposed approach
called Flower and Gargoyl. They are 8 bpp (bits per pixel) IV. CONCLUSION
grayscale images with size 512×512 pixels shown in Fig. 1.
Lossy compression was applied over them with 3 levels of the In this paper a new approach is presented for highly
IPD after the Haar transform with 1 level of decomposition efficient image compression using wavelet and inverse
into 4 sub-bands –from LL to HH. On each level of the pyramid decomposition. The quality and the compression
pyramid the 4 lowest frequencies in the DCT spectrum were level of the images could be smoothly controlled by choosing
retained when working with n = 4, 3 and 2 respectively, that is different levels of both the decompositions and the
DCT transform matrix of 16×16 for 0th level, 8×8 – for the 1st quantization mask for the pyramid at given level. Progressive
and 4×4 – for the last one. image transmission from level to level is possible which
To compare the obtained results in terms of quality and proves to be very useful in the case of narrow-band
compression levels LuraWave SmartCompress 3 [7] communication channels.
application was used to compress the test images using the Incorporating the advantages of the wavelet decomposition
JPEG2000 standard (in JP2 format). The results are given in which minimizes the block effects due to the absence of
Table 1. Lossless compression is also possible by preserving localizing properties with the strong decorrelating properties
the residual after level 2 in the pyramid but which here is of the orthogonal transforms such as DCT it is possible now to
omitted. have strongly compressed digital images which when restored
In the range between 0,5 and 2 bpp for the compression have low presence of artifacts. As it is obvious from
ratio (CR) there is tangible overweight in the PSNR achieved experimental results higher compression ratios are achieved at
by the proposed approach over the JPEG2000 algorithm from equal image quality levels compared to some popular
about 4 dB for the higher ratios to around 0,05 dB. Obviously algorithms from the practice such as JPEG2000. This makes
going towards lossless compression both algorithms tend to the proposed approach a suitable candidate for further
flatten the results they achieve. And the opposite, with the improvement and establishing a new codec design for highly
reduction of the CR the inverse pyramid decomposition over efficient image compression.
the wavelet spectrum of the image produces higher quality
restored images at equal compression levels.
ACKNOWLEDGEMENT
TABLE I
COMPRESSION RESULTS This paper was supported by the National Science Fund of
Flower Test Image Gargoyl Test Image the Bulgarian Ministry of Education, Youth and Science
CR,
PSNR, dB PSNR, dB (Contract – DDVU 02/13 – “Public and Private Multimedia
bpp
IPD JPEG2000 IPD JPEG2000 Network Throughput Increase by Creating Methods for
0,5 34,772 30,876 28,689 27,594 Assessment, Control and Traffic Optimization”).
1 35,698 34,952 29,075 28,632
2 36,615 36,560 29,351 28,691 REFERENCES
The visual analysis of the restored images from both [1] K. R. Rao and P. Yip, The Transform and Data Compression
algorithms (Fig. 1) reveals the same tendency – for low CR Handbook, CRC Press, Boca Raton, USA, 2001.
levels most of the smaller details are preserved in almost [2] G. Wallace, “The JPEG Still Picture Compression Standard”, In
identical way for IPD and JPEG2000. At the same time large IEEE Transactions on Consumer Electronics, Vol. 38, Issue 1,
homogenous areas are clean from any additional artifacts pp. 18-34, 1992.
introduced by the compression. But when CR begins to [3] J. Walker and T. Nguyen, “Wavelet-Based Image
decrease considerably JPEG2000 blurs most of the smaller Compression”, Handbook of Transforms and Data
Compression, Ch. 6, pp. 267 - 312, CRC Press, Boca Raton,
details in the image mainly because of the quantization of the
2000.
wavelet spectral coefficients and specially of the low-level [4] S. Jacob and A. Cheeran, “Wavelet Based Image Compression”,
HH components. In the new approach where IPD is used no In Proceedings of the International Conference and Workshop
quantization of the wavelet spectrum is applied directly. The on Emerging Trends in Technology ICWET’10, Mimbai, India,
second transform in the pyramid (DCT) produces highly p. 999, 2010.
efficient decorrelation of these coefficients and even then they [5] A. Skodras, C. Christopoulos, and T. Ebrahimi, “The JPEG
are being quantized for the given level retaining the error for 2000 Still Image Compression Standard”, IEEE Signal
the next level. In such a way much smoother transition is Processing Magazine, Vol. 18, Issue 5, pp. 36-58, 2001.
guaranteed from level to level not omitting the details from [6] R. Kountchev and R. Kountcheva “Image Representation with
Reduced Spectrum Pyramid”. Book chapter in: "New Directions
the image in intolerable degree. Another major advantage is
in Intelligent Interactive Multimedia", Eds. G. Tsihrintzis, M.
the almost full absence of artifact distortions in the even areas Virvou, R. Howlett, L. Jain, Springer, Berlin, pp. 275-284,
because of the properties of the wavelet spectrum which is 2008.
used in a first place but not the original image as it is with the [7] http://www.luratech.com

145
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

a) b)

c) d) e)

f) g) h)

i) j) k)

l) m) n)

Fig. 1. Visual test results: a) original Flower image; b) original Gargoyl image; c) IPD 0,5 bpp; d) IPD 1 bpp; e) IPD 2 bpp; f) JP2 0,5
bpp; g) JP2 1 bpp; h) JP2 2 bpp; c) IPD 0,5 bpp; d) IPD 1 bpp; e) IPD 2 bpp; f) JP2 0,5 bpp; g) JP2 1 bpp; h) JP2 2 bpp.

146
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Efficient Adaptive Local Binarization Algorithm for Text

Extraction from Image with Complex Background
Antoaneta Popova1
Abstract – This paper describes a fast adaptive local The following section (Sec. 2) describes the modified
tresholding algorithm with the same quality as the Sauvola approach for combining integral images with the local
method and as fast as the global thresholding methods. The adaptive thresholding techniques. The evaluation of the
algorithm is independent of the operator window size and suggested algorithm is described in Sec. 3, followed by
combines the advantages of the Wiener filter preprocessing, the
conclusion in Sec. 4.
adaptive local threshold calculation and the integral images.

Keywords – Binarization, local thresholding, integral images, II. ADAPTIVE LOCAL BINARISATION ALGORITHM
Wiener filter.
AND BLOCK DIAGRAM

I. INTRODUCTION Integral image-based representation is applied in the

modified algorithm, because it allows very fast multi-scale
The binarization converts the input grayscale or color image processing.
image into a bi-level representation. Most document analysis An integral image I(x,y) of an input grayscale image Ig(x,y)
systems use this approach. The next steps in document is defined as the image in which the intensity at a pixel
analysis like Optical Character Recognition (OCR) heavily position is equal to the sum of the intensities of all the pixels
depend on the result of binarization algorithm. Several above and to the left of that position in the original image. So
different methods for image thresholding have been proposed the intensity at position (x, y) can be written as:
the last decade.
In this paper we focus on the binarization of color or x y
grayscale images with superimposed text. There are two I ( x, y )   I g ( x, y) . (1)
groups of binarization methods: global binarization and local i 0 j 0
binarization. Global binarization methods like that of Otsu,
IsoData, K-Means define a single threshold value for the The integral image is computed in a single pass. A
whole image and are very fast [1]. They give good results for cumulative row sum S(x, y) is used to calculate the integral
typical scanned documents, but are not appropriate for our image I(x,y):
goal to develop an efficient and fast tresholding for text
extraction from image with complex background: for example S ( x, y )  S ( x, y  1)  I g ( x, y )
(2)
Web pages or camera-captured documents; images with not I ( x, y )  I ( x  1, y )  S ( x, y).
uniform illumination; noised / degraded images; images with
significant intensity changes. In Fig. 1 a region A of the integral image can be computed
Local binarization methods [2, 3, and 4] try to overcome using the following 4 array references:
problems of the complex background by computing thresholds
individually for each pixel using information from the local A  ( A1  A4 )  ( A2  A3 ) . (3)
neighborhood of the pixel. The achieved results are good, but
they are often slow because the local neighborhood is If the integral image is the sum of pixels in a given area and
computed for each pixel. it can be represented as two addition and one subtraction
This paper presents a modified approach with a good operations.
performance for computing of adaptive local thresholds with
speed close to the global thresholding methods. The suggested
algorithm has considerably lower computational complexity.
The proposed adaptive threshold level computation gives
smooth, faster image binarization and often it has better noise
robustness. The “coarse to fine” approach combining Wiener
filter [5], adaptive local thresholding and integral images [6,
7] is applied.

1
Antoaneta Popova is with the Faculty of Telecommunications at
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, Fig. 1. Integral image feature computation
Bulgaria, E-mail: antoaneta.popova@tu-sofia.bg.

147
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

For example the sum of the 4 right-down pixels (3, 7, 5, 6) The main algorithm steps are the following:
of the input grayscale image block is calculated faster from 1 step: A color image conversion to grayscale image Ig is
the elements of the integral image (right block in Fig. 1) done, allowing color depth 2/24/32 bits:
without using the top row and the left column of the input
image: I g  0,3.R  0.59.G  0.11.B . (7)

3  7  5  6  (1  39)  (9  10) . (4) 2 step: Integral image, local mean and variance calculation
is applied according to the above description.
Then the local mean m(x,y) for any window size w is 3 step: Wiener (optimal) image filtration is selected for a
computed by using two addition and one subtraction preprocessing stage as more effective and faster, with low
operations instead of the summation over all pixel values in complexity and giving high accuracy of the OCR results. The
the window: Wiener filtered image is calculated:

 w w w w  m  ( 2   n 2 ).(I g ( x, y)  m))
 (I ( x  , y  )  (I ( x  , y  ) 
I wf ( x, y )  , (8)
m( x, y )   2 2 2 2  / w2 (5)
 2
 w w w w 
  (I ( x  , y  )  (I ( x  , y  ) 
 2 2 2 2  where  n is a standard noise variance, defined from the
whole image.
Similarly, the local variance  is computed very
efficiently, independent of the local window size w: 4 step: Coarse local image binarization (Sauvola) is applied
first, calculating a threshold for each pixel:
x w / 2 j  y w / 2
 
1
 ( x, y ) 
2
2
I g 2 (i, j )  m2 ( x, y) . (6)    ( x, y )  
w i  xw / 2 j  y w / 2  ( x, y)  m( x, y) 1  k   1 , (9)
  R 
An adaptive local binarization algorithm block diagram is
presented in Fig. 2. where k parameter has a value 0,2 and R is the defined in
advance dynamic range of the standard variance  . This
algorithm uses the mean and the variance around the pixel in a
local area, but threshold is better adaptively calculated for the
background with changed intensity.
A pixel S(x,y) = 1 is accepted as part of the text object if its
grayscale intensity Ig is less than the threshold θ:

S ( x, y )  1 if I g ( x , y )   ( x, y )
, (10)
S ( x, y )  0, else
else the pixel is part of the background S(x,y) = 0.
5 step: Image background Bg calculation and interpolation
is performed to obtain an image with background only – the
previously selected text pixels now are interpolated with the
closest intensity of the background pixels. The pixels in the
text object are only changed in this step. The purpose is to
achieve an equalized background for a more successful
following finer binarization:

Bg ( x, y )  I ( x, y ) if S ( x, y )  0
 x  dx y  dy
   ( I (i, j ).(1  S (i, j )))
 i  x  dx j  y  dy , (11)
 j  y  dy y  dy
if S ( x, y )  1

   (1  S (i, j ))
 i  x  dx j  y  dy
where 2dx and 2dy are the size of the operator window
(typically 40х40 px. to cover minimum two characters). If the
Fig. 2. Local binarization algorithm block diagram pixel is part of the background S(x,y) = 0 it has the same
unchanged intensity I(x,y).

148
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

6 step: Fine and final adaptive local binarization over the the window size w is done in Wiener filter (3 px.) for
grayscale input image is applied, taking in account the preserving the character contours, in the local area for Sauvola
calculated background image. The subtraction between the (15 px.) and in the local area for background calculation (40
background and the input image (Bg(x,y) – I(x,y)) gives an px.).
image with the replaced position parts of the text and In order to demonstrate the advantages (speed and ORC
background (darker) and the image can be assumed to be with accuracy after text image binarization) of our proposed
a constant background. algorithm, 5 existing binarization algorithms (IsoData, Otsu,
For the background the pixels Bg(x,y) and I(x,y) are equal Local Mean, Niblack, Sauvola) were tested.
and the background pixel intensities are not considered. The
equation for achieving the fine and final binarized image is:
Used Binarized image and Recognized text
I f ( x, y )  1 if ( Bg ( x, y )  I ( x, y ))  d ( Bg ( x, y )) algorithm
, (12)
0 else Original
where d(Bg(x,y)) is a local threshold for pixel (x,y).
For images with a constant background the value of the
minimal threshold between text and background is:
Otsu
d  q. f , (13)

where q is a weight coefficient, and δf is a mean distance

between the background and the text and it is calculated as
Am 5. of understanding the world...
follows:

f 
x  y ( Bg ( x, y)  I ( x, y)) . (14)
Niblack

 x  y S ( x, y )
idini FI; - .___ ii: F
For achieving different adaptive thresholds for different
contrasts between the text and background, the mean value b Sauvola
of the intensity of the grayscale pixels Bg(x,y), belonging to
the coarse binarized text, is calculated using the equation:

b
x  y ( Bg ( x, y)(1  S ( x, y))) . (15)
Another ay of understanding the world...

x  y (1  S ( x, y)) Suggested

The threshold for the given pixel (x,y) is equal to q.δf., if the algorithm
intensity of the image Bg(x,y) is bigger than p1.b. Else the
threshold is equal to p2.q. δf. The coefficients p1 and p2 are in Another way of understanding the world...
the range [0; 1]. The above is achieved using the
function: Fig. 3. Original image, binarized by Otsu, Niblack, Sauvola and the
d ( B g ( x, y ))  suggested algorithm with the corresponding recognized texts

 
  In Fig. 3 is presented the original image with superimposed
 1  p2  . (16)
q. f   p2  text on a complex background, the binarized images
  4 B g ( x, y ) 2(1  p1)   comparing the 4 methods - Otsu, Niblack, Sauvola and the
 
 1  exp b(1  p1)  1  p1   suggested algorithm in this paper, and the recognized
    characters from OCR system (shown under the binarized
images). After the Otsu global binarization 9 text symbols
cannot be recognized. Using the Niblack binarization many
III. EXPERIMENTS AND RESULTS black pixels pass from the background to the text object. This
is the worst case and no one of 36 characters/ symbols are
The conducted tests include images with text over the recognized. As it is shown in the Sauvola algorithm,
picture, with darker background parts, and different binarization results in the threshold being higher than the
background artifacts. optimal and 3 characters are almost unreadable in the binary
The algorithm implementation is done using program image, but the OCR system helps and only one symbol is not
language C. During the experimental testing the selection of recognized. In the suggested by us adaptive local binarization

149
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

algorithm, the text is maximally readable and all 36 characters complex background, it is easy to evaluate the results by
are recognized. comparing the recognized texts in images.
In this paper we presented a modified algorithm of the
adaptive computing thresholds for local image binarization.
The new approach in this paper is the application of integral
images in the Wiener filter in order to compute the mean and
the variance in the local processing window. This results in
increasing the speed of the thresholding algorithm regardless
of the local processing window size. The second new
approach in the presented algorithm is the computation of the
Sauvola threshold function with integral images in the stage of
coarse binarization in combination with the Wiener filter. The
proposed algorithm includes a fine adaptive binarization
technique that achieves better thresholding compared to
Sauvola, resulting in a higher recognition accuracy of the
image presented in Fig. 3. As a result all text characters are
recognized correctly after applying the described binarization
algorithm. An additional advantage of the developed
algorithm is the reduced time of the binarization process,
Fig. 4. Original image, binarized by IsoData, Mean and the suggested compared to the other local thresholding algorithms. The
algorithm
decreased time is close to the fast global binarization schemes
like Otsu.
The application of the proposed algorithm, as well as
IsoData and Mean algorithms, on test images with dark areas ACKNOWLEDGEMENT
due to bad scanning shows that the suggested approach better
cleans such backgrounds (Fig. 4). This paper was supported by the National Science Fund of
The typical binarization algorithm calculates local image the Bulgarian Ministry of Education, Youth and Science
areas sequentially around each pixel. The calculation (Contract – DDVU 02/13 – “Public and Private Multimedia
complexity presented as a number of operations O is: Network Throughput Increase by Creating Methods for
Assessment, Control and Traffic Optimization”).
C  O( N 2 . W 2 ) , (17)

where N 2 is the area of the whole image in pixels, W 2 is REFERENCES

the area of the operator window processing each pixel. We
allow maximal image 4000 x 3000 px., an operator window [1] Liang X., Image Binarization using Otsu Method, NLPR-PAL
W2= 20 х 20 px. In this case the number of the arithmetical Group, CASIA, 2009.
[2] V. Vonikakis, I. Andreadis, N. Papamarkos and A. Gasteratos,
operation is 4.9е+9. This speed is equal to 5 sec. for CPU
"Adaptive Document Binarization: A human vision approach",
with 20000 MIPS. 2nd International Conference on Computer Vision Theory and
Using the integral image approach in the proposed Applications (VISAPP), Barcelona, 2007.
algorithm the arithmetical operations are reduced to [3] J. Sauvola and M. Pietikainen, “Adaptive Document Image
Binarization”, Pattern Recognition, vol. 33 (2), pp. 225-236,
CI  3( N 2 ) . (18) 2000.
[4] S. Pavlos, E. Kavallieratou and N. Papamarkos, “An Evaluation
The achieved speed of the suggested algorithm with the Technique for Binarization Algorithms”, Journal of Universal
applying integral images is 3 sec. for the same CPU, that Computer Science, vol. 14, no. 18, pp. 3011-3030, 2008.
[5] E. Smith, L. Likforman-Sulem and J. Darbon, “Effect of pre-
means 1,7 times faster only for one algorithm step.
processing on binarization”, Proceedings SPIE Electronic
Imaging Document Recognition and Retrieval, vol. 7534, pp.
IV. CONCLUSION 75340H-75340H-8, 2010.
[6] K. Derpanis, “Integral image-based representations”,
Department of Computer Science and Engineering, York
An adaptive local binarization algorithm was proposed for University, 2007.
text images with complex background. This algorithm is [7] F. Shafait, D. Keysers and T. M. Breuel, “Efficient
appropriate for document images that are difficult to be Implementation of Local Adaptive Thresholding Techniques
recognized correctly directly as grayscale or after using a Using Integral Images”, Proceedings of 15th International
global thresholding like Otsu. After the applying of different Conference on Document Recognition and Retrieval, vol. 6815,
binarization algorithms to the real images with text and pp. 81510, , San Jose, CA, USA, 2008.

150
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Text Skew Detection using Log-polar Transformation

Darko Brodić1, Zoran N. Milivojević2, Dragan R. Milivojević3
Abstract – The paper proposes the method for text skew transform, vertical projection profiles, etc. The proposed
detection based on log-polar transformation and cross- algorithm shows a good skew estimation in the standard
correlation. The text image is transformed into log-polar domain resolution. Hence, it contributes to theirs robustness.
as well as the control ellipse. Theirs cross-correlation established Organization of this paper is as follows. Section 2 describes
the cost function. The extraction of the cost function maximum
the proposed algorithms for the estimation of the text skew.
represents the text skew value in the region. The method is
characterized by the accuracy and computational time Section 3 defines text experiments. Section 4 compares and
inexpensiveness. discusses obtained results. Section 5 makes conclusions.

Keywords – Document image processing, Log-polar

transformation, Text skew.
II. ALGORITHM

A. Document Image
I. INTRODUCTION Document text image is a product of the image scanning. It
is a digital gray-level image, which is represented by matrix
The printed text is a strongly formed text type with a D. It consists of M rows, N columns, and contains the
articulated regularity in shape [1]. Accordingly, the letters are elements which intensity has L discrete levels of gray. L is the
of the similar size and the distance between text lines is integer from {0, …, 255}, D(i, j) ∈ {0, …, 255}, where i = 1,
generally sufficient. Hence, the spacing between text lines is …, M and j = 1, …, N. After performing the binarization
decent. The orientation of the text lines is similar, which leads procedure the image represented by matrix D is transformed
to the uniform text skew. These attributes represent the into binary image B(i, j). Its elements are equal to 1 if D(i, j) ≥
relatively predicted characteristics, which simplify the printed Dth(i, j), or to 0 if D(i, j) < Dth(i, j), where Dth is given by any
skew identification. local binarization method [4]–[5]. Dth represents local
However, the text skew extraction represents a severe threshold sensitivity decision value. Currently, document
problem. It is a consequence of the digitization process. image is given as binary matrix B featuring M rows and N
Hence, the text skew occurrence is simply unavoidable. The columns.
existence of this phenomena could cause the optical character
recognition system failing. Hence, its identification represents
one of the crucial steps [2]. B. Log-polar Transformation
Existing methods for the text skew identification can be
grouped as follows [3]: projection profiles method, k-nearest The log-polar transformation is a nonlinear and non-
neighbor clustering method, Hough transforms method, uniform sampling of the spatial domain. Nonlinearity is
Radon transforms method, Fourier transformation method, introduced by polar mapping, while non-uniform sampling is
cross-correlation method, and other methods. the result of logarithmic scaling [6]. Consider the log-polar
This paper recommends a new algorithm based on the coordinate system, where denotes radial distance from the
interaction of the log-polar transformation and cross- center and denotes angle. For the input binary image B(i, j),
correlation. Firstly, it converts images into log-polar space. the center point has been extracted as B(m, n). The radius,
Furthermore, two images are cross-correlated in log-polar which ensures the maximum number of pixels to be included
domain to extract theirs similarity. As a consequence, cross- within reference circle of the conversion is assigned as R.
correlation function called cost function has been obtained. Its Center of the circle is given as m = M/2, and n = N/2 [6].
maximum values represent the angle of the text skew Furthermore, the image is converted into polar coordinate
estimation. The result gives the fractured line with two text system. This way, the input binary image B(i, j) has been
skew values: left and right. These two values represent the transformed into polar domain (r, θ) where [6]:
new elements in the skew estimation compared to the
previously mentioned methods like Hough transform, Radon r = (i − m)2 + ( j − n)2 , 0 ≤ r ≤ R , (1)

1 and
Darko Brodić is with the University of Belgrade, Technical
Faculty in Bor, V.J. 12, 19210 Bor, Serbia, E-mail:
 j−n
dbrodic@tf.bor.ac.rs. θ = arctan   , 0° ≤ θ ≤ 360° . (2)
2
Zoran N. Milivojević us with Technical College Niš, Aleksandra i−m 
Medvedeva 20, 18000 Niš, Serbia, E-mail:
zoran.milivojevic@jotel.co.rs. Furthermore, log-polar transform is given as (ρ, θ) where:
3
Dragan R. Milivojević is with Institute for Mining and
Mettalurgy, Zeleni bulevar bb, 19210 Bor, Serbia. ρ = ln r . (3)

151
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Applying a polar coordinate transformation to an image 2. Identification of the center point needed for the log-
maps radial lines in Cartesian space to horizontal lines in the polar transformation.
polar coordinate space. 3. Creation of the binary image with normalized ellipse
(ellipse image).
C. Cross-correlation 4. Log-polar transformation of the text image.
5. Log-polar transformation of the ellipse image.
6. Cross-correlation of the text image with ellipse image
Cross-correlation is a measure of similarity of two images.
in the log-polar domain.
In the discrete form, it is given as [7]:
7. Extraction of the maximum values from the cross-
M −1 N −1 correlation function.
cc(i, j ) = B(i, j ) o E (i, j ) = ∑ ∑ B(k , l ) E (i + k , j + l ) . (4) 8. Identification of the left and right side skew angle from
k =0 l =0
the center transformation point.
However, in our case cross-correlation have to be made Step 1.
between the images in log-polar domain. Hence, eq. (4) The original binary text image B is shown in Fig. 1.
should be transformed adequately. Hence, suppose that in the
log-polar domain matrices of text image B and referent object
5
E are marked as BC and EC. Furthermore, cc (in log-polar
domain) that represents the cross-correlation function can be 10
defined as:

cc(θ ) = Ccoeff ( BC, circshift (EC,θ ) ) ,

15
(5)
20

where ECS is circshift(EC,θ) and Ccoeff (BC,

circshift(EC,θ)) is given as [7]: 25

10 20 30 40 50 60 70 80

∑ ∑ ( BCρθ − BC )( ECS ρθ − ECS )

ρ θ
Ccoeff = , (6)
Fig. 1. The original binary text image.
 2  2
 ∑ ∑ ( BCρθ − BC )  ∑ ∑ ( ECS ρθ − ECS ) 
ρ θ  ρ θ  The text is extracted by the bounding box.
  
Step 2.
If the images are more alike, then the cross-correlation Typical center point B(m, n) needed for transformation is
function cc(θ) will tend to approach 1. extracted according to the pixel density in the center of the
The identification of the rotation in the spatial domain, i.e bounding box.
image space is a complex task. However, the rotation in the Step 3.
log-polar space is mapped into translation. The translation in Furthermore, the binary image with ellipse is created. The
the direction of one axis is an easy task to solve. Suppose that size of the ellipse depends on the original text image. It can be
a referent object is rotated in the space domain. If it is cross- said that ellipse is normalized according to the text image size.
correlated with the text image for the different angles, then it The binary image with ellipse is shown in Fig. 2.
will be readout as the translation in the log-polar space. The
objective is the selection of the referent object. In this paper, 10
the ellipse is selected as a referent object. It is a suitable object
20
because it can overlap text efficiently. However, the ellipse
has to be normalized according to the text image dimension. 30

Furthermore, the ellipse is split into left and right half part 40
y

from the center point of the transformation. This way, those 50

parts of the ellipse are matching with the original image by the
60
cross-correlation. Hence, they establish the left and right skew
estimation. Unlike the other methods, the log polar 70

transformation identifies two skews: left and right one. This 80

fact is the advantage of the proposed method. 10 20 30 40 50 60 70 80

x

D. Algorithm’s Steps Fig. 2. Ellipse image (ellipse has been normalized according to the
text object).
The algorithm for the estimation of the text skew based on
log-polar transformation is as follows: According to the eq. (1)-(3) log-polar transformation of the
1. Text image extraction by the bounding box (text text and ellipse image is achieved.
image).

152
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Step 4. Cross-correlation cost function

0.8
In Fig. 3, the log-polar transformation of the text image is
shown. Maximums
0.7
Left Right

0.6
50

100
0.5

150 0.4
omega

200
0 50 100 150 200 250 300 350 400
θ
250

300 Fig. 5. Cross-correlation of the original text image and ellipse in log
polar domain.
350
20 40 60 80 100 120 140 160 180
r

10
Fig. 3. Log-polar transformation of the original text image.
20

Step 5. 30
In Fig. 4, the log-polar transformation of the ellipse image
40
is shown.
y

50
Left Right

60
50
70

100
80

150 50 100 150 200 250

omega

x
200

Fig. 6. Enlarged original image with the skew line obtained from
250
the log-polar cross-correlation function.
300

350 III. EXPERIMENTS

20 40 60 80 100 120 140 160 180
r
The main goal of the experiments is the evaluation of the
algorithm for the text skew estimation. It evaluates the
Fig. 4. Log-polar transformation of the ellipse.
algorithm’s performance in the skew tracking domain.
Step 6. Experiments were performed mostly on the synthetic
As a result of the cross-correlation of the text and ellipse datasets, which represents the single line of the printed text
image in log-polar domain, a cross-correlation function is sample [8].
obtained. It is so-called cost function. This function is shown The test consists of the single line printed text rotated for
in Fig. 5. the angle β from 0° to 60° by the 5° steps around x-axis [8].
Step 7. Text sample is shown in Fig 7.
From Fig. 5 the cost function has two maximums. These Furthermore, all text samples are given in the standard
maximums represent the two angles of the text skew rate resolution of 300 dpi. The results are evaluated by the
given from the central point of the transformation. These absolute deviation, i.e. error. It is given as:
information has been return from the log-polar domain to the
∆θ A = θ A − θ REF , (7)
spatial image domain.
Step 8.
where θREF is the referent skew of the input text sample and
As a result the left and right angle skew line in drawn in the
θ i.e. θA is the skew of the text sample obtained with a tested
image. This is shown in the Fig. 6.
algorithms. Furthermore, a relative error (RE) [9] is important
for the algorithm evaluation as well. It is given as:

153
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

∆θ A θ − θ REF resolution of text images. Furthermore, the method is

RE (θ A ) = = A . (8) computer time inexpensive.
θ REF θ REF
The future investigation will be toward the estimation of
the handwritten text skew in the unconstrained text.

ACKNOWLEDGEMENT
This work is partly supported by the project funded by the
θREF Ministry of Education and Science of the Republic of Serbia,
No. TR33037.

REFERENCES
Fig. 7. Enlarged original image with the skew line obtained from [1] D. Brodic, “The Evaluation of the Initial Skew Rate for Printed
the log-polar cross-correlation function. Text,” Journal of Electrical Engineering - Elektrotechnický
časopis, Vol. 62, No. 3, pp. 142-148, 2011.
[2] P. Shivakumara, G.H. Kumar, D.S. Guru, P. Nagabhushan, “A
IV. RESULTS AND DISCUSSION Novel Technique for Estimation of Skew in Binary Text
Document Images based on Linear Regression Analysis,”
Sādhanā, Vol.30, No.1, pp. 69–86, 2005.
The result of testing is given in Table 1. [3] A. Amin, S. Wu, “Robust Skew Detection in Mixed
Text/Graphics Documents,” Proceedings of the 8th ICDAR ’05,
TABLE I Seoul, Korea, Vol.1, pp. 247–251, 2005.
TESTING RESULTS [4] Sauvola L.; Pietikainen M., “Adaptive Document Image
Binarization,” Pattern Recognition, Vol.33, No.2, pp. 225-236,
θREF (º) θA (º) ΔθA (º) RE(θA) 2000.
0 0 0 - [5] Khasman A., Sekeroglu B., “Document Image Binarisation
1 1 0 0.000 Using a Supervised Neural Network,” International Journal of
2 2 0 0.000 Neural Systems, Vol.18, No.5, pp. 405-418, 2008.
[6] Mrinal Kanti Bhowmik, Debotosh Bhattacharjee, Mita Nasipuri,
3 3 0 0.000 Mahantapas Kundu, Dipak Kumar Basu, “Classification of Log-
4 4 0 0.000 Polar-Visual Eigenfaces using Multilayer Perceptron,”
5 5 0 0.000 International Journal of Image Processing (IJIP), Vol. 4, No. 1,
10 10 0 0.000 pp. 12-23, 2010.
15 15 0 0.000 [7] Gonzalez R. C., Woods R. E. Digital Image Procesing, 2nd edn.
20 20 0 0.000 . – Prentice-Hall, 2002.
[8] D. Brodić, D.R. Milivojević, Z. Milivojević, “Basic Test
25 25 0 0.000 Framework for the Evaluation of Text Line Segmentation and
30 30 0 0.000 Text Parameter Extraction,” Sensors, vol.10, no.5, pp. 5263–
35 35 0 0.000 5279, 2010.
40 40 0 0.000 [9] V. S. Popov, “Principle of Symmetry and Relative Errors of
45 45 0 0.000 Instrumentation and Transducers,” Automation and Remote
50 49 1 0.020 Control, Vol. 62, No.5, pp. 183–189, 2001.
55 56 1 0.018
60 61 1 0.016

The investigated algorithm shows good results in the whole

testing angle range. Hence, the presented method is promising
in the domain of the accuracy. Furthermore, it is computer
time non-intensive.

V. CONCLUSION

This paper gives the analysis of the text skew estimation

techniques based on the log-polar transformation method.
This method estimates the similarity of the text image and
ellipse in the log-polar domain. As a result, the cross
correlation cost function is obtained. Given method shows
good results for the skew estimation of the printed and hand
printed text. Hence, it proves its accuracy in the standard

154
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Directional Transforms Applicability in Image Coding

Ivo R. Draganov1
Abstract – In this paper an attempt for review of the most
popular directional transforms is made in relation to their
applicability in image coding along with some directions for II. DIRECTIONAL LINEAR ORTHOGONAL
future improvement of their incorporation into complete TRANSFORMS
compression algorithms. Some comparison considerations are
given for the DDCT, DWHT, DDWT and others based on image One of the simplest 2D-transforms is considered to be the
quality, execution time and memory efficiency when applied for
image compression. These transforms are considered even more
Hadamard transform which is generalized in [10] by using the
effective if used in combination with some spectral coefficients following jacket matrix:
rearrangement as suggested for further elaboration. 1 1 ... 11
1 * ... *± 1
Keywords – Image Coding, Directional Transform, DDCT, 
DWHT, DDWT. K = ... ... ... ... ...  , (1)
 
1 * ... * ± 1
I. INTRODUCTION  1 ± 1 ... ± 1 ± 1
One of the most popular and widely used idea in image which later is used by Monadjemi and Moallem [17] for
coding (compression) is the decorrelation of the intensity texture classification where the case of sequency-ordered
(color) values and possibly removing non-significant portions matrix of rank=3 (8x8 size) is applied to extract features from
of the transformed data for which the human eye is less the image. They achieved classification accuracy of 90.5 %
sensitive. Such classical techniques are the Discrete Cosine against 90.0 % for approach using the Gabor filter and only
Transform (DCT) used in JPEG compressors and Discrete 77.5 % for the ordinary Walsh-Hadamard transform-like
Wavelet Transform (DWT) in JPEG2000 algorithm [1]. Due features. More than that, the execution time for the Directional
to their separability along horizontal and vertical direction it is Walsh-Hadamard Transform (DWHT) is more than 10 times
possible to apply them by using 1D masks reducing the less than for the Gabor filter which is considerable difference.
amount of memory needed in one computational pass and The more advanced transform – DCT – has been extended
reducing the complexity of the code structure, e.g. by using to its directional form by Zeng and Fu [8, 15] by modifying
simpler look-up tables, etc. the weighting factors according to:
Nevertheless of these enhancements there are large amount  1/ Nk , i = 0
of cases where inside the coded image exist periodic αˆ (i ) =  , k = 0,1,...,2 N − 2 , (2)
structures with one and the same pattern alternating along an  2 / Nk , i ≠ 0
arbitrary direction different from horizontals and verticals, where N is the size of the transform vector and k – the current
e.g. stripes, slopes, etc. In these cases it is appropriate to apply number of the direction the transform is being applied. They
a transform in the same direction in which the dominant also introduce a DC correction given in [15] by:
pattern spreads, that is a directional transform. 2N −2 2N −2
Two large groups can be defined for the existing directional ∆DC = ∑ N k B Q (0, k ) / ∑ Nk , (3)
transforms – the group of the linear orthogonal transforms [7- k =0 k =0
15, 17-22] and that of the directional wavelet transforms [2-6, where B denotes a column-vector with the spectral
16]. Here a brief review is made for both groups revealing coefficients along the current direction.
their basic principle, advantages and range of applicability in Depending on the direction type, e.g. diagonal down-left,
image coding. diagonal down-right, etc. a certain amount of modes is
The rest of this paper is organized as the following – in defined, usually 8. In Fig. 1 some of the possible modes are
section 2 a description of some of the most popular directional given by their basic direction. Combining the proper ones
linear orthogonal transforms is given followed by directional inside different images Zeng and Fu [15] obtain difference for
wavelet transforms in section 3 and then in section 4 – the Peak Signal-to-Noise Ratio (PSNR) from 0.5 to 1.5 dB
conclusion is made. compared to that produce by the JPEG coder. This is true for
static images compressed inside the range from 0.1 to 2 bpp
where the actual PSNR changes between 31 and 42 dB and
1 the transform block size is fixed to 8x8 pixels. Similar
Ivo R. Draganov is with the Faculty of Telecommunications, 8
Kliment Ohridski Blvd., 1000 Sofia, Bulgaria, E-mail: experiment is done for motion pictures where the comparison
idraganov@tu-sofia.bg is made between the H.263 codec and its modified version
using the DDCT. The compression ratios achieved are
between 0.1 and 6 bpp for the different videos where PSNR

155
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

varies between 28 and 45 dB. The advantage of using DDCT

is obvious along all the selected range where the PSNR
dominance is from 0.1 to 1.5 dB.

Fig. 2. Two types of pixel connections determining the primary type

operations in the Fast DDCT: a) direct and b) non-direct
a) b) c)
Dremeau et al. propose in [13] a new compression
algorithm based on extended DDCT to rectangular bases and
then use bintree segmentation along with dynamic
d) e) f) programming for optimal bases selection according to a rate-
distortion criterion. They compare their approach with the
JPEG and JPEG2000 coders. The test images compression
Fig. 1. Different DDCT modes: a) Diagonal Down-Left, b) Diagonal ratios achieved are between 0.01 and 2 bpp while the PSNR is
Down-Right, c) Vertical-Right, d) Horizontal-Down, e) Vertical- changing between 15 and 44 dB. All along this range their
Left, f) Horizontal-Up
approach proves to be better than both the other with between
1 and 5 dB.
Some simplified form of the DDCT using only diagonal
Another approach employing the directional linear
directions is fully described in [7] showing that it is actually
transforms is described in [9] for predicting visual residuals in
optimal orthonormal transform because of the minimized
a large number of cases – moving vehicles, people and other
object function maximizing smoothness. Also it is proven that
objects in city environment, landscapes, etc. The approach is
no separability can be achieved here and thus it has no explicit
tested with videos for both intra and inter coding and the
functional form and any fast algorithm.
results show enhancement as for the total PSNR with 1 to 7
More productive attempts to speed up the DDCT are
dB in comparison to AVC coder. The direction in which the
developed using ligting-based schemes. In [11] Xu et al. use
transform is going to be applied is found adaptively in each
factorization of 8-point DCT into 35 primary operations
frame.
formalized in:
r r r Some considerable expanding of the directional transforms
Y = DCT ( X ) = O35 o O34 o ... o O2 o O1 ( X ) , (4) is done when considering the challenges met in high
r r resolution image coders design. In coders such as HD Photo
where X is the input vector and Y - the transformed one.
There are two types primary operations – the first represents overlapping is introduced. In [18] similar technique is
direct connection between two pixels along a direction of undertaken in combination with DDCT and the achieved
processing and is defined according [14] by: results are very promising. In a range from 0.1 to 7 bpp the
PSNR benefit here is from 1 to 20 dB over the HD Photo itself
O( X [ni ], X [n j ], α ) = which is remarkable result.
In [19] a hierarchical class structure is introduced for the I-
 1   and B-frames from video processed by 1D directional unified
  1   transform along with bidirectional intra prediction. There are
    5 classes – from A to E. The total rate gain only for the
  1   bidirectional prediction is 3.72 %, for the directional unified
   r
r β  X  , (5)
transform – it is 5.64 % and the cumulative effect for both is
= X ← 
 
8.76 %. This scheme is currently used by the Joint
 1
   
Collaborative Team on Video Coding (JCT-VC) of ITU-T.
  γ 1  
Some further development of the spatial prediction inside a
   video frame is done in [20] where such an optimal prediction
 1   is looked for along with an adaptive transform by Han et al.
 
  1  Here a hybrid transform is constructed alternating between
 sinusoidal transform regarding the frequencies and phases of
where β = 1 and γ = α; the second type is when direct path the harmonics which are precisely defined by the boundary
could not be selected and scale parameter α is equal to β while effects between adjacent blocks inside the frame. Inter-block
γ = 0 representing the crossings over mid-lying pixels (Fig. 2). correlations are exploited in an effective manner which is
Experimental results based on that approach are presented proven by the fact that PSNR is higher by 1 to 5 dB over the
in [11, 12, 14] concerning the quality comparison with the classical approaches using DCT for video in a range of 0.3 to
JPEG algorithm in the range from 0.5 to 2.5 bpp for lossy 1.8 bpp compression ratios.
compression. The PSNR for static images changes generally Some comparable results are presented in [21] and in [22]
from 28 to 42 dB. In virtually all the cases DDCT produces where sparse orthonormal transforms and direction-adaptive
PSNR positive difference from 0.2 to 2.0 dB. partitioned block transform are introduced.

156
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

III. DIRECTIONAL WAVELET TRANSFORMS interest (ROI), modified fuzzy C means segmentation, ROI
based modified EZW algorithm ending with modified
The second large group of directional transforms includes Huffman coding. The authors managed to raise the
the wavelet ones. All of them are based on the classical compression ratio from almost 2 times at relatively constant
wavelet transform realized by the lifting implementation (Fig. PSNR around 33 dB for the test MRI images in comparison to
3) [6]. the classic EZW approach and the SPIHT algorithm.
Another approach is the robust adaptive directional lifting
wavelet transform used for image denoising in [3] by Wang et
al. The adaptive directional lifting is at the base of this method
x(n) combining the directional spatial prediction and the
↓2 +
conventional lifting scheme which removes the spatial
l(n) redundancy leaved by the directional attributes. Additional
novelty here is the classification at pixel level and the inter-
z P1(z) U1(z) scale correlation which assure more robustness of the
orientation estimation algorithm. The transform itself is
- applied at pixel level and affecting only those pixels which
h(n) belong to texture regions of interest. The PSNR for filtered
↓2 + images is increased by 7 dB for some typical for the practice
cases.
a) In [4] some optimization is done for the directional lifting
x(n) with reduced complexity. While the major disadvantage of the
+ ↑2 +
l(n) direction-adaptive discrete wavelet transform is the need of
- exhaustive search for the optimal prediction direction which
makes it too complex in contrast to the classical DWT, here
U1(z) P1(z) z-1 lowering of this complexity is aimed. Prediction of the
optimal direction is done using gradient-based technique over
a formal model of the prediction errors generated by the
h(n) directional lifting of input wedge image. Stevens et al. [4]
+ ↑2 proved practically that the prediction step remains very simple
and fast and the total complexity reduction has a factor of 11/4
b) preserving the prediction accuracy. The difference with the
original test images and the coded ones with exhaustive search
and with the optimized algorithm is about or less than 0.5 dB.
Fig. 3. One-pass link for the lifting scheme of the wavelet transform:
a) analysis filter bank and b) synthesis filter bank Similar algorithm to that described in [3] is presented in [5]
by Chang and Girod. They use local adaptation of the filtering
The prediction P(z) and update U(z) filters are at the base of directions to the image content based on directional lifting.
the analysis and synthesis filter banks of the DWT. One of the The advantage is that energy compaction is more for sharp
most often filters used in this process are the 9/7 biorthogonal image features. Additionally anisotropic statistical image
wavelet filters described in [6] by: model is created for quantifying the gain achieved by adapting
the filtering directions. In such a way the authors claim that
this algorithm is even more effective than similar ones
P1 = +1.58613(1 + z −1 ) developed earlier and gain of up to 2.5 dB for the PSNR is
+1 achieved. No loss of image structure is reported in the
U1 ( z ) = −0.05298(1 + z ) processed pictures.
P2 = −0.88291(1 + z −1 ) . (6) Kamisli and Lim propose in [6] the directional wavelet
transform to be used for prediction residuals in the video
U 2 ( z ) = +0.44350(1 + z +1 ) coding process. They clearly distinguish the coding of
prediction residuals of frame intensities such as the motion
s1 = 1.23017 s 2 = 1 / s1 compensation residual and the resolution enhancement
residual. Special attention is dedicated to the specific
Here the odd samples are predicted by two neighbouring even characteristics of the different prediction residuals and how
pixels which are first averaged and then the result is scaled they differ from those of the image (frame) itself. Adapting
while the even ones are predicted by averaging of two the model for the directional transforms used then produce
neighbouring odd pixels of the prediction residual. better results than to unify one and the same algorithm for
In [2] an efficient embedded coding is suggested for both. The experiments carried out by Kamisli and Lim
medical image compression using the contourlet transform. It indicate that coefficient savings over the classical DWT are
is an extensive scheme which incorporates discrete contourlet between 1 and 40 % with an average close to 30% and
transform, laplacian pyramid, directional filter bank, some considerably more in some cases when DDWT model is
post noise removal steps, optionally extraction of region of adapted to the specific type of residual being processed.

157
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

IV. CONCLUSION the 17th IEEE International Conference on Image Processing,

Hong Kong, PRC, pp. 185-188, Sept. 26-29, 2010.
[10] K. Horadam, “A Generalised Hadamard Transform”, In
In this paper a brief review is presented of the directional Proceedings of the IEEE International Symposium on
transforms used in image coding. They can be divided in two Information Theory, Adelaide, Australia, pp. 1006-1008, Sept.
large groups consisting of linear orthogonal transforms and 4-9, 2005.
wavelet approaches respectively. A lot of combinations exist [11] H. Xu, J. Xu, and F. Wu, “Lifting-Based Directional Dct-Like
with other popular in practice techniques such as local and Transform for Image Coding”, In Proceedings of the IEEE
holistic decompositions, lifting schemes, groupings based on International Conference on Image Processing (ICIP 2007), Vol.
spatial and time correlation, etc. Depending on the application 3, San Antonio, TX, USA, pp. III-185 – III-188, Oct. 16-19,
being developed all of them have their place in practice 2007.
[12] B. Chen, H. Wang, and L. Cheng, “Fast Directional Discrete
revealing even broader opportunities for future study and
Cosine Transform for Image Compression”, Optical
addendum. Some especially perspective approach is Engineering, Vol. 49, No. 2, pp. 020501-1 – 020501-3, Feb.
considered the hierarchical spectral multistage decomposition 2010.
of the wavelet spectrum of an image where classical linear [13] A. Dr´emeau, C. Herzet, C. Guillemot, and J.-J. Fuchs, “Sparse
orthogonal and wavelet transforms are combined together. Optimization with Directional DCT Bases for Image
Substituting transforms represented of fixed matrix Compression”, In Proceedings of the IEEE Int'l Conf. on
coefficients with adaptive ones such as Karhunen-Loev Acoustics, Speech and Signal Processing (ICASSP), Dallas, TX,
transform will produce even more efficient algorithms as for USA, pp. 1290-1293, March 14-19, 2010.
the quality and compression ratios of the coded images. [14] H. Xu, J. Xu, and F. Wu, “Lifting-Based Directional DCT-Like
Transform for Image Coding”, In IEEE Transactions on Circuits
and Systems for Video Technology, Vol. 17, No. 10, pp. 1325-
ACKNOWLEDGEMENT 1335, October 2007.
[15] B. Zeng and J. Fu, “Directional Discrete Cosine Transforms for
Image Coding”, In Proceedings of the IEEE International
This paper was supported by the National Science Fund of
Conference on Multimedia and Expo (ICME 2006), Toronto,
the Bulgarian Ministry of Education, Youth and Science Ontario, Canada, pp. 721-724, July 9-12 2006.
(Contract – DDVU 02/13 – “Public and Private Multimedia [16] A. Monadjemi, Towards Efficient Texture Classification and
Network Throughput Increase by Creating Methods for Abnormality Detection, PhD Thesis, University of Bristol, UK,
Assessment, Control and Traffic Optimization”). Oct. 2004.
[17] S. Monadjemi and P. Moallem, “Texture Classification Using a
Novel Walsh/Hadamard Transform”, In Proceedings of the 10th
REFERENCES WSEAS International Conference on COMPUTERS,
Vouliagmeni, Athens, Greece, pp. 1002-1007, July 13-15, 2006.
[1] K. R. Rao and P. Yip, The Transform and Data Compression [18] J. Xu, F. Wu, J. Liang, and W. Zhang, “Directional Lapped
Handbook, CRC Press, Boca Raton, USA, 2001. Transforms for Image Coding”, IEEE Transactions on Image
[2] M. Tamilarasi and V. Palanisamy, “An Efficient Embedded Processing, Vol. 19, No. 1, pp. 85-97, January 2010.
Coding For Medical Image Compression Using Contourlet [19] A. Tanizawa, J. Yamaguchi, T. Shiodera, T. Chujoh, and T.
Transform”, European Journal of Scientific Research, Vol. 49, Yamakage, “Improvement of Intra Coding by Bidirectional Intra
No. 3, pp. 442-454, 2011. Prediction and 1 Dimensional Directional Unified Transform”,
[3] X. Wang, G. Shi, Y. Niu, and L. Zhang, “Robust adaptive Input Document to JCT-VC, JCTVC-B042, 2nd Meeting:
directional lifting wavelet transform for image denoising”, IET Geneva, CH, July 21-28, 2010.
Image Processing, Vol. 5, Iss. 3, pp. 249-260, 2011. [20] J. Han, A. Saxena, and K. Rose, “Towards Jointly Optimal
[4] R. Stevens, A. Munteanu, J. Cornelis, and P. Schelkens, Spatial Prediction and Adaptive Transform in Video/Image
“Optimized Directional Lifting with Reduced Complexity”, In Coding”, In Proceedings of the IEEE Int'l Conf. on Acoustics,
Proceedings of the 16th European Signal Processing Conference Speech and Signal Processing (ICASSP), Dallas, TX, USA, pp.
(EUSIPCO 2008), Lausanne, Switzerland, pp. 1-5, August 25- 726-729, March 14-19, 2010.
29, 2008. [21] O. Sezer, O. Harmanciy, and O. Guleryuz, “Sparse Orthonormal
[5] C.-L. Chang, and B. Girod, “Direction-Adaptive Discrete Transforms for Image Compression”, In Proceedings of the 15th
Wavelet Transform for Image Compression”, In IEEE IEEE International Conference on Image Processing (ICIP
Transactions on Image Processing, Vol. 16, No. 5, pp. 1289- 2008), San Diego, CA, USA, pp. 149-152, Oct. 12-15, 2008.
1302, May 2007. [22] C.-L. Chang, M. Makar, S. Tsai, and B. Girod, “Direction-
[6] F. Kamisli, and J. Lim. “Directional Wavelet Transforms for Adaptive Partitioned Block Transform for Color Image
Prediction Residuals in Video Coding”, In Proceedings of the Coding”, In IEEE Transactions on Image Processing, Vol. 19,
16th IEEE International Conference on Image Processing (ICIP), No. 7, pp. 1740-1755, July 2010.
Cairo, Egypt, pp. 613-616, November 7-10, 2009.
[7] I. Selesnick and O. Guleryuz, “A Diagonally-Oriented DCT-
Like 2D Block Transform”, Proc. SPIE, Vol. 8138, 2011.
[8] J. Fu and B. Zeng, “Directional Discrete Cosine Transforms: A
Theoretical Analysis”, In Proceedings of the IEEE International
Conference on Acoustics, Speech and Signal Processing
(ICASSP 2007), Honolulu, HI, USA, pp. I-1105 – I-1108, April
15-20, 2007.
[9] R. Cohen, S. Klomp, A. Vetro, H. Sun, “Direction-Adaptive
Transforms for Coding Prediction Residuals”, In Proceedings of

158
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

An algorithm and a program module for calculating the

border height of the mass centre of a vessel
Emiliya Koleva1, Mariya Nikolova2, Mariya Eremieva3, Viktoriya Sabeva4
Abstract – This paper presents an algorithm and a program
module based on “Method for calculating the stability at moderate
and big heeling angles of a vessel”. The program is built to help the
command staff of the ship by considerably simplifying the
calculations connected to designing the cargo plan and respectively
the stability of the vessel. The program is realized on Matlab.

Keywords – Static Stability Curve, Dynamic Height of the

Mass Centre, Metacentric height, Matlab

I. INTRODUCTION
Fig. 1 – Arm of the form
The researches related to the normalization of the stability
of a ship meet significant difficulties caused by the ignorance
of its hydrodynamic interaction with the water. The empirical-
statistic method widely covers in the introduction of certain
norms for stability in the design and exploitation of the
vessels. The research of the stability could be most clearly
made with graphical interpretation of the so called Static
Stability Curve (SSC). The direct building of the curve is too
hard for the command staff because of the many calculations.
Therefore, a new method is presented in “Method for
calculating the stability at moderate and big heeling angles of
a vessel”. It simplifies the graphic work on creating SSC and
initiating a diagram of the dynamic height of the mass centre
(DHMC). An algorithm and a program module are created
Fig. 2 – KN-curves of freight passenger ship [1]
using this method. The program module is realized on Matlab
[3].
2. For each  i are taken heights of the mass centre KG j in
II. ALGORITHM realistic exploitation borders for the height of the mass centre
G throughout 0.1m .
The sequence of the algorithm based on the new method for 3. The arm of the height (lG ) is calculated for each KG j ,
obtaining the border graphical dependence between the mass
centre and the displacement of the ship  is the following: heeling angle  and  i by the formula [1,3]
1. The values ( KN ) j (Fig. 1) are determined from the (lG )  j  KG j sin  , (1)
KN -curves for each displacement  i from 1 " empty" to where  is from 10 to 40 throughout 10 . [3](Fig. 3)
0 0 0

 n " full" ship in m at intervals of   500m .(Fig. 2)

3 3

1 2
Mariya Nikolova, 3Mariya Eremieva,
Emiliya Koleva, viktoriya_sabeva@yahoo.com,
eremievam@abv.bg,
4
Viktoriya Sabeva
mpn@abv.bg, are with the Engineering Faculty at Naval
emiliya_f@abv.bg.
Academy “Nikola Yonkov
eremievam@abv.bg, Vaptsarov”, 73 Vasil Drumev str, Varna
viktoriya_sabeva@yahoo.com,
9026, Bulgaria,emiliya_f@abv.bg.
mpn@abv.bg,
e-mails respectively: Fig. 3 – Arm of the height
emilia_f@abv.bg, mnp@abv.bg, viktoriya_sabeva@yahoo.com,
eremirvam@abv.bg

159
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

4. The arm of the righting moment GZ   i, j is evaluated for For each  i are built j in number diagrams respectively

each  i and each   10 ,20 ,30 ,40 . (Fig. 4) [2]

0 0 0 0 for each l j ,i i.e. same KN for each angle  i [2].
0 0
To form the initial area from 0  10 is necessary to lay
vertically the initial metacentric height GjMi at
0
1rad  57.3 . The vertex G j M i is connected to the origin of
the coordinate system. At this area the curve must be tangent
of the connection between the vertex and the origin. (Fig. 7)

Fig. 4 – Arm of the righting moment

5. The height of the transverse metacentre KM i is

determined depending on the kind of the vessel. (Fig. 5)

Fig. 7 Static Stability Curve

8. Next step is to calculate the areas S 00 300 , S 00 400 and

S 300 400 and compare them with the ones required in IMO. If
the three areas are the same as the required or one of them is
the same and the others are bigger KG j is considered as a
solution.[6]
9. The obtained KG j responds to a specific

Fig. 5 – Initial transverse metacentre displacement  i . The graph of KG j  f ( i ) is built for the
whole diapason 1   n . The area over the graph is
impermissible for the cargo height[1,4].

III. PROGRAM MODULE BASED ON THE ALGORITHM

For the calculations in this report are used values from the
Naval Academy’s training ship’s documents.
3
The displacements i are from 1  4000m to
 n  10000m at intervals of   500m .[6]
3 3

Fig. 6 – Determining the height of the transverse The heights of the mass centre KG j from KG1  6m to
metacentre [1]
KG m  9m at intervals of 0.1m .[6]
The heights of the transverse metacentre are taken from the
6. The transverse metacentric heights G j M i are determined
KM -curve of a freight passenger ship for each displacement
as
G j M i  KM    KG
i j (2)
 i [6].
The program is created following the steps shown in the
for each  i . flowchart. (Fig. 8)
Finally an acceptable zone for dynamic stability is
7. The Static Stability Curve (SSC)
GZ 
determined. (Fig.9)
i, j  f ( ) . (3)

160
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

CONCLUSIONS

1) Simplifies the evaluation of the stability at moderate and

big heeling angles of the ship.
2) The program module can be applied to all kinds of ships,
for which is necessary to build SSC.
3) The graph of the DHMC could be introduced in the
ship’s documents as reliable evaluation of stability.

REFERENCES

Fig. 9 - Diagram of the dynamic height of the mass centre [1] M. Zhelyazkov, P. Kaloyanchev, “Book of problems based on
ships theory”, Military publisher, 2001
[2] Matlab help
[3] P. Petkov, V. Petkova, I. Draganov, “Theory of the ships and
ships construction”, Steno, 2008
[4] STCW Module 7-Ship Construction, Stability and Damage
Control.
[5] STCW Module 17-Introduction to Ships.
[6] Rezolution А 749 (18) of IMO-requirements for SSC.

161
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Begin

Enter values of
 , KN and i, j , k  1
KM

no yes
Plot the i≤13
graph for all ( KN k ) i
KG max

i = i+1 [(lG ) k ] j  KG j sin  k

End

Store KG max i
GZ 
i, j  
 KN k i  lG k  j

no
k  k 1
yes
k≤4
If yes
S1 , S 2 , S3  IMO
no

j = j+1

Calculating S1  S00 300 ,

 
yes
S 2  S00 400 and G j M i  KM i  KG j no j ≤31
S3  S300 400

Fig. 8 – Flowchart of the algorithm

162
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Power consumption analysis of fault tolerant

real-time systems
Sandra Djosic1, Milun Jevtic2 and Milunka Damnjanovic3
Abstract – In this paper we analyze the power consumption of technique means less time for fault tolerance, and vice versa.
fault tolerant real-time systems. We consider real-time systems Therefore, there is a tradeoff between low energy
where the fault tolerance is achieved running the task affected by consumption and high fault-tolerance. In accordance with that,
a transient fault again using time redundancy. The power we designed one heuristic-based DVFS algorithm and used it
consumption analysis is done using one heuristic-based dynamic
voltage and frequency scaling algorithm which we designed. The
for RTSs analysis. The analysis refers to the power
simulation results show that our proposed algorithm can be consumption and fault tolerance through time redundancy for
successfully used for power consumption according to fault different number available operating frequency levels of the
tolerance analysis. processor used in the RTSs.
The rest of the paper is organized as follows. The first part
Keywords – Dynamic voltage and frequency scaling, Fault of Section II describes real-time system, power, fault and
tolerance, Real-time systems. feasibility models we used in the paper. The second part of
Section II introduces our proposed heuristic-based DVFS
algorithms. Section III gives the simulation results and finally,
Section IV presents our conclusions.
I. INTRODUCTION
Real-time systems (RTSs) have been designed in order to II. MODELS AND ALGORITHM DESCRIPTION
be safe and extremely reliable. They are usually realized as
real time systems with the ability of tolerating some faults. A A. Models description
fault-tolerant RTS has to ensure that faults in the system do
not lead to a failure. Faults could be transient, permanent or For a system model, we assume one uniprocessor RTS with
intermittent faults. Among these, the transient faults are much variable processor’s operating frequency fj (j = 1, ..., m) where
more common than faults of other two types. Transient faults fj < fj+1. Changing the operating frequency of the processor the
have the feature that they occur and then disappear so fault voltage also changed and it could be switched between m
tolerance can be achieved running the task affected by a values. This system can be used for one real-time task set
transient fault again (i.e. re-executing the task). It means that execution. We assume a set of n periodic real-time tasks,
time redundancy can be used as fault-tolerance techniques by Γ={τ1,..., τn} where each tasks are defined by a period Ti,
using free slack time in the system schedule to perform worst case execution time (WCET) Ci, deadline Di and
recovery executions, [1], [2]. priority pi. We assume that Di ≤ Ti, for i = 1, 2, ..., n. The
Energy efficiency is also crucial to many real-time systems. WCET of real-time tasks corresponds to executing the task at
Dynamic Voltage and Frequency Scaling (DVFS) is the most the maximum frequency fm. For simplicity, we assume that the
popular and widely deployed technique for reducing power WCET of a task scales linearly with the processing speed. So,
consumption of processors [3], [4], [5]. Nowadays, DVFS is a
if we scale the operating frequency by a factor α, then WCET
commonly used technique for energy management and is
must be scaling by factor 1/α, i.e.
supported by many commercial processors [6].
Ci (fj) = Ci (fm) fm / fj.
Fault tolerances through time redundancy as well as energy
Power consumption of an active processor can be modeled
management through frequency and voltage scaling have been
as
well studied in the context of real-time systems. But simply
PA(f) = Pd(f) + Pind,
applying fault recovery techniques and energy minimization
where Pd(f) and Pind are frequency dependent power and
techniques one after the other results in inferior quality. These
frequency independent power respectively [7]. Frequency
techniques use free slack time and since free slack time is a
dependent power is
limited resources, it is obvious that more slack time for DVFS
Pd(f) = V2(f) Cef f
where V is supply voltage and it is a function of operating
1
Sandra Djosic is with the University of Nis, Faculty of Electronic frequency, Cef is the switch capacitance and f is the operating
Engineering, Aleksandra Medvedeva 14, 18000 Nis, Serbia, frequency. Beside power, for DVFS techniques energy is
E-mail: sandra.djosic@elfak.ni.ac.rs. equally important and it is defined as the integral of power
2
Milun Jevtic is with the University of Nis, Faculty of Electronic over time.
Engineering, Aleksandra Medvedeva 14, 18000 Nis, Serbia,
We assume that faults can occur during execution of any
E-mail: milun.jevtic@elfak.ni.ac.rs.
3
Milunka Damnjanovic is with the University of Nis, Faculty of
task. We consider transient faults and assume that the
Electronic Engineering, Aleksandra Medvedeva 14, 18000 Nis, consequences of a fault can be eliminated by simple re-
Serbia, E-mail: milunka.damnjanovic@elfak.ni.ac.rs. execution of the affected task at its original priority level and
at its original oparting frequency. The re-execution of the

163
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

corrupted task must not violate timing constraints of any task Input: operating frequency levels fj (j=1..m),
in Γ. characteristics for n real time tasks (Ci, Di, Ti, pi),
For checking the feasibility of fault tolerant real-time task fault tolerant constraint (TF)
set we use the response time analysis (RTA). In the RTA, the _______________________________
fault-tolerance capability of a RTS is represented by a single
parameter, TF, which corresponds to minimum time interval (1) for each Task in TaskSet set Task’s_Freq to fm and set
between two consecutive faults that the RTS can tolerate. Task’s_Key to true;
More about RTA can be found in [8], [9]. The basic equation (2) repeat step (3) to (7) until there are true Task’s_Key in the
characterize for RTA is Eq. (1). TaskSet;
(3) for each unlock Task in TaskSet do
(4) temporarily set Task’s_Freq to Lower_Task’s_Freq;
⎡ Rn ⎤ ⎡ Rn ⎤
Rin + 1 = Ci +
(1) (5) if new TaskSet is not feasible
∑ ⎢ i ⎥C + ⎢ i ⎥ max ( C )
⎢ T j ⎥ j ⎢ T F ⎥ j∈hp( i )∪ i j (6) then set Task’s_Key to false;
j∈hp( i ) ⎢ ⎥ ⎢ ⎥
(7) else calculate ΔPower as Power(Task’s_Freq) –
Power(Lower_Task’s_Freq);
With Eq. (1) the response time Ri of a task τi could be (8) find Task with maximum ΔPower and set Task’s_Freq to
calculated. This equation has three main addends. The first is Lower_Task’s_Freq;
WCET Ci for a task τi. The second presents interference due
to preemption by higher priority tasks. We use hp(i) to denote _______________________________
the set of tasks with higher priorities than i, hp(i)={τj∈Γ⎪pj > Output: TaskSet with new frequency assigne to each Task
pi}. The third addend refers to possible faults in the system. If
we assume that inter-arrival time between faults is TF then
⎡R ⎤ Fig. 1. Pseudo code of the proposed algorithm
there can be at most ⎢ i ⎥ faults during the response time Ri
⎢ TF ⎥ The algorithm starts with assigning the maximum operating
of task τi. Since these faults could occur during the execution frequency, fm, to each real-time task, step (1). Also, at the
of task τi or any higher priority task which has preempted τi, beginning, all tasks are allowed to change the frequency - we
each fault may add max ( C j ) to the response time of say that all tasks are unlocked. An iteration of the algorithm
j∈hp( i )∪i decreases the frequency of one task for one frequency level.
task τi. So, the third addend in Eq. (1) presents an extra time The chosen task is one for which the frequency decrement
needed tasks recovery due to faults. yields maximum power reduction among all unlocked tasks
Since Ri appears on both sides Eq. (1) is recurrence provided that tasks set remains feasible. To find such task, the
relations which starts with Ri0 = Ci . The solution is found when algorithm checks all currently unlocked task. For example,
frequency index of one unlock task τi is temporarily decreased
Rin + 1 = Rin . If during the iteration process we get that
for one frequency level, i.e. from fj to fj-1, step (4), and
Rin +1 > Di then task τi is infeasible and iteration process must feasibility of task-set is tested using Eq. (1), step (5). If task-
be terminated. set is not feasible, τi is locked, step (6). Otherwise, if task-set
is feasible, the difference between power consumption of τi at
lower (fj-1) and higher (fj) frequency is calculated, step (7).
B. Algorithm description Then, τi’s frequency is changed back to fi. After checking all
tasks, one that remains unlocked and provides the maximal
In order to solve the tradeoff problem between low energy power reduction is selected, and its frequency index is
consumption and high fault-tolerance, we propose one decremented, step (8). Additionally, the selected task is locked
heuristic DVFS algorithm. The proposed algorithm has to find if its new frequency equals 1, i.e. corresponds to the lowest
appropriate execution frequency for each task, from the real- execution frequency, f1. After that, the algorithm enters the
time tasks set, such that energy consumption is minimal when next iteration. The algorithm finishes when there are no more
faults are presence. Fig. 1 gives the algorithm in pseudo code unlocked tasks. The frequency assignment to each task is
form. algorithm’s output. We previously proved the proposed
For this purpose we created a heuristic-based algorithm to heuristic algorithm and more about that can be found in [10].
The RTA is the basic of our proposed algorithm. This analysis
is used to guarantee feasibility of real-time tasks set and fault
tolerance. The input parameters for the algorithm are: III. SIMULATION RESULTS
- frequency fj (j=1,..., m) where fj < fj+1 and m is number
of frequency levels; We realized simulator based on our proposed heuristic
- characteristics for all n real-time tasks from the set: DVFS algorithm. The input parameters of the simulator are
period Ti, worst case execution time Ci, priority pi and number of real-time task and their real-time characteristics:
deadline Di, for i=1,..., n; minimum inter-arrival time Ti, worst case execution time Ci
- minimum time interval between two consecutive faults on maximum operating frequency fm, deadline Di and priority
TF. pi. Also, input parameters are processor’s voltage and
frequency levels and fault constraints TF.

164
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA
TABLE I
TASKS SET FROM GENERIC AVIONICS PLATFORM

Ti=Di Ci
τi pi
(ms) (ms)
Nav_Status 1 1000 1
BET_E_Status_Update 2 1000 1
Display_Stat_Update 3 200 3
Display_Keyset 4 200 1
Display_Stores_Update 5 200 1
Fig. 2. Power consumption according to number of frequency
Nav_Steering_Cmds 6 200 3 levels in the absence of faults in the RTS
Tracking_Target_Upd 7 100 5
It can be concluded that power reduction is better when
Display_Hook_Update 8 80 2 more voltage levels are included. The maximum energy
Display_Graphic 9 80 9 savings is 42.8% for 5 levels and the minimum savings is
31.5% for only 2 frequency levels. The energy reduction is
Nav_Update 10 59 8
significant even for low number of frequency levels and this
clearly shows the effectiveness of our proposed algorithm.
We performed simulations with a number of synthesized Our next step, in the simulation process, was to consider
real-time task sets and few real-world applications. The possible faults appearance in the RTS. This is represented by a
characteristics of one real-world application are summarized single parameter, TF, which corresponds to minimum time
in Table I. It is a task set taken from the Generic Avionics interval between two consecutive faults that the RTS can
Platform (GAP) used in [11]. For the processor’s frequency tolerate. For the input parameters of the simulator we used the
levels we used data for Transmeta Crusoe procesor from [12]. same task set and the same processor. Fig. 3 shows the
The relevant parameters for the processor are listed in Table simulation results for different number of available frequency
II. levels. We used the same sets of frequencies (667MHz,
300MHz), (667MHz, 600MHz, 300MHz), (667MHz,
TABLE II 600MHz, 400MHz, 300MHz), (667MHz, 600MHz, 533MHz,
PROCESSOR FREQUENCIES, VOLTAGES AND POWER 400MHz, 300MHz). The x-axis of the Fig. 3 represents the
ratio of TFmax to TF. TFmax is minimum time interval between
CPU CPU two consecutive faults that the task set can tolerate on
Voltage
Frequency Power maximal executing frequency and TF is input simulation
(V)
(MHz) (W) parameter. This axis represents the normalized TF value which
300 1.2 1.3 is proportional to fault tolerance of the task set. As fault
400 1.225 1.9 tolerance proportional to time redundancy this axes also could
533 1.35 3 represent free slack time in the systems. The y-axis represents
600 1.5 4.2 the power saving with respect to the power consumption at
667 1.6 5.3 maximum frequency calculated in percents.
According to number of available frequency levels the
First, we assumed that there were no faults in the system. simulation was done for four possible scenarios. All four
With this assumption, we used our proposed algorithm to find scenarios indicate the same fact that power reduction leads to
the appropriate execution frequencies for each real-time task less fault tolerance and vice versa. Now due to simulation
that lead to the maximum energy savings. Fig. 2 shows the results, we can better perceive the tradeoff between power
simulation results for GAP task set and Transmeta Crusoe consumptions and fault tolerance. For example, let’s suppose
processor. We performed simulation for different number of that power reduction demands are between 40% and 45%. It
available frequency (voltage) levels. That is represented on can be seen, from the Fig. 3, that processor with 4 or 5
the x-axis where 2, 3, 4 and 5 frequency levels include set of frequency levels could fulfill these demands. Also, fault
frequencies (667MHz, 300MHz), (667MHz, 600MHz, tolerances vary for the given power reduction interval. The
300MHz), (667MHz, 600MHz, 400MHz, 300MHz), best is to choose one with maximal tolerances.
(667MHz, 600MHz, 533MHz, 400MHz, 300MHz) Also, it can be concluded that power reduction is better
respectively. The y-axis represents the power reduction when more voltage levels are included. With larger number of
calculated in percents. This reduction is presented as power frequency levels there are more possible task-frequency
saving with respect to the power consumption at maximum mapping, so the chance of finding solutions with lower energy
frequency. becomes higher.

165
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 3. Power consumption according to fault tolerance for different number of frequency levels

IV. CONCLUSION [3] K. Woonseok, S. Dongkun, Y. Han-Saem, K. Jihong, M. Sang,

“Performance Comparison of Dynamic Voltage Scaling
Algorithms for Hard Real-Time Systems”, Proceedings of the
In this paper the power consumption of the real-time Eighth IEEE Real-Time and Embedded Technology and
systems according to fault tolerance through time redundancy Applications Symposium (RTAS’02), pp. 219 – 228, 2002.
analysis are given. The analysis is based on the heuristic [4] A. S. Ahmadian, M. Hosseingholi, A. Ejlali, “A Control-
DVFS algorithm which we realized. The proposed algorithm Theoretic Energy Management for Fault-Tolerant Hard Real-
offers the possibility to study the trade-off between energy Time Systems”, 2010 IEEE International Conference on
efficiency and fault tolerance for real-time task sets. Computer Design, pp. 173-178, 2010.
Generally, this trade-off in discrete systems is NP-hard, so the [5] R. M. Santos, J. Santos, J. D. Orozco, “Power saving and fault-
heuristic-based approach is imposed as possible solution for tolerance in real-time critical embedded system”, Journal of
system Architecture 55, pp. 90-101, 2009.
RTSs analysis.
[6] “Intel PXA270 Processor Electrical, Mechanical and Thermal
On the basis of proposed heuristic-based algorithm we Specification Data sheet”,
realized simulator. Our simulations results show that power www.phytec.com/pdf/datasheets/PXA270_DS.pdf, 2005.
reduction is better when more operating frequency levels are [7] Z. Dakai, R. Melhem, D. Mosse, “The Effects of Energy
included. This is valid for the case of the absence or the Management on Reliability in Real-Time Embedded Systems”,
presence of the faults, in the RTS. Our opinion is that this Proceedings of the 2004 IEEE/ACM International conference
simulator could be successfully used in the RTS design on Computer-aided design, pp. 35-40, 2004.
process. [8] S. Đošić, M. Jevtić, “Analysis of Real-Time Systems Timing
Constrains”, SSSS2010, 3rd Small Systems Simulation
Symposium 2010, February, 12-14, Faculty of Electronic
ACKNOWLEDGEMENT Engineering, Niš, Serbia, pp 56-60, 2010.
[9] G. Lima, A. Burns, “An Optimal Fixed-Priority Assignment
This paper is supported by Project Grant III44004 financed Algorithm for Supporting Fault-Tolerant Hard Real-Time
Systems”, IEEE Transaction on Computers, Vol. 52, No. 10, pp.
by Ministry of Education and Science, Republic of Serbia. 1332-1346, October 2003.
[10] S. Đošić, M. Jevtić, “Dynamic voltage scaling for real-time
REFERENCES systems under fault tolerance constraints”, accepted for
publication on 28th International Conference on
Microelectronics, MIEL2012.
[1] S. Đošić, M. Jevtić, “Scheduling in RTS Using Time Redundancy [11] Locke, C. D., Vogel, D. R., Mesler, T. J., “Building a
for System Recovery After Faults”, Proceedings of papers, Indel Predictable Avionics Platform in Ada: A Case Study”,
2004, Banja Luka, pp. 146-149, November 2004. Proceedings of IEEE Real-Time Systems Symposium, pp. 181–
[2] S. Đošić, M. Jevtić, M. Damnjanović, “Analysis of possibilities 189, 1991.
to overcome the transient faults in real-time systems with time [12] Ying Zhang, Krishnendu Chakrabarty, “Task Feasibility
redundancy”, XLVI International scientific conference on Analysis and Dynamic Voltage Scaling in Fault-Tolerant Real-
information, communication and energy systems and Time Embedded Systems” Proceedings of Design, Automation
technologies, ICEST 2011, Proceedings of Papers, volume 2, and Test in Europe Conference and Exhibition, Vol.2, pp. 1170
pp 417- 420, Serbia, Niš, June 29 - July 1, 2011. – 1175, 2004.

166
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

CHALLENGES OF PERSONALIZATION AND COLLABORATION

LEARNING PROCESS BY USING BLOGS
Teodora Bakardjieva1, Boyka Gradinarova 2
Abstract – Semantic blogging is a recently emerging technology • Combination of solitary and social interaction.
that attempts to solve the problems of traditional blogging by
integrating the features of the semantic web. However, the Blogging is increasingly finding a home in education (both
semantic capabilities currently implemented for semantic in school and university), as not only does the software
blogging are still limited. It is difficult to obtain blog entries
remove the technical barriers to writing and publishing online
relevant to a topic in an aggregated and organized form. A new
framework for semantic blogging has been developed capable of - but the 'journal' format encourages students to keep a record
organizing results relevant to user requirement. At the centre of of their thinking over time. Blogs also of course facilitate
this approach are the challenges of personalization and critical feedback, by letting readers add comments - which
collaboration. Rather than integrating different tools into a could be from teachers, peers or a wider audience.
centralized system, the idea is to provide the learner with tools
and hand over control to him/her to select and use the tools the Students use of blogs are far ranging. A single authored blog
way the learner deems fit. can be used to provide a personal space online, to pose
questions, publish work in progress, and link to and comment
Keywords – blogging, semantic relations, personalization, on other web sources. However a blog needn't be limited to a
collaboration. single author - it can mix different kinds of voices, including
fellow students, teachers and mentors, or subject specialists.
Edu-blogging pioneer Will Richardson (author of the main
I. INTRODUCTION TO BLOGS books devoted to Blogs, Wikis and Podcasts) in 2001 used the
blog software Manila (http://manila.userland.com) to enable
his english literature students to publish a readers guide
Blog posts or blogs are primarily textual and can vary
(http://weblogs.hcrhs.k12.nj.us/bees) to the book The Secret
widely in their content. They can be devoted to politics, news
Life of Bees. Richardson asked the book's author, Sue Monk
and sharing opinions or dedicated to technical developments.
Kidd, if she would participate by answering questions and
Blog entries are usually maintained in chronological order, but
commenting on what the students had written - to which she
are usually displayed in reverse chronological order. Nardi et
agreed. The result was a truly democratic learning space.
al. (2004) identified five reasons why blogs are used:
Richardson marked 10 years since his first blog post, a full
• to update others on activities and whereabouts;
decade of writing and sharing online. He defines the education
• to express opinions to influence others; reform: “We don’t need better, we need different”
• to seek others’ opinions and feedback; (Richardson, 2011)
• to “think by writing”;
• to release emotional tension. Today’s students are immersed in the digital age, but can our
The learning specialists Fernette and Brock Eide cited educational system keep up? Best-selling author Will
by Will Richardson (2006) identified the Potential Richardson's comprehensive collection of posts from his
benefits of using blogs in educational process as acclaimed blog (http://weblogg-ed.com) outlines the
following: educational reform we must achieve to stay ahead of the
• Can promote critical and analytical thinking. curve:
• Can promote creative, intuitive and associational
thinking (creative and associational thinking in • Project-based learning
relation to blogs being used as brainstorming tool • Student-created media that develops critical thinking
and also as a resource for interlinking, commenting
on interlinked ideas). • Extending learning beyond the classroom and school
• Can promote analogical thinking. hours
• Potential for increased access and exposure to
quality information. • Cooperative and collaborative learning

• Student empowerment and career readiness

1 The necessary shift will not magically happen, but experts

Teodora Bakardjieva is Director of· Institute of Technology at
Varna Free University, гр. Варна 9007, к.к. Чайка, Bulgaria, E- agree that it must happen now. This compilation will inspire
mail: bakardjieva@vfu.bg. educators and parents to engage in the technology their
2
Boyka Gradinarova is with the Department of Computer children already embrace, and to take an active role in
Sciences and Technology at Technical University of Varna,1 transforming education to meet the challenges of the digital
Studentska Str.Varna 9000, Bulgaria. E-mail Bgradinarova@tu- revolution..
varna.bg

167
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

II. OBSERVATIONS AND DISCUSSIONS instances in the ontology of the domain of application.
Integrated with the ontology is the inference engine, which
Herring et al. (2004) defined three types of blogs: personal can deduce implicit relations from the ontology. All the blog
journals, “filters” (because they select and provide entries related to the relevant ontology instances are obtained
commentary on information from other websites) and from the blogontology mapping. The total relevant blog
“knowledge logs”. The majority of blogs are the online diary entries obtained are finally organized into an aggregated and
type. Bloggers are interested in reading new information, navigable collection by the organizer. The system also
sharing knowledge and being connected with other users. produces output in RSS format which computers can
While blog writers are more extroverted, blog readers are understand and aggregate.
more consumerist.

The use of blogs and semantic blogs has recently been

associated with a decentralised form of knowledge
management (Cayzer, 2004, Breslin & Decker 2007). Metadata
Semantic blogging is a technology that builds upon blogging Schema Search Blog
entries
and enriches blog items with metadata. For publishing
User
information such as research publications, there is need of Blog RDF
query
Query Blogging
some structure and semantic blogging provides this. Items Ontology Infrastructur Metadata
Processor
may be classified using ontologies. Semantic links may exist Mapping Store
between items (Cayzer, 2004b). Semantic blogging uses
desirable features of both blogging and the semantic web to
deal with the challenges of traditional blogging. The semantic RSS Output
Ontology Organizer
web is well suited for incrementally publishing structured and
semantically rich information. On the other hand, the easy
Inference
publishing nature of blogging can boost the semantic web by
publishing enough data and resources (Cayzer 2004a; Cayzer, Organized
2004b). results

Semantic blogging can help users discover items of interest

Fig. 1. System architecture of the semantic blogging framework
in blogs. Navigation through the blogosphere can be more
flexible and meaningful due to interconnections among
various items and topics. Aggregation of useful materials
Some edu-blogs that are used at Varna Free University
across multiple blogs and the semantic web is possible.
(Fig. 2, Fig. 3, Fig. 4):
Semantic blogging can extend blogging from simple diary
browsing to informal knowledge management (Cayzer,
2004b). Publication is easy in semantic blogs too because only
some additional metadata data have to be added compared to
traditional blogs. The users do not need to put any effort to
enjoy the additional features provided. Hence, there is not
much effort added in using a semantic blog instead of a
conventional one. The rich metadata and semantic structure
work behind to give the user the added value experience of
semantic blogging. However, the semantic capabilities
currently implemented for semantic blogging are still limited.
It is difficult to obtain blog entries relevant to a topic in an
aggregated and organized form.

There is newly developed framework for semantic blogging

capable of organizing results relevant to user requirement
(Shakya, 2006). Attempts for implementation of that
framework are made at Varna Free University (VFU) to
provide more effective navigation and search by exploring
semantic relations in blogs.

The system is built upon a blogging infrastructure backed

up by an RDF metadata store. The metadata schema enriches Fig. 2. Edu-blog for the Ranking System for the Bulgarian
the blog entries input. The metadata schema also helps the Universities
query processor to search by metadata. Users input queries to
the system according to their information requirement. The
query processor searches for matching blog entries and

168
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

REFERENCES
[1] Bates, T. (2011), E-Learning Outlook for 2011
http://www.tonybates.ca/2011/01/16/e-learning-
outlook-for-2011/
[2] Bates, Tony (2010), Blackboard acquires Elluminate and
Wimba: the end of LMSs?
http://www.tonybates.ca/2010/07/11/blackboard-
acquires-elluminate-and-wimba-the-end-of-lmss/
[3] Boulos, M. N. K., Maramba, I. & Wheeler, S. (2006)
Wikis, blogs and podcasts: A new generation of web-based
tools for virtual collaborative clinical practice and education,
BMC Medical Education, 6(41).
http://www.biomedcentral.com/ content/pdf/1472-
6920-6-41.pdf (Last visited January 2011)
[4] Breslin, J.G, S. Decker (2007), "The Future of Social Networks
on the Internet: The Need for Semantics", IEEE Internet
Fig. 3. Edu-blog for Choreography Computing, vol. 11, no. 6, pp. 86-90pp.
[5] Cayzer, S. (2004a). Semantic blogging and decentralized
knowledge management. Communications of the ACM, 47(12):
47-52. http://portal.acm.org/
citation.cfm?id=1035164&coll=GUIDE&dl=ACM&CFID=4
9 229987&CFTOKEN=13649580&ret=1#Fulltext
[6] Cayzer, S. (2004b). Semantic Blogging: Spreading the Semantic
Web Meme. http://citeseer.ist.psu.edu/698724.html Last
visited July 2011
[7] Chao, J. (2007) Student project collaboration using Wikis.
Proceedings of the 20th Conference on Software Engineering
Education and Training (CSEE&T2007), Dublin, Ireland: July
3-5.
[8] Chatti, M.A., Jarke, M. and Frosch-Wilke, D., (2007). The
Future of e-Learning: a Shift to Knowledge Networking and
Social Software, International Journal of Knowledge and
Learning, IJKL, 3 (4), p. 404-420
[9] Chatti, M.A., Srirama, S., Kensche, D. & Cao, Y. (2006).
Mobile web services for collaborative learning, Proceedings of
the 4th International Workshop on Wireless, Mobile and
Ubiquitous Technologies in Education (WMUTE 2006), 16–17
Fig. 4. Edu-blog for Spatial Design November, Athens, Greece.
[10] Godwin-Jones, R. (2003). Blogs and Wikis: Environments for
Online Collaboration. Language Learning and Technology
7(2): 12-16. http://llt.msu.edu/vol7num2/
III. CONCLUSION TO BLOGS pdf/emerging.pdf Last visited July 2011 Stephen Downes, E-
Learning 2.0:Notes from Stephen Downes
Teachers are using blogs to provide up-to-date information http://www.downes.ca/
and commentary on their subject areas, as well as posting http://www.speedofcreativity.org/
questions and assignments and linking to relevant news stories 2006/10/08/elearning-20-notes-from-stephen-downes/
and websites. [11] Gradinarova. B. (2009). Importance of web 2.0- mediated
competence for the educational demands on learners. In
They are preferred tools because they have advantage to Proceedings of the 10th International Conference on Computer
provide reliable and safe information protected from spam. Systems and Technologies and Workshop for PhD Students in
Computing on International Conference on Computer Systems
and Technologies (CompSysTech '09),
Ontology has been introduced to utilize semantic relations, [12] Kakizawa, Y. (2007), “In-house use of Web 2.0: Enterprise
enhanced by inference. Blog entries are mapped to the 2.0”, NEC Technical Journal, Vol. 2 No. 2, pp. 46-9.\
ontology using language processing. Search results are [13] Kirkpatrick, M. (2006) The flu wiki: A serious application of
organized by introducing semantic aggregation. Blog entries new web tools.
are enriched by metadata and an annotation mechanism has http://marshallk.blogspot.com/2005/07/flu-wiki-
also been developed. The framework has been tested and serious-application-of-new.html (Last visited March 2011)
evaluated by implementing a system for the Institute of [14] Lamb, B. (2004) Wide open spaces: Wikis, ready or not.
EDUCAUSE
Technology domain ontology at VFU. Experiments have
Review,39(5) (September/October), 36-48.
shown quite good results. Single sample ontology is created http://www.educause.edu/ pub/erferm04/erm O452.
for demonstration. asp?bhcp=l

169
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

[15] McAfee, A.P. (2006), “Enterprise 2.0: the dawn of emergent

collaboration”, MIT Sloan Management Review, Vol. 47 No. 3, [26] Pettenati M.C., M. Ranieri, Informal learning theories and tools
pp. 1-28. to support knowledge management in distributed CoPs, E.
[16] McAfee, A.P. (2006), “Enterprise 2.0: the dawn of emergent Tomadaki and P. Scott (Eds.): Innovative Approaches for
collaboration”, MIT Sloan Management Review, Vol. 47 No. 3, Learning and Knowledge Sharing, EC-TEL 2006 Workshops
pp. 1-28. Proceedings, ISSN 1613-0073, p. 345-355, 2006.
[17] McFedries, P.: The Web, Take Two Technically Speaking. [27] Razmerita, L. (2005), “Exploiting semantics and user modeling
IEEE Spectrum, June (2006) for enhanced user support”, electronic version, Proceedings of
[18] Mentzas, G., Kafentzis, K. and Georgolios, P. (2007), HCI International 2005
“Knowledge services on the Semantic Web”, Communications [28] Rogers, E. (2011) How Web 2.0 Is Influencing E-Learning
of the ACM, Vol. 50 No. 10, pp. 53-8. http://ezinearticles.com ?HowWeb-2.0-Is-Influencing-E-
[19] Meyer, M. (2010) Creating an Open Web 2.0 Cloud E-Learning Learning&id=5968742 (Last visited March 2011)
Experience, 2010 Educause Annual Conference [29] Shakya A.(2006), A Semantic Blogging Framework for better
http://www.educause.edu/E2010/Program/SESS124 Utilization of Information, Thesis submitted in partial
(Last visited March 2011) fulfillment of the requirements for thedegree of Master of
[20] Mott, J. (2010) Envisioning the post-LMS era: the Open Engineering
Learning Network Educause Quarterly, Vol. 33, No. 1 http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.
[21] Nabeth, T., Angehrn, A. and Roda, C. (2002), “Towards 1.1.115.7250 &rep=rep1&type=pdf (Last visited July 2011)
personalized, socially aware and active knowledge management [30] Surowiecki, J. (Ed.) (2005), The Wisdom of the Crowds,
systems”, in Stanford-Smith, B., Chiozza, E. and Edin, M. Anchor Books, New York, NY
(Eds), Proceedings of E-business and E-work – Challenges and [31] Taylor, C. (2005). It’s a Wiki, Wiki World. Time 165(23): 40-
Achievements in E-business and E-work, Vol. 2, IOS Press, 42.
Amsterdam, pp. 884-91. [32] Vanderwal, T., (2005). Explaining and Showing Broad and
[22] Olson, C. (2006) New tools for learning. Narrow Folksonomies,
http://faculty.eicc.edu/golson/tools.htm (Last visited http://www.vanderwal.net/random/entrysel.php?blog=
March 2011) 1635 (Last visited March 2011)
[23] O'Reilly, T.: What Is Web 2.0. In Internet: [33] Wheeler, S, Kelly P., & Gale, K. (2005) The influence of
http://www.oreillynet.com/pub/a online problem- based learning on teachers' professional
/oreilly/tim/news/2005/09/30/what-is-web-20.html practice and identity. ALT-J 2005, 3(2):125-137.
[24] Oren Eyal , Max Vёolkel, John G. Breslin& Stefan Decker [34] Wiki, Matrix - Compare Them All. (2007) Retrieved April 19,
(2006), Semantic Wikis for Personal Knowledge Management, 2007 from
S. Bressan, J. KЁung, and R. Wagner (Eds.): DEXA 2006, http://www.wikimatrix.org/
LNCS 4080, pp. 509–518, 2006, Springer-Verlag Berlin [35] Will, Richardson, Blogs, Wikis, Podcasts and Other Powerful
Heidelberg Web Tools for Classrooms http://weblogg-ed.com/
http://www.johnbreslin.org/files/publications/20060919 [36] Will, Richardson, Learning on the Blog
_dexa2006.pdf http://www.corwin.com/books/Book235915
[25] Paul, C. and Schofield, A. (2010) e-Portfolios for teacher ?siteId=corwin-press&subject=C00 &q=richardson
candidates Vancouver BC: UBC Faculty of Education (retrieved
from http://ctlt.ubc.ca/2009/09/22/e-portfolios-for-
teacher-candidates/, December 6, 2010)

170
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Implementation of Web 2.0 in the Bitola Museum

- Successful Marketing Tool
Pargovski Jove1, Irena Ruzin2 and Aleksandra Lozanovska3
Abstract – This paper presents the Web 2.0 strategies and presentation of the cultural heritage in the Municipality of
implemented in the Bitola museum. The project started in 2010 Bitola and the nearest regions.
and it has already proved itself as successful marketing tool, It is the largest museum in South West region of Republic
resulting with increased number of visits, higher level of of Macedonia and first museum institution, awarded as “The
communication and friendlier image of the museum in the
public.
Best Museum” by Macedonian National Committee of ICOM
in 2010. The Municipality of Bitola also awarded the Museum
Keywords – Web 2.0, Museum, Social media, facebook as “The Best Cultural Institution in 2010”.
The main reason for these awards was the new established
permanent exhibition, which is among the richest museum
I. INTRODUCTION collections in Republic of Macedonia.
In the framework of the Institute and Museum of Bitola the
The permanent exhibition in the Bitola museum was following edifices are included:
renovated and reopened in the end of 2009. It is one of the - The Museum building - Old Barrack,
largest and richest museum collections in Republic of - Archeological site Heraclea Lyncestis,
Macedonia, chronologically describing the history of the - Art gallery (Yeni Mosque),
Bitola region. As soon the exhibit was presented, an - Hajdar Kady Mosque,
appropriate strategy was developed, toward the popularization - Memorial House of Goce Delcev
of the museum and its activities. - Memorial House of Stevan Naumov Stiv
The evolution of the internet technologies and introduction - Memorial museum in village Smilevo
of Web 2.0, have opened a whole new realm of opportunities - The Magaza gallery
towards the increased operational effectiveness of the
museums. Web 2.0 refers to the transition from static HTML In 2010 the museum started several innovative projects,
web pages to a more dynamic, second generation of the World with main goal is to increase its operational effectiveness. The
Wide Web, where people could collaborate and share digitization program, the modernized website and the social
information online. This introduces a new moment in the media activities, made this museum the first candidate from
current practices of the museum, influencing many changes in Macedonia, nominated for the prestigious EMA Micheletti
the decision making processes, influenced by the constantly award in 2012.
monitored public opinion.
Also another advantage of this kind of approach is more
“friendlier” image of the museum, resulting with increased
number of collaboration projects and cooperation with the
public on many levels.
According to many experts, the Bitola museum is the leader
in the area of innovative approach, aimed towards
popularization of the cultural heritage in Republic of
Macedonia.

II. NI INSTITUTE AND MUSEUM BITOLA

The Institute for protection of cultural monuments and Fig. 1. NI Institute and Museum Bitola
Museum Bitola is a national museum institution which main
goal is the protection, systematization, scientific processing III. MUSEUMS AND WEB 2.0
1
Pargovski Jove is with the Cultural Heritage Protection Office of It is a fact that museums are constantly struggling with the
Republic of Macedonia, Gjuro Gjakovic No 61, Skopje 1000, decline in the number of visits and reduced interest in their
Macedonia, E-mail: j.pargovski@uzkn.gov.mk. activities. The evolution of the internet technologies has
2
Irena Ruzin is program director at NI Institute and Museum
oversaturated the market with events and knowledge from all
Bitola, Republic of Macedonia, Kliment Ohridski bb, Bitola 7000
Macedonia, E-mail: irenaruzin@yahoo.com.
over the world, so an evident change can be recorded in the
2
Aleksandra Lozanovska is director at Gauss Institute – Bitola, people’s everyday lives. The reduced interest for the museums
Republic of Macedonia, Pitu Guli 27, Bitola 7000 Macedonia, exhibits and activities can also be contributed to the old
E-mail: director@gaussinstitute.org.

171
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

fashioned rigid approach of the cultural heritage presentation Internet—a more mature, distinctive medium characterized by
and the lack of appropriate marketing strategies. user participation, openness, and network effects. [3]
Many museum activities are created not having any kind of Basically Web 2.0 refers to the transition from static HTML
market research or user survey, but based only on the web pages to a more dynamic, second generation of the World
curator’s idea. This also is the case with the museum exhibits, Wide Web, where people could collaborate and share
where the curators often have an abundance of objects which information online.
they want to present in the exhibit, but there is always the There have been many successful examples where the
constraint of space and money. In some way this can be implementation of the new techniques, have increased the
categorized as “telling the curator’s version of the story”, quality of work of the museums. The Bitola museum is one
which can be good or bad but never complete. such example, this kind of strategy, resulted with many
How visitors are experiencing the exhibits and can they benefits for the museum.
contribute in some way, are the factors that are many times
neglected.
The modern technologies had in great deal influenced the
IV. THE BITOLA MUSEUM WEBSITE
peoples life’s and also have introduced many new ways of
cultural heritage presentation and visitor collaboration. The old website of the Bitola museum existed since 1999
According to ICOM, “Museum is non-profit, permanent and it was static HTML website offering only one way
institution in the service of society and its development, open presentation of the museum.
to the public, which acquires, conserves, researches, In 2006 for the first time in Republic of Macedonia, the
communicates and exhibits the tangible and intangible “Virtual exhibits” were presented, offering new approach
heritage of humanity and its environment for the purposes of towards the presentation of the cultural heritage. The virtual
education, study and enjoyment”. [1] exhibits have proved themselves as a successful tool, resulting
Having this definition in mind, the museums must use all with increased number of “online” and “real” visits to the
the available means and technologies in order to be successful museum.
in all previously mentioned areas. With the information age, With the introduction of the new museum exhibit and the
many museums along with all other type of information have necessity of its adequate presentation, new strategy was
gradually moved from real into virtual space. developed, based on the proven virtual technologies,
Many critics have argued that museums need to move from supported by the use of Web 2.0. The new website of the
merely supplying information to providing usable knowledge Bitola museum was presented at the beginning 2010.
and tools that enable visitors to explore their own ideas and
reach their own conclusions [2]. This argument seems
especially relevant today, when technology gives individuals
access to communication, information gathering and analysis.
The evolution of the internet technologies and introduction
of Web 2.0, is opening a whole new realm of opportunities
towards the increased operational effectiveness of the
museums.

Fig. 3. The web site of the Bitola museum

(www.bitolamuseum.org)

According to the website statistics, as predicted most

visited sections from the start were the virtual tours, photo
galleries, the permanent, temporary exhibitions and the
educational section.
It was also noticed that large quantities of website traffic
come from the social media, facebook especially. By analysis
of these visits, it was determined that users were also
significantly contributing to the quality of the published data.
People were viewing, sharing and commenting the various
Fig. 2. A tag cloud (a typical Web 2.0 phenomenon in itself) sections of the website, constantly contributing towards the
presenting Web 2.0 themes quality of the published content.
Web 2.0 is a set of economic, social, and technology trends
that collectively form the basis for the next generation of the

172
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

V. THE BITOLA MUSEUM ON FACEBOOK Soon after all the galleries and videos were transferred to
the facebook account, which made this account very popular
Being on the Internet is not just having a website. Almost and generated large circle of friends in a short time. This is
every company today has a Facebook profile or a Twitter also another advantage of the use of social media, creation of
account in order to stay in touch with its clients. With circle of friends with similar interests. The online friends are
estimated 1.2 billion users, the social media sites represent a not limited by the place and time barrier. For example city
significant tool for promotion of the companies, their products like Bitola has a very large scientific and nonscientific
and activities. community outside its borders, holding large quantity of
However, Web 2.0 is not only about social connections or knowledge and assets in different areas.
promotion. Constantly evolving, the Web 2.0 technologies Also trough the analysis of the comments, it was
provide new smart and interactive formats to show determined that in some areas of interest a large quantity
information. information was contained in the “normal” users also. Either
Social media is not just about opening up another marketing in form of historical fact they have witnessed, some person
channel. It enables customer participation on many levels. For they knew or a story they have heard.
museums, it offers various ways to support the museum
experience and even to extend it beyond the actual visit. It is
not about virtualizing the museum, but about developing new
models of participation and feedback [4]
With more than 830 million users Facebook is the most
used social media site, and it was the first choice for social
profile for the Bitola museum.
In Republic of Macedonia the number of facebook users in
March 2012 was 936 300, or 49,15 % of the total population
and 88,55 % of the internet users [5]. According to Social
Bakers, more than 50% of the facebook users are under the
age of 34, which is target group for many museums. It is
known fact that the younger population is spending less time
visiting the museums, so new approaches must be developed Fig. 5. Facebook profile of the Bitola museum
in order to popularize the cultural heritage.
Using the sport as always interesting topic, some photos
from the early days of the local soccer club “Pelister” were
published. The museum had those photos but with only small
legend describing their date and the clubs that played. Soon
after, some of the players or their close relatives and friends
were tagged, and in very short time the images were
completely described, involving all kind of information. Even
greater asset was the generated circle of friends tagged on the
photos, which can be used in future researches.

Fig. 4. User age distribution on Facebook in Macedonia

The facebook profile of the Bitola museum was created

the same time when the website was launched. At first it was
used as a tool for boosting the website traffic, but the
preliminary results uncovered its true potential.
Small test was conducted, where one photo gallery form
the website was shared as external link, compared to the same
photos from the gallery uploaded and shared as album on Fig. 6. Donated photo from the online friends
facebook. The first approach resulted in increased number of
visits on the website but with no comments. The facebook Figure 6 is good example of the benefits of using Web 2.0
album was consecutively shared, but also individual images strategies, since it is photo donated to the museum by
were tagged and commented. By analysis of the facebook facebook friend. The photo also carries the facebook tags of
album valuable information was gathered concerning the many members, friends and relatives of the team and
content of the photos, but also on the visitor’s behavior. For complete list of names of all team members.
example, some photos were shared and commented more than The facebook approach proved as complete success and
others, meaning they were more interesting to the public. soon after new quizzes and educational games were

173
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

developed, which embraced the user interactivity with the - 92 % wanted to continue with workshops and to be
museum. In some way this created more “friendly” image of involved for the next year
the museum, resulting with more users donating images to the - All of the participants had facebook profile and more
museum. than 70 % have shared their museum activity on their
During the campaign “Bitola then and now”, more than profile
thousand old photos were collected by the museum from the - More than 70 % of schools asked for extra workshop
online friends. More than 70 % of these photos were new for on other subject (art, geography…)
the museum and as a result of this campaign temporary - More than 80 % of the participants have never been in
exhibition was organized in occasion of the day of liberation a museum before the workshops
of Bitola – 4th November. - More than 40 % of the participants, claimed that their
Trough the online communication the museum gathered a close relative or friend have visited the museum on
large quantity of new information and managed significantly their suggestion after the workshop.
to increase its inventory trough donations and acquisitions. After the workshop a meeting was held with the
The online friends have accepted with great pleasure this participants, where the mutual experience was shared. Many
new approach and many of them have contacted the museum of the participants said they liked the Web 2.0 activities
concerning different matters. Based on this communication a implemented by the museum and today some of them are also
large quantity of new ideas have evolved, which are among the most active participants and supporters. According
constantly implemented by the museum. According to many to them, “The Bitola museum represents an excellent place for
experts, the Bitola museum is the leader in the area of cultural heritage education and sometimes fun by participating
innovative approach in Republic of Macedonia in the online quizzes”.
The number of visitors is increased thanks to the different
types of educational and interactive programs for all
categories and excellence promotion in the media and
VII. CONCLUSION
International fairs.
It is evident that the new internet technologies have
changed the people’s everyday lives. If the museums want to
VI. EDUCATIONAL WORKSHOP : “MUSEUM be successful in their work, they are left with only two
DETECTIVE” choices: to resist the changes and have only minor individual
efforts in this area, or to embrace the new technologies and
The museum of Bitola has significant experience in the make the required organizational change. The benefits of the
organization of educational games and activities for the second approach are evident in many fields of the museum
younger audience. In 2011 new project was conducted, aimed work and significantly increase the quality of work in the
towards the education and familiarization of the museum by museums.
the young audience. Those educational workshops are part of
long-term project “Learning about us” REFERENCES
The main goal of the ”Museum detective“ was to provoke
the participants to be interactive using different types of [1] http://icom.museum, International Council of Museums.
techniques and gain knowledge about the cultural heritage [2] Lynda Kelly, Ph.D., “How WEB 2.0 Is Changing the Nature of
(during the “game”). Museum Work”,
Before the task all participants attended lectures according http://museumstudies.si.edu/webcast_052110.html.
to their age and the possibilities for further research were [3] John Musser, Tim O’Reilly & the O’Reilly Radar Team “Web
presented to them. They were divided in groups and every 2.0 - Principles and Best Practices”
group had received a task to locate specific object from the [4] Vogelsang, A. and B. Minder. “Audience+: A Holistic
museum exhibition. The museum collection of the Bitola Approach to Developing Social Media Guidelines for Swiss
Museums. In J. Trant and D. Bearman (eds).” Museums and the
museum is among the richest in Republic of Macedonia with
Web 2011: Proceedings. Toronto: Archives & Museum
abundance of objects. Having in mind the competition aspect Informatics. Published March 31, 2011.
of the game, it was determined later that the participants who [5] http://www.socialbakers.com
did pre-game online research, were more successful than their
competitors. A large number of participants were involved in
these workshops and even greater interest is recorded for the
second phase planned for May 2012.
This project resulted with some interesting conclusions, or
expressed trough statistics:
- In the following 4 months 35 % more primary and
secondary school visits were recorded in the museum,
than the same period in 2010
- 78 % of the participants have visited the museum
before the workshops to learn more
- 92 % have used the museum web site before the
workshops

174
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Attacking the cloud

Vlad Andrei Poenaru1, George Suciu1, Cristian George Cernat1, Gyorgy Todoran1 and
Traian Lucian Militaru1
Abstract – Cloud is emerging as a new technology for serving the European research project “Cloud Consulting” [1]. We
web applications. There is little literature on how DDOS affects will use our platform hosted on several servers running
clouds and how other types of attacks fare versus a well designed Ubuntu Linux – Apache – MySQL template with current
cloud architecture. Is cloud resilient enough to be a solution or software release. On our cloud testing environment we
normal web architectures with good caching and reverse proxy is
going to be better always? This is what paper is trying to answer
provide the platform for processing information from
by testing two architectures, SlapOS and an architecture with hundreds different sensors, enabling the analysis of security
nginx in front and a few web servers behind it against different data through a large sample of logs.
types of attacks DDOS, slowloris, RA flood attacks. We demonstrate that open source cloud platforms are well-
developed and mature technologies offering a secure
Keywords – cloud, DDOS, cloud architecture, flood attacks. environment for deploying a growing number of applications.

II. PROBLEM FORMULATION

I. INTRODUCTION
SlapOS is an open source Cloud Operating system which
In this paper we present a study about distributed denial of was inspired by recent research in Grid Computing and in
service attacks (DDOS) in open source cloud platform particular by BonjourGrid [2] a meta Desktop Grid
SlapOS, the first open source operating system for Distributed middleware for the coordination of multiple instances of
Cloud Computing. This will include writing security testing Desktop Grid middleware. It is based on the motto that
scripts, collecting results and automating scripts for improving ”everything is a process”. SlapOS is now an OW2 project.
security of software deployment and configuration on cloud Fig. 1 shows the current architecture.
nodes.
We develop a test platform for cloud computing and use it
as a case study for testing and monitoring different security
threats. We use different types of attacks and monitor
important information such as CPU load, number of processes
and intrusion level from installed sensors. The sensors will
transmit intrusion detection data from our cloud platform in
real-time, display it in our web-based visualization application
and get detailed recommendations when and where security
threats did occur - resulting in optimized automating patching.
Also we will introduce in this article SlapOS, the first open
source operating system for Distributed Cloud Computing.
SlapOS is based on a grid computing daemon called slapgrid
which is capable of installing any software on a PC and Fig. 1. The SlapOS Architecture
instantiate any number of processes of potentially infinite
duration of any installed software. Slapgrid daemon receives SlapOS defines two types of servers: SlapOS Nodes and
requests from a central scheduler the SlapOS Master which SlapOS Master. SlapOS Nodes can be installed inside data
collects back accounting information from each process. centers or at home. Their role is to install software and run
SlapOS Master follows an Enterprise Resource Planning processes. SlapOS Master acts as a central directory of all
(ERP) model to handle at the same time process allocation SlapOS Nodes, knowing where each SlapOS Node is located
optimization and billing. SLAP stands for “Simple Language and which software can be installed on each node. The role of
for Accounting and Provisioning”. SlapOS Master is to allocate processes to SlapOS Nodes.
This structure has been implemented for cloud-based SlapOS Nodes and SlapOS Master exchange are
automation of ERP and CRM software for small businesses interconnected through the HTTP and XML based SLAP
and aspects are under development under the framework of protocol. SlapOS Master sends to each SlapOS Node a
description of which software should be installed and
executed. Each SlapOS Node sends to SlapOS Master a
1
The authors are with the Faculty of Electronics, description of how much resources were used during a given
Telecommunications and Information Technology at Politehnica period of time for accounting and billing purpose.
University of Bucharest, Bd. Iuliu Maniu, nr. 1-3, Bucharest 060042, From a user point of view, SlapOS Node looks like an
Romania, E-mails: vlad.wing@gmail.com, george@beia.ro,
online shop for Cloud Computing resources. The user
cernatcristi@gmail.com, todoran.gyorgy@gmail.com,
gelmosro@yahoo.com. connects to SlapOS Master through a simplified front end,

175
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

selects which software he or she needs. SlapOS Master then operating system agnostic and to automate application
allocates the software onto a SlapOS Node and provides the configuration process in a reproducible way.
connection information to the user. The allocated software can
be of any type: virtual machine, database server, application
server, web cache front end, etc.
From a developer point of view, as seen in Fig. 2, SlapOS is
a simple and universal API to create instances of any software
daemon through a programmatic interface.

Fig. 3. The SlapOS kernel

Each time slapgrid receives a request from SlapOS master

to run a software as a new process, it calls first buildout to
create all configuration files for that process then delegates to
supervisord the execution of the process. Supervisor is a
client/server system that allows its users to monitor and
control a number of processes on UNIX-like operating
systems. It provides a higher abstraction and flexibility than
traditional sysinit.
After some time, a typical SlapOS Node will include
multiple software applications and, for each software
application, multiple instances, each of which running in a
different process. For example, both Mediawiki and OS
Commerce could be installed onto the same SlapOS Node,
Fig. 2. An example of SlapOS front-end
with six instances of each being run as processes. By running
software instances as processes, rather than by creating a
A simple code allows a developer to request a new instance
virtual machine for each software instance as one would do
of a memcache server by invoking the request method of
with Amazon EC2 [6], SlapOS is able to use hardware
SlapOS API. Memcache [3] is a widely adopted key-value
resources and RAM in particular more efficiently.
store protocol which is used to cache values in large scale web
SlapOS Master runs ERP5 Cloud Engine, a version of
infrastructure. It is usually installed and configured by system
ERP5 open source ERP capable of allocating processes in
administrators using packaging systems such RPM or DEB. In
relation with accounting and billing rules. Initial versions of
this example, a single method call does in a few seconds what
SlapOS Master were installed and configured by human.
a human system administrator would have done in few
Newer versions of SlapOS Master are implemented
minutes at best.
themselves as SlapOS Nodes, in a completely reflexive ways.
A SlapOS Master can thus allocate a SlapOS Master which in
III. PROBLEM SOLUTION turn can allocate another SlapOS Master, etc.:
After running security testing scripts and collecting results,
SlapOS is implemented as an extension of widely adopted as shown in Table 1, we conclude that our open source cloud
open source software: GNU/Linux, Buildout [4] and platform delivers better performance in attacks against it.
Supervisord [5] and as depicted on Fig. 3. The only new
software introduced by SlapOS is Slapgrid, a daemon in TABLE I
charge of implementing the SLAP protocol on each SlapOS
Node.
Each time slapgrid receives a request from SlapOS master Architecture SlapOS cloud Nginx Web
to install a software, it downloads a description of that CPU load
software in the form of so-called buildout profile. It then runs (1) DDOS (1) 85% (1) 95%
the buildout bootstrap process to install the software. Buildout (2) slowloris (2) 90% (2) 99%
is a Python-based build system for creating, assembling and (3) RA flood (3) 60% (3) 80%
deploying applications from multiple parts, some of which attacks
may be non-Python-based. Buildout can be used to build C, Number of 200K 14K
C++, ruby, java, perl, etc. software on Linux, MacOS, processes
Windows, etc. (slowloris)
Buildout can either build applications by downloading their Exploits detected 5238/5545 5211/ 5545
source code from source repositories (subversion, git,
mercurial, etc.) or by downloading binaries from package
repositories (rpm, deb, eggs, gems, war, etc.). Buildout excels
in particular at building applications in a way which is

176
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

IV. CONCLUSION
SlapOS can be described as a cloud operating system in
which “everything is a process” unlike Unix in which
“everything is a file”. If one has to manage thousands of
servers with thousands of processes, hundred different
applications in multiple different releases or versions, SlapOS
can help you a lot by making the whole security management
process well specified, automated and under control.
Therefore cloud security is shared with the processes of the
applications running on the nodes.
The second result with SlapOS is that the best way to create
a reliable and secure cloud computing system is to follow the
original principles of the Internet: distribution and simplicity.
Our system can also help keeping track of exploit
development, optimize patching for zero-days threats and to
produce log auditing to improve security risk management.
As future work we envision an early warning system of
cloud attacks that applies intrusion prevention measures based
on sensor information from different partitions on the
distributed nodes.

ACKNOWLEDGEMENT
This paper is presented as part of the project “Valorificarea
capitalului uman din cercetare prin burse doctorale
(ValueDoc)” Project co-financed from the European Social
Fund through POSDRU, financing contract
POSDRU/107/1.5/S/76909 and part of the project “Cloud
Consulting”.

REFERENCES
[1] George Suciu, Octavian Fratu, Simona Halunga, Cristian
George Cernat, Vlad Andrei Poenaru, Victor Suciu, “Cloud
Consulting: ERP and Communication Application Integration in
Open Source Cloud Systems”, 19th Telecommunications Forum
- TELFOR 2011, IEEE Communications Society, pp. 578-581,
2011
[2] Heithem Abbes, Christophe C´erin, and Mohamed
Jemni.Bonjourgrid as a decentralised job scheduler. In APSCC
08.Proceedings of the 2008 IEEE Asia-Pacific Services
ComputingConference, pages 89–94, Washington, DC,
USA,2008. IEEE Computer Society.
[3] Memcached: a free and open source, high-performance,
distributed memory object caching system.
http://memcached.org/
[4] Buildout - software build system reloaded
http://www.buildout.org/
[5] Supervisor: A Process Control System http://supervisord.org/
[6] "The impact of virtualization on network performance of
Amazon EC2 data center", Tze Ng, Guohui Wang, IEEE
INFOCOM 2010 - 029th IEEE International Conference on
Computer Communications, Vol. 29, no. 01, March 2010

177
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

An Implementation of SMS Communication with Patients

in a Medical Information System
Ivica Marković1, Aleksandar Milenković2 and Dragan Janković3
Abstract – In order to achieve high quality in providing
healthcare services an appropriate communication between staff
and patients is required. For a large majority of patients SMS II. SMS COMMUNICATION – CONS AND PROS
communication with a healthcare institution is a good solution
because it uses simple and widely adopted technology. On the We must agree that SMS is, in terms of information
healthcare institution side, automating this communication can technology, relatively old standard (with the first officially
significantly reduce amount of work which needs to be done by transferred message in 1992.) [1]. It is also a fairly limited
the personnel. Therefore a carefully designed IT solution for this technology allowing users to transfer only 160 ASCII
purpose can help improving certain processes in a healthcare
institution.
characters. Actually most of modern mobile devices do allow
their users to use different character sets and multi-part
Keywords – Medical information system (MIS), SMS messages, but it is a good practice when designing a system to
communication, Patients scheduling. keep in mind the worst case scenario. And there are many
more characteristics of SMS which encourage us to use this
kind of communication integrated with a healthcare
information system. Here are listed some of them.
I. INTRODUCTION
A. Number of devices which support SMS
SMS (abbreviated from Short Message Service) stands for a
communication protocol which enables mobile phones users
According to data from [3], there are less than 2 bilions of
to send and receive short textual messages [1]. There are
Internet users and more than 5 bilions of mobile phone users
many different standards for exchange of textual messages
in the world registered in 2010. It is further estimated that
and they are continuously being developed and improved by
98% of those phones have text messaging capabilities.
adding new features (e.g. sending and receiving formatted
text, images, audio and video). Nevertheless, in the context of
this paper, text messaging typically refers to SMS, with its B. Geographic availability of mobile networks
160 ASCII characters limitation.
In this paper we firstly discuss the advantages of using The fact is that mobile phone networks cover significantly
SMS in communication between a healthcare institution and larger areas comparing to fixed phone networks and cable or
the patients by using SMS protocol. Then we analyze wireless Internet. Although Internet coverage is constantly
architecture and implementation of our system for automated growing, the advantage on the SMS side comes from its
sending and receiving of textual messages. Finally we minimal demands regarding the quality of the network. This is
describe one typical use case of our system. especially important in areas with low population density
It is important to notice that our system for automated where Internet service providers do not have economic
sending and receiving of textual messages is implemented as a interests to build their networks. In such areas quality of
part of medical information system Medis.NET. Medis.NET mobile phone network is often below the level required for
is being developed at Faculty of Electronic Engineering in Niš mobile phone conversation and SMS remains as the only one
and more details about the system can be found in [2]. way of communication.
Anyway, system for automated sending and receiving of
textual messages is a complete system itself and can be simply C. Simplicity of usage
used in conjunction with other medical information systems as
well as with other information systems which can be The most of mobile phone users agree that it is very simple
improved by adding SMS communication with their users. to type and send/receive a short textual message. Also, for a
number of these users it is too complicated to use Internet or
1
Ivica Marković is with the University of Niš, Faculty of send/receive an e-mail.
Electronic Engineering, Aleksandra Medvedeva 14, P.O. Box 73
18000 Niš, Serbia, E-mail: ivica.markovic@elfak.ni.ac.rs.
2 D. Immediacy of action and response
Aleksandar Milenković is with the University of Niš, Faculty of
Electronic Engineering, Aleksandra Medvedeva 14, P.O. Box 73
18000 Niš, Serbia, E-mail: aleksandar.milenkovic@elfak.ni.ac.rs. Most of the mobile users keep their mobile devices always
3 by themselves and turned on so the users can receive
Dragan Janković is with the University of Niš, Faculty of
Electronic Engineering, Aleksandra Medvedeva 14, P.O. Box 73 important information with the least possible delay and act
18000 Niš, Serbia, E-mail: dragan.jankovic@elfak.ni.ac.rs. accordingly. This is an advantage of SMS communication
over e-mail and web where most of the users are not always

178
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 1. System for automated sending and receiving of textual messages

online and healthcare institutions lack a possibility to quickly Core messaging service also maintains the data about
notify such users of some important information if needed. patients’ subscriptions to our system and stores them in a
small database on the same server.
E. Broadcast possibilities Third role of core messaging service is to choose
appropriate interface for sending messages. As shown in
Fig.1. there are two possible interfaces – SMS gateway and
It is possible to quickly and easily transfer information to a
GSM modem.
large number of users through SMS (for example to all
SMS gateway is a service provided by a GSM service
patients registered in a healthcare institution database). In this
case SMS outperforms traditional phone calls. provider company. SMS gateway can be exposed as a web
application or a web service and depending on that our core
messaging service accesses it by using HTTP for web
III. IMPLEMENTATION application or SOAP for web service.
GSM modem is a specialized type of modem which accepts
Our system for automated sending and receiving of textual a Subscriber Identity Module (SIM card), and operates over a
messages operates as a “middle layer” between existing subscription to a mobile operator, just like a mobile phone.
Medis.NET information system and patients (Fig. 1.). From the mobile operator perspective, a GSM modem
Left part of the diagram shows existing Medis.NET behaves just like a mobile phone. From our perspective GSM
information system. Medis.NET manages patient’s Electronic modem compared to standard mobile phone has some
Health Records (EHR), patient’s visits scheduling, work advantages (can be connected at the same time both to
scheduling for employees, maintains financial data regarding external power supply and to our server by a cable
provided services etc. All the data is kept in a central database connection).
and can be accessed by the employees of healthcare institution Provided with these two interfaces, our core messaging
through Medis.NET client application. service can make a choice which one to use based on
Central part of the diagram shows our system for automated configuration parameters. Usual configuration settings are
sending and receiving of textual messages. Core logic of our such that SMS gateway is used with a higher priority because
system is running as a service on application server node. This of its lower price per message and higher throughput. GSM
service listens to existing Medis.NET database for data modem is used only as a back-up interface in case of failure of
modifications. If an important data for a patient who is the SMS gateway.
subscribed to this service is modified then the service sends
appropriate text message to the patient (e.g. some biochemical
laboratory analysis for the patient are finished and service
IV. A TYPICAL USE CASE SCENARIO
sends some most important results in the message also
reminding him/her to collect full list of results at the In this chapter one interesting usage of our system for
automated sending and receiving of textual messages will be
healthcare institution).
described. We use it here to improve an existing system for
patients’ visits appointments.

179
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

the institution the patient can use our SMS service as shown
on Fig. 2.
1. Patient registers for our service by sending an SMS
containing service keyword and his Social Security
Number.
2. SMS gateway forwards the message to core
messaging service.
3. Core messaging service checks appointments and
position in waiting queue for this patient.
4. Core messaging service estimates time when the
patient will be examined and stores his registration in
messaging database.
5. Core messaging service creates a confirmation
message containing estimated time when patient’s visit
should start and information that he will receive one more
message 30 minutes before that.
6. SMS gateway forwards confirmation message to the
patient.
7. Doctor at healthcare institution examines patients
and removes examined patients from the waiting queue
using Medis.NET information system (Fig. 3.).

Fig. 2. Using system for patient notification

Medis.NET information system already has a sophisticated Fig. 3. Medis.NET application used by medical personnel shows
module for scheduling employees work time and appointing lists of scheduled patients, patients in the waiting room and
patient’s visits [4]. This system supports automatic generating examined patients
of work time schedule for employees and splits generated
work time into time slots for patients. Duration of time slot is
configurable, based on employee’s specialization and work 8. Modifications from Medis.NET client are transferred
area (e.g. 15 minutes for general practice medicine, 20 to Medis.NET database.
minutes for dentistry etc.). For most specializations and work 9. Core messaging service detects modifications in
Medis.NET database.
areas this approach works fine but there still some
10. Based on new data in the waiting queue, core
departments which make appointments for their patients on
daily bases and not for precise hour and minute. Also for messaging service makes new estimation of the waiting
some work areas time needed per patient can significantly time.
vary, for example from 15 to 30 minutes. 11. If new estimation of the waiting time is about 30
In such cases our system for automated sending and minutes core messaging service sends invitation message
which contains new estimation for examination start time.
receiving of textual messages can enable patients to quickly
check their appointment status, get estimation when they will 12. SMS gateway forwards invitation message to the
be examined and/or get an invitation message 30 minutes patient (Fig. 4.).
before they should come to doctor’s office.
Take for example a patient who already knows that he has
an appointment today, but healthcare institution can’t provide
him exact time of his visit. Instead of waiting several hours at

180
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

REFERENCES
[1] G. Le Bodic, Mobile Messaging Technologies and Services:
SMS, EMS & MMS, Chichester, West Sussex, England, John
Wiley & Sons Ltd, 2005.
[2] P. Rajković, D. Janković, V. Tošić, “A Software Solution for
Ambulatory Healthcare Facilities in the Republic of Serbia",
Healthcom 2009, Conference Proceedings, pp.161-168, Sydney,
Australia, 2009.
[3] P. Vrgović, I. Jošanov-Vrgović, B. Jošanov, “SMS Information
Service: Innovative Thinking for the Successful Solution”,
Management Information Systems, vol. 6, no. 3, pp. 22-28,
2011.
[4] I. Marković, S. Cvetković, D. Janković, “An Implementation of
a Scheduling Tool in a Medical Information System”, ICEST
2010, Conference Proceedings, pp.327-330, Ohrid, Macedonia,
2010.
Fig. 4. Invitation message received on patient’s phone [5] A. Stone, "Mobile scaffolding: an experiment in using SMS text
messaging to support first year university students", ICALT
2004, IEEE International Conference on Advanced Learning
Technologies Proceedings, Joensuu, Finland, 2004.
V. CONCLUSION [6] A. Dickinger, P. Haghirian, J. Murphy, A. Scharl, "An
investigation and conceptual model of SMS marketing",
HICSS-37, Proceedings of the 37th Annual Hawaii International
In this paper we presented our implementation of a system Conference on System Sciences, Hawaii, 2004.
for automated sending and receiving of textual messages. [7] L. Prabhakaran, W. Yan Chee, K. Chong Chua, J.
Motivated by existing systems in different areas such as e- Abisheganaden, W. Mun Wong, "The use of text messaging to
learning, e-banking, marketing etc. ([5], [6], [7]) we designed improve asthma control: a pilot study using the mobile phone
our system as an improvement of a medical information short messaging service (SMS)", Journal of Telemedicine and
system. Telecare, vol. 16 no. 5, pp. 286-290, 2010.
Use case presented here is planned to assist medical
personnel in scheduling patients’ visits and to reduce waiting
time for the patients. Our plans for future work include some
more services for patients based on the system for automated
sending and receiving of textual messages. One possible
extension is related to biochemical laboratory. When some
analyses for the patient are finished the service would send
most important results in the short message also reminding
patient to collect full list of results at the healthcare institution.
It is also possible to create a notification service which would
warn a patient that a period of time has expired since his/her
last visit to a doctor (for example 6 months after visiting
dentist) and remind patient to consider scheduling a new visit.
Another possibility is to create a service which would
broadcast warning messages about changes in weather
conditions to all registered patients with chronicle diseases
which could be affected.
Although we listed plans for extending our messaging
service in healthcare area, it is important to notice that the
service is not limited to this area and that it could easily be
coupled with other kinds of information systems.

181
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

A Comparative Analysis of Mobile AR Software with the

Application to the Archeological Site Medijana
Dušan Tatić1, Časlav Stefanović2, Dragan Stanković3
Abstract - As Augmented Reality is becoming more and more II. COMPARATIVE ANALYSIS OF AR SOFTWARE
popular, it is also getting different ways of use. A particular area
of application is in the field of digitalization of cultural heritage.
Consequently, there are numerous tools for the application
The development of mobile devices over the last few years
development in this area. This paper provides an overview and a has contributed to the implementation of the augmented
comparative analysis of the software for mobile augmented reality technology on these devices. Thus, the mobile
reality. Based on this analysis, we selected a software tool to augmented reality software has also undergone development,
augment and improve the archaeological sites, by integrating 3D which has facilitated the process during which users’, in an
models of the artifacts with images of the real environment. As a interactive manner (using images from the real world), search
case study, this paper presents the implementation details for the for information (without the use of a keyboard and mouse).
design of the fountain in the audience part of the imperial palace Studying the field of augmented reality from the aspect of
at the archaeological site of Medijana, Niš, Serbia.
development and implementation of AR applications we can
Keywords – Augmented Reality (AR), AR browsers, Mobile AR identify two groups: AR browsers and AR software
SDK, Cultural heritage development kits.

I. INTRODUCTION A. AR browsers

Augmented Reality (AR) is defined as a real-time direct or This group includes AR browsers (ARBs), that are,
indirect view of a physical real-world environment that has applications of augmented reality aimed at helping the search
been enhanced by adding virtual computer-generated of the real environment by using mobile devices cameras. By
information to it [1]. Improvements in computer following the search of the images of the real environment,
graphics technology have lead to the world of virtual they are augmented with multimedia contents (html, image,
reality becoming more and more real, while the audio, video or 3D models). AR browsers index the content
possibilities of combinations with real (actual) worldly through media streams (termed channels, layers, or worlds).
environment opens a new dimension in the understanding of ARBs, generally, accesses remote resources using web
the concept of reality (the real world). In cases where the protocols and web services.
combination of real space and virtual elements is needed, the All ARBs work in a similar way and consist of the
practical application indicates that there are no fixed limits. following components (Fig. 2.):
According to Milgram [2], the possible combination states 1. The ARB represents the client on the mobile device.
are shown in Fig. 1. Through the implementation of virtual 2. The publishing website - a website on which developers
objects into a real environment we create the augmented can register new channels (layers, worlds), and manage their
reality. channels and accounts.
3. The ARB server is the basis of this system and represents
the interface towards other components (ARB, the publishing
web site, and external servers).
4. The external server – a platform on which are stored all
the multimedia contents (images, text, audio, and video) as
well as the description of the location (POI - point of interest).
Fig. 1. Milgram’s reality-virtuality continuum [2].

Augmented reality technology (AR) has achieved great

progress over the last few years. The ever-increasing
implementations of augmented reality are a result of the
development of both hardware and software solutions. In this
paper we show the implementations of augmented reality in
the field of cultural heritage by using mobile devices.
1
Dušan Tatić, Faculty of Electronic Engineering, University of
Niš, Serbia, e-mail: dusan@dragongroup.org
2
Časlav Stefanović, Faculty of Electronic Engineering, University
of Niš, Serbia, e-mail: caslav.stefanovic@gmail.com
3
Dragan Stanković, Faculty of Technical Science, Kosovska
Mitrovica, Serbia, e-mail: sfsgagi@gmail.com Fig. 2. AR browser system overview.

182
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

On the basis of previous research [4], shown in Fig. 3., we formed in the case of the Layar browser. It supports 2D multi-
carried out an analysis, of the three most frequent AR media contents. This browser uses its own ARML language
browsers from the first group: Layar, Junaio, and Wikitude. (Augmented Reality Markup Language) which is based on
KML (Google) and enables the addition of more data onto the
POI including addresses, telephone-numbers, links, different
icons and images. Using the ARML database we can
exchange data between the client (the mobile device), the
Wikitude server, and external servers. This ARB can be used
on iOS and Android platforms.

B. AR software development kits

The second group of software includes software

development kits (SDKs) used to create mobile applications
running on iOS and Android devices, providing mobile AR
functionality. This software is by its structure and
Fig. 3. Usage of AR browsers [4]. functionality significantly more complex than the software
from the first group. Unlike the AR browser, the AR software
A.1 Layar AR browser development kit enables greater freedom in the development
of applications.
Layar [5] was one of the first ARB that appeared on the From this group we will present two types of AR SDK
market, designed for the Android and iOS operating systems. software: metaio Mobile SDK and Vuforia SDK.
It enables both location based and vision based tracking. This
browser, the creation of channels also known as layers is
enabled. POIs are shown, in the form either of an icon or a 3D B.1 metaio Mobile SDK
object, on the display of the mobile device. By clicking on a Using this tool we can develop applications both for the
POI it is possible to obtain additional information, to activate Android and the iOS platforms. This software uses fast and
some of the multimedia contents or launch another modular libraries and algorithms since it was designed to be
application. The Layar browser gets data on the layer channel used on mobile devices. It offers location based, marker-
with the JSON structured data. based, and markerless 2D and 3D object tracking. In addition,
the metaio Mobile SDK [8] enables both QR Code and
A.2 Junaio AR browser Barcode scanning. With the help of this software, the AR
browser Junaio was also developed. Fig. 4. shows the metaio
Junaio [6] is a powerful AR browser which offers the Mobile SDK architecture.
support for location based tracking and both marker and
markerless image recognition. For location based tracking,
except for the outdoor version, an indoor version was also
created, that is, one which supports AR even in those
locations where GPS is not available and where the compass
does not work. In order to determine the user’s position in
these conditions, we use the Latitude Longitude Altitude -
Markers (LLA Markers).
Location based channels show points of interest. Virtual
objects floating at the position of a POI. These virtual objects
can take the form of: texts, thumbnails, or can be animated
and static 3D objects. Each of these POIs can be linked to
other images, sounds, videos, and websites. Fig. 4. Metaio Mobile SDK architecture [8].
GLUE (image based) channels are used for vision tracking
and enable one to attach or "glue" virtual 3D models to any The metaio Mobile SDK can be used in different ways
real object. The Junaio server transforms markers into depending on the type of application. For simpler applications
tracking xml which then helps the Junaio browser to perform we use the Application Layar. Applications can be configured
an optical analysis. To describe layers and POIs we use a based on an XML configuration file. The high level API
specific XML code. development enables the realization of complex applications.
At this level, knowledge of programming is required. This
A.3 Wikitude AR browser SDK platform also allows access to low level components.
This approach is used in the case of special demands, when
The Wikitude World browser [7] is used for location based some other 3D graphics engine is required, as well as in
tracking. POIs are grouped into worlds, just like layers are connection with capturing and tracking modules.

183
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

B.2 Vuforia SDK B. Implementation

Qualcomm [9] has developed a platform for augmented Taking into consideration the characteristics of the
reality known as Vuforia. Vuforia SDK supports the Android described software systems, we have opted for the Junaio AR
and iOS platforms. It uses natural-feature tracking and frame browser. Using this AR browser we have designed an
markers. In addition to these characteristics, Vuforia also application for the archaeological site Medijana. The
supports the option of virtual buttons, which registers the application was designed so that when the visitor, while
contact between the user and the marked parts of the marker. touring the archaeological site, using the camera on the
The tracking of objects which are rectangular in shape or mobile device, detects a mosaic, the screen of the device
something with a similar shape is also made possible. For the shows a 3D model of the Roman fountain in the exact location
purpose of tracking the Vuforia SDK via one’s Target where it was once located (Fig. 6.).
management system, a dataset configuration XML file is In order to realize the application, we used tracking: NFT
created. The Vuforia uses OpenGL ES for rendering 3D (natural feature tracking), that is, vision based markerless
objects. The Vuforia has an AR extension for Unity that tracking. In our case the tracking object is the image of the
enables vision detection and tracking functionality. mosaic.
The AR view content represents a 3D model of the
fountain. The model of the fountain is given in the *.obj
which is encrypted with the help of the online Junaio Model
Encrypter. These AR content is stored on the external server.
The structure of the application follows the one given in
Fig. 2. On the external server we generated a program code
(php) which is used to connect with the Junaio server. This
code enables the execution of other processes: the
authentication process, the selection of multi-media contents
as well as the transfer of the necessary parameters (XML).
The creation of channels is provided through the user
Fig. 5. System overview of Vuforia SDK [9].
interface on the Junaio server. We selected the Junaio GLUE
channel. When forming the channel, we defined the other
III. EVALUATION AND IMPLEMENTATION necessary data: the URL of the external server, the link for the
homepage, and refresh time to search the request to the
A. Evaluation external server. It is possible to define the channel for it to be
After experimenting in practice with the software briefly either of the type public or private. For additional searches of
discussed above, a comparative analysis of AR browsers and the channel the necessary tags are defined.
AR software development kits can be summarized as in Table This application is experimentally verified on the tablet
I. The most important elements of the software are shown: Asus transformer with the Android Honeycomb platform.
Platform – the system platform on which the described
software is implemented. Tracking – The type of tracking
which can be supported by the given software. Vision based –
tracking which is based on the natural features (NFT) or
classical (fiducial) markers [10]. Location based – this type of
tracking is realized with the help of the GPS sensor and
compass which are the constituent parts of a mobile device.
AR view content – the type of multi-media contents used to
augment reality. 2D means classic multimedia contents (html,
image, audio, video) while the 3D refers to the possibility of
displaying 3D objects. 3Dani marks the possibility of
displaying 3D animations.
We can conclude that the Layar and Junaio, judging by the
given elements, are very similar. They can be implemented on
the same platforms and support the same type of tracking
except that Junaio supports location based tracking in the
closed space where GPS is not available. From Table 1 we Fig. 6. Application example.
can note that Wikitude supports only location based tracking Augmented Reality is one of the upcoming technologies the
and only has the possibility of 2D contents. usage of which becomes simpler thanks to the development of
In the case of AR software development kits Metaio and mobile devices. The described application realizes a new
Vuforia, they support the same platforms. We can note a quality of the presentation of museum exhibitions. By using
difference can only be found in the case of the location based augmented reality technology is it possible to perform the
tracking which, for now, is not supported by Vuforia AR restructuring of the archaelogical sites and objects which do
SDK. not exist and whose reconstruction is complex.

184
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

TABLE I
A COMPARATIVE ANALYSIS OF MOBILE AUGMENTED REALITY SOFTWARE

AR browsers
platform tracking AR view content
name vision based location based
marker marker 2D 3D 3D ani
outdoor indoor
based less
Layar

Junaio

Wikitude

AR software development kits

Metaio Mobile SDK

Vuforia AR SDK

IV. CONCLUSION REFERENCES

In this paper we carried out an analysis and a comparison of [1] J. Carmigniani, B. Furht, M. Anisetti, P. Ceravolo, E. Damiani,
mobile augmented reality software. An application was and M. Ivkovic, Augmented reality technologies, systems and
designed for the virtual projection of the fountain in the applications, in: Multimedia Tools and Applications, pp. 341-
auditorium of the emperor’s palace at the archaeological site 377, 2011.
of Medijana. We have shown one of the possible ways for [2] P. Milgram, H. Takemura, A. Utsumi, and F. Kishino,
displaying cultural and historical heritage at the Augmented reality: A class of displays on the reality-virtuality
archaeological sites. continuum, Telemanipulator and Telepresence Technologie,
By using these applications, the desire and interest for the vol. 2351, pp. 282–292, 1994.
acquisition of new knowledge regarding cultural heritage or [3] R. Azuma, A Survey of Augmented Reality, In Presence:
history can be increased. Learning is made more complete and Teleoperators and Virtual Environments, vol. 6, pp. 355–385,
extensive at the very location, enhancing the learning process. 1997.
In order to use this technology it is necessary to perform the [4] Augmented Reality Browser Survey, Jens Grubert, Tobias
digitalization of cultural heritage, which contributes to its Langlotz, Raphael Grasset, Institute for Computer Graphics
preservation. and Vision Graz University of Technology, Austria , 2011.
This type of systems offers the visitor the possibilty of
selecting his own choice of additional contents regarding the [5] Layar, http://www.layar.com/browser/, Mar. 2012.
object being viewed, in accordance with the available time [6] Junaio, http://www.junaio.com/develop, Mar.2012.
and his personal interests.
[7] Wikitude, http://www.wikitude.com/developer/wikitude-
Further research on this topic should be focused on the use augmented-reality-for-developers, Mar. 2012.
of more inclusive software from the AR software
development kit group for the realization of multi-media [8] Metaio Mobile SDK, http://www.metaio.com/software/, Mar.
guides based on AR technology which could replace standard 2012.
audio guides. [9] Vuforia AR SDK, https://ar.qualcomm.at/qdevnet/, Mar. 2012.
[10] Michael Gervautz, and Dieter Schmalstieg, Anywhere
ACKNOWLEDGMENT Interfaces Using Handheld Augmented Reality, Computer,
Feb. 2012.
The work presented here was supported by the Serbian
Ministry of Education and Science (project III 044006).

185
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

A comparative analysis of dynamic programming

languages for application in multi-agent systems
Ana Stankovic1, Dragan Stankovic2, and Dusan Tatic3
Abstract – In recent years the interest for dynamic Nevertheless, definition provided by Wooldridge and Jennings
programming languages has risen together with increasing [3], [4] is becoming widely accepted by a growing number of
popularity of Web 2.0 applications. Many web frameworks based researchers, which is why it can be regarded as being one of
on popular dynamic programming languages such as Rails the most complete definitions. According to this definition, an
(developed in Ruby), Grails (Groovy) or Django (Python), were
agent is:
created in order to improve the efficiency of web applications
development by promoting Agile methodology of work and “a computer system that is situated in some environment,
simple, maintainable code. In this paper we have analysed if the and that is capable of autonomous action in this environment
same efficiency gains can be achieved in development of multi- in order to meet its design objectives”.
agent systems (MAS). In our analysis we have observed the Interest in studying multi-agent systems usually stems from
quality and size of the code written in dynamic programming the interest in artificial (software or hardware) agents, such as
languages: Groovy, Python, and Ruby, by comparing it with the the agents living on the Internet, for example. Examples of
code written in statically typed Java programming language. The those agents are trading agents, game-playing agents that
analysis is based on independent implementations of the assist or replace human players in multi-player games,
asynchronous dynamic programming algorithm in all four
autonomous robots in multi-robot environments and the like.
programming languages. Obtained results can be generalized to
other MAS algorithms. Software agents can be regarded as a natural extension of
the concept of software objects. Object-oriented programming
Keywords – multi-agent systems, dynamic programming has introduced abstraction entities – objects to the structural
languages. programming paradigm. Similarly, agent-based programming
introduces new entities – agents, which, in contrast to objects,
have an independent execution thread. Therefore, in
comparison to objects, agents have the ability of acting in a
I. INTRODUCTION goal-directed manner, for example, by interacting with other
agents, reading sensors or sending commands to effectors,
Dynamically typed programming languages have recently while objects only passively respond to procedure calls. In
turned out to be really suitable for specific scenarios such as short, it can be stated that agents represent intelligent,
Web development, application frameworks, game scripting, adaptable software applications, designed with the purpose of
interactive programming, rapid prototyping, dynamic aspect- meeting different, user-defined requirements.
oriented programming and any kind of runtime adaptable or In most of the cases, even separate action of agents can be
adaptive software. The main benefit of these languages is the useful. Nevertheless, agents achieve their fullest potential by
simplicity they offer to model the dynamicity that is interacting with other agents, thus making multi-agent
sometimes required to build high context-dependent software. systems. Most of these systems are heterogeneous because
Common features of dynamic languages are meta- they consist of different types of agents that have different
programming, reflection, mobility and dynamic functions within the observed system. Agents act either in
reconfiguration and distribution [1]. Out of special interest to synergy with the purpose of achieving the common goal or
us when considering dynamically typed programming competitively with the purpose of achieving contradictory
languages in the context of multi agent systems is rapid goals.
prototyping with additional benefit of adaptability which is What follows are detailed definitions of important concepts
not the subject of this paper. associated with agents and multi-agent systems. After that,
Russel and Norvig in [2] define artificial intelligence as a comparison of one simple MAS-used search algorithm
scientific study of agents that are able to perceive the implemented in dynamic programming languages Ruby,
environment and perform actions. Since the research of multi- Python and Groovy with statically typed Java programming
agents is still in its infancy, there is no universal consensus on language will be performed.
an unequivocal definition of the concept of agent.

1
Ana Stankovic is with the Faculty of Information Technologies at
II. CONCEPT OF AN INTELLIGENT AGENT
Metropolitan University, Tadeusa Koscusca 63, 11000 Belgrade,
Serbia, E-mail: ana.stankovic@metropolitan.ac.rs It has already been pointed out that there is no universally
2 accepted definition of the term agent. Debates concerning this
Dragan Stankovic is with the Faculty of Technical Sciences,
University of Pristina, Kneza Milosa No. 7, Kosovska Mitrovica, issue are still under way. In fact, while there is a general
Serbia, E-mail: sfsgagi@gmail.com consensus that autonomy is something that is always
3
Dusan Tatic is with the Faculty of Electronic Engineering, associated with agents, not all the details have been cleared
University of Nis, Aleksandra Medvedeva 14, 18000 Nis, Serbia, E- up. Perhaps the main reason for which one universally
mail: dusan@dragongroup.org

186
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

accepted definition is difficult to find lies in the fact that intelligent assistants, e-trade, production and modeling of
agents are used for finding solutions to problems within business processes [7], [8].
various domains. Therefore, for some applications, the ability
of agents to learn on the basis of their previous experience is
very important. However, for some other applications learning
III. DYNAMIC PROGRAMMING LANGUAGES
is not only unnecessary but also undesirable at times (an
example supporting these facts is found in the air-traffic In the last few years, the development of Web 2.0
control system; the passengers would probably not like the applications has brought about the increase of interest in
situation in which the system modified flights schedule at run dynamically typed programming languages. Great number of
time on the basis of previously learned facts). frameworks that enable efficient development of web
Agents can be regarded as an approach to structuring and applications and promote Agile application development
development of software that offers certain advantages and methodology have been developed. Terms such as DRY
that is suitable for certain types of applications (some papers (don’t repeat yourself), KISS (keep it simple but not simpler)
see agents as evolutionary in relation to objects) [5]. Agents’ and convention over configuration have been adopted by the
characteristics to reduce the interdependence of application programmers and these stand for the major directions the
components can represent their most advantageous programmers follow in the process of application
characteristics. Agents are autonomous, which can be development. Growing interest in programming languages
regarded as some kind of encapsulation [5]. While objects that increase productivity of web developers has resulted in
have their own methods that are controlled by external further expansion of their use in the desktop applications
entities, agents do not allow external entities to control them. domain making languages such as Ruby, Python and Groovy
When an agent gets a message, being autonomous, it decides extremely popular nowadays. What follows is a short
what is to be done with that message by itself. description of these programming languages and their
Interdependence of application components is reduced not important features that improve efficiency of rapid
only by the agents’ autonomy, but also by their robustness, prototyping of multi-agent systems algorithms.
reactivity and proactiveness. For example, when an agent
enters goal-directed phase, agent itself is responsible for the A. Ruby
process of realization of that goal. It is not necessary to
perform constant supervision and checking. Analogously, Ruby is a dynamic programming language which is
object can be regarded as a reliable employee that has no characterized by a complex but very expressive grammar and
initiative or sense of responsibility; supervision of that a good core class library with a rich and powerful API. Ruby
employee requires increased level of communication. On the is based on elements of Lisp, Smalltalk and Perl, but its
other hand, agent can be regarded as an employee that takes grammar is such that C and Java Programmers find it easier to
initiative and has the sense of responsibility. Therefore, learn. Ruby is a programming language that is completely
supervision of that employee does not require increased level object-oriented, but it is also suitable for procedural and
of communication, which is why it can be stated that there functional programming styles . Ruby includes powerful
exists lower level of interdependence. metaprogramming mechanisms and can be used for the
Reduced interdependence leads towards software systems creation of new languages which are suitable for certain
that are more modular, more decentralized and more easily domains or for the creation of DSLs (Domain Specific
changeable. This resulted in the fact that agents started being Languages). [9]
used in wide specter of applications, especially in applications
that are regarded as open systems, that is, applications which
B. Python
have been designed and written by different authors without
their mutual communication. Of course, this entails the
Python is a dynamic, object-oriented programming
introduction of certain standards. Examples of these systems
language which can be used for various forms of software
include semantic web and grid computing.
development. It offers strong support for integration with
The fact that some agents are proactive and reactive makes
other languages and tools, huge range of standard libraries and
their mode of problem solving similar to human. That feature
can be learned in a few days. Many programmers who had
resulted in a great number of applications in which agents are
had experience with Python programming reported substantial
used as substitutes for humans within some limited domains.
productivity gains and easier and more maintainable code
One such example is an application in which software agents
development. [10]
are used to replace human pilots in military simulations [6].
Another example are computer games. The game Black &
White uses agents that are based on BDI (Belief-Desire- C. Groovy
Intention) model. Another field within which agents have
been practically applied is the film industry. Producers of the Groovy is a developing dynamically typed programming
film Lord of the Rings used the software package Massive to language for the Java Virtual Machine. It builds upon the
generate armies of orcs, elves and humans. Each individual strength of Java but it also possesses additional features that
character was modeled as an agent. Other types of are inspired by languages such as Python, Ruby and
applications where agents show their advantages include Smalltalk. It supports DSLs and test driven development. Its

187
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

main advantage lies in the fact that it smoothly integrates with solution will be considered here, in which each node performs
Java objects and libraries. In fact, “Groovy is Java and Java is a local computation with insight only into the state of
Groovy”. Groovy is the second referent language for Java neighboring nodes. The principle of optimality underlies the
platform (Java programming language being the first), which solutions that will be illustrated: “if node x belongs to the
further explains its relation with Java [11]. shortest path from s to t, then the part of the path from s to x
(or from x to t) must also be the shortest path between s and x
D. Support for DSL creation (that is, x and t). This principle enables an incremental divide-
and-conquer procedure, also known as dynamic programming.
Let h*(i) represent the shortest distance from any node i to
Common feature of all languages that have been described
the goal node t. In that case, the shortest distance from i to t
so far is that they can create DSLs, that is, new programming
languages that enable more efficient development of via node j neighboring i is shown as: f*(i, j) = w(i, j) + h*(j),
applications for specific domains. This feature is important and h*(i)=minj f*(i,j). Having these facts in focus,
ASYNCHDP algorithm has each node perform the procedure
because it enables, for example, the creation of a specific DSL
shown in Fig. 1. Within this procedure, each node i maintains
for the domain of multi-agent systems, which, in turn, enables
a variable h(i) that stands for an estimate of h*(i).
more efficient modeling and development of multi-agent
It can be proved that ASYNCHDP procedure always
applications without losing the interoperability of the code
written in that new DSL with standard libraries. converges to the true values, that is, that h will converge to h*.
In this case, convergence will require additional step for each
node in the shortest path, which means that in the worst case
IV. DISTRIBUTED PATH FINDING PROBLEM convergence will require n iterations. However, this is not so
good for realistic problems. Not only will convergence be
Majority of problems that occur within multi-agent systems slow, but this procedure also assumes the existence of agent
are focused on how to meet some global constraints in a for each node. In typical search spaces it is not possible to
distributed way, that is, how the agents can optimize some enumerate all nodes in an efficient way and allocate each of
objective function in a distributed manner. In most cases, it is them a separate process. (For example, chess has
achieved with the help of four families of techniques and approximately 10120 positions). For that reason, programmers
specific problems. Those techniques are:
- Distributed dynamic programming (applied here to the procedure ASYNCHDP (node i)
path planning problem)
- Distributed solutions for Markov Decision Problems if i is a goal node then
(MDP – Markov Decision Problems) h(i)  0
- Algorithms of optimization algorithms of economic
else
functions (matching and scheduling problems)
- Coordination on the basis of social laws and conventions initialize h(i) arbitrarily (e.g. to  or 0)
(example of traffic regulations) repeat {
With the purpose of illustration, distributed dynamic
programming will be applied to the path planning problem. forall neighbors j do
Path planning problem consists of a weighted directed graph f(j) w(i, j) + h(j)
with a set of n nodes N, directed links L, a weight function w:
h(i)  minj f(j)
LR+ and two nodes s, t N. The goal is to find a directed
path from s to t that will have minimal possible total weight. }
Generally speaking, a set of goal nodes TN can be
considered, and the shortest path from s to any of the goal
Fig. 1. Asynchronous dynamic programming algorithm
nodes tT can be looked for.
This kind of abstract framework can be applied in various often turn to heuristic versions of the procedure that require
domains. It can certainly be applied in cases of some specific smaller number of agents.
networks (for example, transportation or telecommunication
network). Nevertheless, it can be applied to other problems as
well. For example, in a planning problem the nodes can be VI. RESULTS
states of the world and the arcs can be the actions that the
agent performs. In that case, the weights stand for the cost of We have implemented the above mentioned algorithm in
each action (for example, the time needed for the action) (37) Java, Groovy, Python, and Ruby and used the number of lines
(38). of code as a measure for evaluating their rapid prototyping
abilities. It can be further discussed whether the number of
lines of code is a measure that can be suitable for the estimate
V. ASYNCHRONOUS DYNAMIC PROGRAMMING of efficiency of some programming language in a specific
domain (multi-agent systems). The lines of code will depend
The problem of finding the best path is the problem that has on developer’s experience with certain programming language
been thoroughly studied in computer science. Distributed and the applied code style rules. The code that we used here

188
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

and the way it was generated are sufficient for the process of REFERENCES
drawing general conclusions. One such conclusion is that
there is a big difference between statically typed Java and [1] F. Ortin, “Type Inference to Optimize a Hybrid Statically and
dynamically typed languages that were considered here. The Dynamically Typed Language”, The Computer Journal, vol. 54,
graph shown in Fig. 2 illustrates this difference in the best No. 11, 2011
possible way. [2] S. Russell, P. Norvig, “Artificial Intelligence - A Modern
Approach”, Prentice Hall, 2009.
[3] M. Wooldridge, “An Introduction to MultiAgent Systems”,
Chichester : John Wiley & Sons, 2002.
[4] M. Wooldridge, N. Jennings, “Intelligent Agents: Theory and
Practice”, Knowledge Engineering Review, 10(2): pp. 115–152,
300 1995.
[5] M. Wooldridge, “Intelligent Agents. In: G. Weiss (Ed.),
196 Multiagent Systems. A Modern Approach to Distributed
200 Artificial Intelligence”, The MIT Press, Cambridge,
Massachusetts, pp. 27-78, 1999
84 88 [6] G. Tidhar, C. Heinze, M. Selvestrel, “Flying together: modelling
78
100 air mission teams”, Applied Intelligence. 1998, Vol. 8, 3, pp.
195–218.
[7] W. Shen, D. Norrie, “Agent-based systems for intelligent
0 manufacturing: a state-of-the-art survey”, Knowledge and
Information Systems, An International Journal. 1999, Vol. 1, 2,
Java Groovy Python Ruby pp. 129–156.
[8] N. Jennings et al., “Autonomous agents for business process
LoC test application
Fig. 2. Lines of code (LoC) for the complete
management”, International Journal of Applied Artificial
Intelligence. 2000, Vol. 14, 2, pp. 145–189.
that relies on AsynchDp algorithm for different programming
[9] D. Flanagan, Y. Matsumoto, “The Ruby Programming
languages
Language”, O'Reilly Media, 2008.
[10] [Online] http://www.python.org/
[11] [Online] http://groovy.codehaus.org/
VII. CONCLUSION

In the domain of rapid prototyping of MAS algorithms

dynamically typed programming languages clearly have
advantages over classic statically typed languages, such as
Java. These advantages, as we have shown, are noticeable
even in the simplest examples. Apart from the advantages that
are reflected in reduced number of code lines and higher
productivity, big advantage is also seen in the increased code
readability and subsequent easier influx of broad community
of developers in what was previously done, as well as in more
efficient code maintenance and iterative improvement.
The conclusion that can be drawn is that dynamically typed
programming languages should be given preference over
statically typed languages whenever possible.
The future will probably bring increased interest in
functional programming languages in the domain of multi-
agent systems as well. Environments and tools with the most
efficient support and broadest community of users will
become dominant while the remaining projects will disappear
in time.

ACKNOWLEDGEMENT

The work presented here was supported by the Serbian

Ministry of Education and Science (projects III44006 and
III42006).

189
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

GPU Accelerated Construction of Characters of

Finite Abelian Groups
Dušan B. Gajić1 and Radomir S. Stanković1
Abstract – In group theory and Fourier analysis on finite using the graphics processing unit (GPU). This choice of
Abelian groups, the group characters are an essential concept. hardware is made due to the fact that contemporary GPUs are
In many applications, as for instance, spectral processing of logic highly parallel computing engines which can simultaneously
functions (binary or p-valued), it is often required to construct serve as programmable graphics processors and scalable
the table of group characters for the specified group. This can be
parallel computational platforms [1, 8, 13]. For a given group
a computationally demanding task, both in terms of space and
time, when dealing with large groups, since the group characters G, the construction of group characters can be expressed in
are viewed in matrix notation as rows of (pm × pm) matrices, terms of the Kronecker product of characters of its subgroups
where p is the cardinality of the set where the given logic of smaller orders. In this formulation, the algorithm for the
function and its variables take values, and m is the number of construction of group characters expresses a substantial
variables. The graphics processing unit (GPU), as a highly inherent parallelism and, therefore, the GPU is a natural
parallel computational platform, may facilitate this complex choice of hardware for the implementation of this algorithm.
task. The experimental comparisons of the proposed
This paper discusses the application of the GPU processing to implementation on the GPU and the C/C++ implementation of
the construction of tables of group characters for finite Abelian the same algorithm processed on the central processing unit
groups represented as a direct product of cyclic subgroups of
order p. We exploit the Kronecker product structure of these
(CPU) confirm this assumption.
tables permitting redistribution of the related computing tasks The rest of the paper is organized as follows. The
over GPU resources. Experimental results confirm that the background theory is introduced in Section 2. In Section 3, we
presented solution offers a considerable speed-up over the C/C++ propose a mapping of the algorithm for the construction of
implementation of the same character construction method group characters to the GPU and discuss the details of the
processed on the central processing unit (CPU). respective programming implementation. The experiments are
discussed in Section 4. We close the paper with Section 5, by
Keywords – Abstract harmonic analysis, finite Abelian groups, presenting some conclusions and possible directions for
group characters, Kronecker product, GPU computing, further research.
OpenCL.

II. BACKGROUND THEORY

I. INTRODUCTION
In this section, we give a brief introduction to the
Abstract harmonic analysis is a mathematical discipline theoretical background of the paper. For more detailed
that evolved from the classical Fourier analysis by the discussion of these topics, we recommend classical works
replacement of the real group R with an arbitrary locally such as [3, 15, 17], or more recent references [6, 9, 14].
compact Abelian or compact non-Abelian group [3, 6, 9, 14, We consider finite Abelian groups of the
15]. This implies the transition from the exponential form G = C pm = ({0,1,..., p − 1}m , ⊕ p ) , where Cp is the cyclic
functions, used in classical Fourier analysis and viewed as the group of order p, and ⊕ p is the componentwise addition
group characters of R, to the group characters, in the case of
modulo p.
Abelian groups, and the group representations, in the case of
non-Abelian groups [6, 9]. Abstract harmonic analysis The group characters χω( p ) ( z ) , z = 0, 1,..., pm-1, of the group
provides foundations for the formulation of many methods G are defined as [9, 16, 17]:
with significant applications in electrical engineering and  2π m −1 
computer science [9, 16, 17, 18, 19]. In these methods, it is χω( p ) ( z ) = exp  i ∑ ωm −1− s zs  , (1)
often required to construct the group characters of various  p s =0 
Abelian groups and use them in further computations. With where i = −1 , ωs , zs ∈ {0,1,..., p − 1} , and
that motivation, this paper presents a method for an efficient m −1 m −1

construction of group characters of finite Abelian groups ω = ∑ ωs p m −1− s , z = ∑ zs p m −1− s . (2)

s =0 s =0

1
Dušan B. Gajić and Radomir S. Stanković are with the University Example 1 The group character tables, for the cyclic
of Niš, Faculty of Electronic Engineering, Aleksandra Medvedeva groups Cp of orders p = 2, 3, and 4, are given in Table I, where
14, 18000 Niš, Serbia, E-mails: dule.gajic@gmail.com,
i = −1 , e1 = −0.5 ⋅ (1 − i 3) = exp(2π i / 3) , and e2 = e1∗ =
radomir.stankovic@gmail.com.
−0.5 ⋅ (1 + i 3) = exp(4π i / 3) .

190
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

TABLE I CHARACTER TABLES OF CYCLIC GROUPS B. Algorithm Mapping

Cyclic The key task in porting algorithms to the GPU is their
C2 C3 C4
group efficient mapping to the SPMD (single program, multiple
1 1 1 1  data) processing model and the multi-level memory hierarchy
1 1 1 1 i −1 −i 
Character 1 1  1 e of GPUs [1, 2, 8, 12, 13]. In the GPU SPMD model, a single
1 −1  e2   
data parallel function called a kernel is executed over a stream
table  
1
1 −1 1 −1
1 e2 e1    of data by many threads in parallel. A thread is the smallest
1 −i −1 i 
execution entity and represents a single instance of the kernel.
The execution of the kernel is controlled by the host program
The group G= C pm is the direct product of m elementary
processed by the CPU.
cyclic subgroups Cp. It follows, see for instance [3, 6, 15], that The mapping of the algorithm for the construction of group
the character table of the group G is the Kronecker product of character tables to the GPU is explained using Example 2.
m character tables of its cyclic subgroup Cp.
The matrix C32  in (3) has the following block structure:
Example 2 For the group C32 , the character table can be
computed as the Kronecker product of two character tables of  B00 B01 B02 
its cyclic subgroup C3. In this way, only the character table of C32  =  B10 B11 B12  . (4)
 
C3 is computed through (1) and the character table for C32 is  B20 B21 B22 
generated as:
Blocks Bx,y (x, y = 0, 1, 2) are the character tables for C3
1 1 1  1 1 1  multiplied by the elements of the matrix [C3]. Therefore, each
C32  = [C3 ] ⊗ [ C3 ] = 1 e1 e2  ⊗ 1 e1 e2  = block can be represented as:
   
1 e2 e1  1 e2 e1   a00 a01 a02 
 1 1 1 1 1 1 1 1 1  Bx , y = cx , y ⋅ [C3 ] = cx , y ⋅  a10 a11 a12  = cx , y ⋅  ai , j  , (5)
  
1 ⋅ 1 e1 e2  1 ⋅ 1 e1 e2  1 ⋅ 1 e1 e2  
 a20 a21 a22 
 1 e1  1 e2 e1  1 e2 e1  

e2
 where cx , y , ai , j ∈ {1, e1 , e2 } , x, y, i, j = 0,1,2.
 1 1 1 1 1 1 1 1 1  To each block we assign a thread t = (x, y, ai,j), x, y, i, j = 0,
1 ⋅ 1 e1 e2  e1 ⋅ 1 e1 e2  e2 ⋅ 1 e1 e2   . (3) 1, 2. Each thread performs a multiplication of [C3] by a scalar,
  as in (5). Threads are organized into a two-dimensional (x, y)
 1 e2 e1  1 e2 e1  1 e2 e1  
  array corresponding to the matrices to be computed. Fig. 1
 1 1 1 1 1 1 1 1 1  represents the mapping of the character table computations to
1 ⋅ 1 e1 e2  e2 ⋅ 1 e1 e2  e1 ⋅ 1 e1 e2   the GPU threads. Each thread processes a single block, which
   is indicated by a different color in Fig. 1.
 1 e2 e1  1 e2 e1  1 e2 e1  
This property of the character table will be exploited in the
mapping of the computation of the character table to the GPU.

III. GPU CONSTRUCTION METHOD

A. GPU Computing
The technique of performing general-purpose algorithms on
graphics processors, known as GPGPU (general-purpose
computing on GPUs) or GPU computing, has recently become
a subject of a fast growing research interest and practical
application [1, 13].
This interest is mainly the result of two factors. First is the Figure 1. Mapping of the computations to the GPU threads for
evolution of the GPU hardware towards a scalable, Example 2.
programmable, and highly parallel computing platform [1,
13], and the second is the development of the Nvidia CUDA For the group C32 in Example 2, we have nine threads in the
[13] and OpenCL (Open Computing Language) [10] first and only step of the algorithm (since this example
programming frameworks, based on the C/C++ language, involves only one Kronecker product), each performing the
which made the immense GPU computational resources more operation from (5) in parallel. In this case, indices of memory
accessible. For the implementation purposes, we use OpenCL, locations, where a thread t(x, y, ai,j) stores the first element
since it allows the development of the code that is both
( cx , y ⋅ a0,0 ) of the block, are computed as:
accelerated and portable across heterogeneous processing
platforms (GPUs, FPGAs, DSPs) [8, 10]. startElement ← x ⋅ 33 + y ⋅ 3 . (6)

191
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Indices of memory locations for the rest of the elements 1. Host program, which executes on the CPU and creates
( cx , y ⋅ ai , j , x, y, i, j = 0, 1, 2, except for the case i = j = 0) in a and controls the context for the execution of kernels as
computed block are determined as: well as allocates and transfers data to the GPU
memory.
nextElement ← startElement + i ⋅ 32 + j . (7) 2. Device program, which is processed on the GPU and
The results of the computations are stored in the GPU implements the SPMD kernels.
global memory which has a linear layout. Formulas for the In the presented OpenCL implementation, the host program
computation of the memory location indices ((6) and (7)) lead determines the character table for the cyclic subgroup Cp
to the GPU global memory access pattern which is, for through (1). Notice that not all of the characters of Cp need to
Example 2, depicted in Fig. 2. Coloring of the blocks and the be computed by using (1), since, e.g., e p −i = ei ∗ , for i = 1, 2,
memory locations in this figure corresponds to the thread
coloring in Fig. 1. …,  p / 2  − 1 . Thus, we compute half of the rows of the
character table for Cp, while other rows are determined by
using this property.
The host allocates GPU global memory space for two (pm ×
m
p ) matrices that are used as buffers to store the results of the
application of the Kronecker product. This minimizes the
communication between the host and the device, which is a
bottleneck in the GPU computing [8, 12, 13]. Note that we
have to reserve the space for (pm × pm) matrices at the
beginning of the computation, since the size of the GPU
buffers cannot be changed after their creation, otherwise, we
would have to create buffers and transfer data between the
host and the device for each step of the algorithm, as the
Figure 2. GPU global memory access pattern for Example 2. resulting intermediate matrices increase in size. To minimize
the memory bandwidth occupation on the GPU itself, we use
In the general case, in the kth step of the algorithm, we the technique of buffer swapping [7]. For odd-numbered
perform the Kronecker product of a (pk × pk) matrix by the steps, the first matrix is used as the input to the kernel and the
(p × p) matrix, and the result is a (pk+1 × pk+1) matrix. second matrix as the output. For even-numbered steps, the
Therefore, there are p2 active threads in the first step of the order is reversed.
algorithm, while in the kth step, there are p2k active threads. The character table for Cp is stored in a (p × p) matrix and it
The index of the GPU memory location for the first entry is used as the second operand in the Kronecker product
( cx , y ⋅ a0,0 ) of the block is determined as: operation in each step. Since it is of a small size, we keep it in
the constant GPU memory, which is cached. This allows
startElement ← x ⋅ p k + 2 + y ⋅ p, (8) much faster access and leads to improved program
The indices of the memory locations for the other elements performance [12].
( cx , y ⋅ ai , j , i, j = 0, 1,…, p-1, except for the case i = j = 0)) in a The Algorithm 1 presents a pseudo-code for the device
program. Code in lines 2 and 6 implements (8) and (9),
block are:
respectively. Since the characters of finite Abelian groups are
nextElement ← startElement + i ⋅ p k +1 + j ⋅ p . (9) complex numbers, elements of C pk  , C p  , and C pk +1  are
C. Features of the Mapping stored in the GPU buffers using the float2 OpenCL vector
data type [10]. The first component in the vector variable
The proposed method for computing the character tables stores the real part and the second component the imaginary
has the following features: part of the complex number.
1. The character table is stored as a vector of length p2m
obtained by the concatenation of rows of [ C pm ]. This Algorithm 1 Pseudo-code for the device program
allows reading the values of characters directly without 1: x, y ← acquire thread indices in the two-dimensional grid
any reordering. 2: startElement ← x · pk+2 + y · p
3: adr1 ← x · pk + y
2. Elements of [ C pm ] computed by threads with the same 4: for i = 0 to p-1 do
first index and the successive second index are stored 5: for j = 0 to p-1 do
in neighboring memory locations. This automatically 6: nextElement ← startElement + i · pk+1 + j
allows memory coalescing, due to which multiple data 7: adr2 ← i · p + j
k +1
accesses to the GPU global memory are performed as a 8: 
C p  (nextElement).re ← C p  (adr1).re · C p  (adr2).re –
k

single memory transaction [2, 12].

C pk  (adr1).im · C p  (adr2 ).im
D. Algorithm Implementation 9: C pk +1  (nextElement).im ← C pk  (adr1).re · C p  (adr2).im +
A GPGPU program consists of two parts: C pk  (adr1).im · C p  (adr2 ).re

192
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

TABLE II SPECIFICATION OF TEST PLATFORMS

Platform A B
AMD Phenom II N830 Intel Core i7-920
CPU
triple-core (2.1GHz) quad-core (2.66GHz)
RAM 4GB DDR3 1066MHz 12GB DDR3-2000
OS Windows 7 Ultimate (64-bit)
IDE MS Visual Studio 2010 Ultimate
SDK AMD APP 2.6 Nvidia GPU Computing 4.0
GPU ATI Radeon 5650 Nvidia GTX 650 Ti
engine speed 650 MHz 900 MHz
memory 1 GB GDDR3 800 MHz 1 GB GDDR5 4.2 GHz
processors 80 384

Figure 3. Computation times for the groups C3m , m = 1, 2,…, 8, on

I. EXPERIMENTAL RESULTS CPUs and GPUs for the test platforms specified in Table II.
The experiments reported in this section are performed
using two hardware platforms, labeled A and B, respectively, REFERENCES
and specified in Table II. The GPU kernel performance
analysis is done through the application of AMD APP Profiler [1] T. M. Aamodt, “Architecting graphics processors for non-
2.4 (for A) and Nvidia Parallel Nsigth 2.1 (for B), in graphics compute acceleration”, in Proc. 2009 IEEE PacRim
Conf. Comm., Comp. & Sig. Proc., Victoria, BC, Canada, 2009.
accordance with instructions provided in [2, 12]. [2] AMD, “AMD Accelerated Parallel Processing OpenCL
The referent C/C++ implementation uses the complex data Programming Guide”, available from: http://developer.amd.com
type from the Standard Template Library (STL) for the /sdks/AMDAPPSDK, [accessed 1 April 2012].
representation of the values of group characters. This data [3] T. Apostol, Introduction to Analytic Number Theory, Springer-
structure best corresponds to the float2 OpenCL vector data Verlag, New York, USA, 1976.
type [10] used for the same purpose in the GPU [4] M. Clausen, “Fast generalized Fourier transforms”, Theoretical
implementation. The referent C/C++ implementation is Computer Science, No. 67, 1989, pp. 55-63.
compiled for the x64 platform using the MS C++ compiler set [5] J. W. Cooley and J. W. Tukey, ”An algorithm for the machine
to the maximum level of performance-oriented optimizations. calculation of complex Fourier series”, Mathematics of
Computation, No. 90, 1965, pp. 297-301.
The results of the experiments performed on both test [6] D. S. Dummit and R. M. Foote, Abstract Algebra, John Wiley &
platforms for the construction of the character table for the Sons, 2003.
groups C3m , m = 1, 2,…, 8, are presented in Fig. 3. Notice that [7] D. B. Gajić, R. S. Stanković, "GPU accelerated computation of
for p = 3 and m = 8, the size of the character table is pm × pm = fast spectral transforms", Facta Universitatis - Series:
38 × 38 = 6561 × 6561, and, therefore, to complete the task, we Electronics and Energetics, Vol. 24, No. 3, University of Niš,
Niš, Serbia, 2011, pp. 483-499.
have to compute and store 43 046 721 complex numbers. The [8] B. R. Gaster, L. Howes, D. Kaeli, P. Mistry, and D. Schaa,
OpenCL implementation processed on the GPUs outperforms Heterogeneous Computing with OpenCL, Elsevier, 2011.
the referent CPU C/C++ implementation on both platforms [9] M. G. Karpovsky, R. S. Stanković, and J. T. Astola, Spectral
and for all values of m used in the experiments. The speed-up Logic and Its Applications for the Design of Digital Devices,
is almost constant throughout the range for m, and it goes up Wiley-Interscience, 2008.
to a factor of 7.8 × , on the test platform A, and up to a factor [10] Khronos,”OpenCL Specification 1.2”, Khronos OpenCL
of 8.2 × on the platform B. Working Group, 2011.
[11] D. K. Maslen and D. N. Rockmore, “Generalized FFTs – A
survey of some recent results”, in DIMACS Workshop in Groups
II. CONCLUSIONS and Computation, 1998, pp. 183-238.
In this paper, we propose a method for the construction of [12] Nvidia, “OpenCL Best Practices Guide”, available from:
http://developer.nvidia.com/nvidia-gpu-computing-
characters of finite Abelian groups of the
documentation, [accessed 1 April 2012].
form G = C pm = ({0,1,..., p − 1}m , ⊕ p ) , using the graphics [13] J. Owens, M. Houston, D. Luebke, S. Green, J. Stone, and J.
processing unit (GPU) as the computational platform. We Phillips, “GPU computing”, Proc. of the IEEE, Vol. 96, No. 5,
identify the sources of the parallelism available in the 2008, pp. 279–299.
[14] C. C. Pinter, A Book of Abstract Algebra, Dover, 2010.
algorithm for construction of the character table for G
[15] W. Rudin, Fourier Analysis on Groups, Wiley, 1990.
formulated in terms of the Kronecker product. Based on this [16] R. S. Stanković, and J. T. Astola, Spectral Interpretation of
analysis, we devise a mapping of the computations to the Decision Diagrams, Springer, New York City, USA, 2003.
SPMD processing model of the GPU and develop an OpenCL [17] M. R. Stojić, M. S. Stanković, and R. S. Stanković, Diskretne
implementation of the algorithm. The experimental results transformacije u primeni, Nauka, Beograd, 1993, (in Serbian).
obtained through the comparison of the proposed solution and [18] M. A. Thornton, “Spectral transforms of mixed-radix MVL
the referent C/C++ implementation of the same algorithm functions“, in Proc. IEEE Int. Symp. on Multiple-Valued Logic
show speed-ups of up to 7.8 × and 8.2 × , depending on the (ISMVL), Tokyo, Japan, May, 2003, pp. 329-333.
platform, when using the GPU and, thus, confirm the validity [19] M. A. Thornton, R. Drechsler and D. M. Miller, Spectral
Techniques in VLSI CAD, Kluwer Academic Publishers, 2001.
of the proposed approach.

193
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Modern Processor Architectures Overview

Danijela Jakimovska1, Aristotel Tentov1, Goran Jakimovski1, Sashka Gjorgjievska1 and
Maja Malenko1
Abstract – Complexity of the modern processor architectures provide with the hardware, while VLIW processors with
and the constant race of the various computer technologies software support.
resulted in a wide range of computer architectures, each with its Architectures like Explicit Parallel Instruction Computing
advantages and disadvantages, but with the ultimate goal to (EPIC) and Data Flow were designed to solve some of the
increase overall computer systems performances. This paper
problems detected in the previous parallel architectures. EPIC
provides an overview of these architectures, including RISC,
CISC, Superscalar, VLIW, EPIC and Data Flow processor processors overcomes the hardware dependencies and
architectures. All of them are presented together with their furthermore, Data Flow machines provide concurrency in
drawbacks. The concept of parallelizing instruction's execution is program execution.
also described, in order to emphasize the goal for executing more
than one instruction in a single clock cycle, and in that way to
increase overall system performances. Moreover, examples are II. RISC, CISC AND SUPERSCALAR
given of commercial implementations of the modern computer
architectures. A. CISC Architecture
Keywords – Processor architectures, RISC, CISC, ILP
CISC is designed to use complex instruction set which
makes the assembly languages closer to the operations and
data structures of the High Level Languages, [1]. CISC
I. INTRODUCTION instruction set is proposed to take advantage of microcode,
and is consisted of many variable-length instructions, which
Modern microprocessors are one of the most complicated can specify a sequence of operations. The CISC instructions
systems that have been ever created by the human beings. are characterized with complexity, in terms of instruction
Microprocessors have the main role in each system, since they formats and addressing modes, and therefore require serial
handle the instruction and data flow, control the (slow) decoding algorithms, [2]. The memory references are
communication with memory and external devices and thus usually combined with other operations, such as add memory
coordinate the whole system operation. Computer architects data to register. CISC processors generally have few registers,
face with the challenge to maximize computer performance, and some of them may be special-purpose, which restricts the
while retaining the cost, power and functional requirements. ways how they can be used, [3].000 Accordingly, the CISC
Regarding this, they should consider three aspects of architecture complicates the instruction's decoding and
computer architecture design, including: instruction set scheduling, and therefore is not very suitable for pipelining
architecture, organization (memory system, memory [4].
interconnect, internal processor), and hardware logic design. The family line of x86 processors is the leader in CISC
Therefore, in order to optimize the architecture design, an computing, used both as general purpose processors and for
architect must be familiar with many technologies, such as embedded systems. Using the x86 processors in embedded
compilers, operating systems, logic design and packaging. design is relatively novel approach and is mainly purposed for
One of the first computer architectures, such as the Intel IA- the 32- and 64-bit designs such as Intel’s Atom, VIA's Nano,
32, belongs to the Complex Instruction Set Computer (CISC) Athlon's Neo, and VIA's C7.
design which takes advantage of the microcode and is
consisted of a wider range of variable-length instructions. In B. RISC Architecture
order to reduce the complexity of the instructions, Reduced
Instructions Set Computing (RISC) was introduced, speeding The researches intended to improve the existing (mainly
up the process of decoding the instruction. CISC) architectures led to reducing the set of instructions and
Superscalar processor and Very Long Instruction Word their complexity, and thus RISC was designed. RISC
(VLIW) architectures were introduced in order to achieve instruction set consists of simple fix-sized operations, which
Instruction Level Parallelism (ILP) exploiting the pipeline are easy (quick) for decoding, and therefore suitable for
mechanism. The main issue in ILP is detecting and pipelining. Contrary to CISC, RISC instructions are less
overcoming data dependencies, which Superscalar processors complex, support simple addressing modes, and do not require
a microcode for their implementation, [2], [4]. RISC
1
D. Jakimovska (danijela@feit.ukim.edu.mk), A. Tentov processors have a relatively large number of general-purpose
(toto@feit.ukim.edu.mk), G.Jakimovski(goranj@feit.ukim.edu.mk), registers. RISC instructions reference to the main memory
S. Gjorgjievska (sashka@feit.ukim.edu.mk) and M. Malenko only via simple load and store operations. This is the main
(majam@feit.ukim.edu.mk) are with the Faculty of Electrical reason why the RISC processors are usually referred as load-
Engineering and Information Technologies, Dept. of Computer store architecture designs, [5].
Science and Engineering, Karposh 2, Skopje, Republic of Macedonia

194
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Widely known success stories behind the RISC architecture III. VLIW AND EPIC ARCHITECTURES
are AMD 29K, ARM, SPARC, Power/PC, and MIPS. ARM
takes the lead in embedded systems, such as smart phones and
tablet computers, due to its low cost and low power A. Very Long Instruction Word
consumption.
Unlike the Superscalar architecture, which relies on
hardware to detect and overcome data dependencies, Very
C. Instruction Level Parallelism Long Instruction Word (VLIW) processors, [3] use software
solutions (compilers) to mark independent instructions.
Further research led to an idea, which explains that splitting Usually compilers translate the code into intermediate
the work of a single processor to multiple processors will language, optionally optimize it, and as well generate machine
increase the productivity and will speed up the execution of code for the specific architecture, [6]. VLIW processors use
the instructions. The main setback here is that programs were compilers that inspect the source code and thus concentrate on
written and were meant to be executed in sequential manner, scheduling and optimizing the raw source code before
that is, one instruction at a cycle. This restriction of the translating it.
sequential execution came from the data dependencies Scheduling is done using two structures: control flow and
between the variables in a program. data flow graphs, [6]. Control flow graphs divide the source
Pipelining is a mechanism which enables instruction level code into basic blocks that must be sequentially executed
parallelism, since parallel instructions are executed in parallel (usually the delimiter is a branch or a label), [6]. The data
over multiple cycles. The theoretical increase of the flow graph shows the dependencies between registers within a
instruction execution speed is proportional to the pipeline basic block and those which are independent can be executed
length, [2]. However, there are three potential hazard in parallel, [6]. Here, the disadvantage is that a single basic
problems that can occur, during the pipeline execution of block usually contains four to six operations and therefore,
parallel instructions. Data hazards appear when an instruction limits the amount of parallelism that can be achieved. To
result depends on the previous instruction; structural hazards maximize parallelism, a technique called global scheduling is
happen when there is not enough hardware space for the used, where instructions are moved from one block to another,
parallel instructions execution and control hazards are result [6].
of the unexpected program counter change, [1]. There are Transmeta’s Crusoe ® and Texas Instruments 320C6x line
several mechanisms targeting these problems, and the of processors are the commercial breakthrough behind the
simplest solution is to stall the pipeline. VLIW architecture. Transmeta’s Crusoe processor uses hybrid
hardware-software implementation of the VLIW architecture.
D. Superscalar architecture It is a 128 bit architecture that uses Code Morphing Software
that detects and resolves data dependencies. The Code
RISC architecture is very suitable for exploiting parallelism Morphing software also implements routines for power
through pipelining, since all the RISC instructions are simple management and thermal dissipation, which makes this
and take roughly the same time to finish. The instruction processor ideal for mobile devices. The Texas Instruments
execute stage is usually the most time consuming pipeline 320C6x line of processors is a general purpose DSP processor
operation, so if the processor architecture employs multiple using the VLIW architecture, which is mainly used as a
execution units, it would always have a busy one, while the research processor. It has wide range of debugging tools and
others would be idle, [2]. As a solution to this problem, compilers available for research.
computer architects proposed a superscalar architecture,
characterized with parallel instructions execution on multiple B. Explicitly Parallel Instruction Computing
executing units. Although this architecture utilizes more
execution units simultaneously, the number of hazards is The explicitly parallel architecture was designed to
increased and the processor has to retire the instructions in overcome some essential limitations of the VLIW
program order if they are re-ordered (dynamic scheduling). architecture, such as hardware dependence. The EPIC
Besides that, handling branch operations becomes very architecture solves the hardware dependence problem, by
problematic, since a typical program executes a branch after defining several mini-instructions which can be combined in
each six or seven instructions. A possible solution would be to groups, depending on the template type field, [8]. Therefore,
utilize a special hardware branch predictor, which would the processors which are characterized with greater
predict the target with some probability, on behalf of the parallelism capabilities will simply exploit more bundles in
previous branch results. However, this approach has still some parallel. EPIC processors utilize dispersal technique to issue
disadvantages, such as speculatively wrong executed two bundles at a time, and split-issue mechanism if the mini-
instructions. The processor architects overcome this difficulty, instruction cannot be executed. Furthermore, the instructions
by extending the processor architecture with reorder buffer. can be predicted, reducing the cost of a branch operation.
This buffer is intended to store the results of speculatively However, the cost for wrong prediction is very high, because
executed instructions, and to update the real state after the branches appear very often. EPIC processors provide
correct instructions has completed, [1]. hardware support for the control speculation of loads and they
allow parallel issuing of multiple prioritized branch

195
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

operations. Speculative load failures are resolved with poison IV. DATAFLOW ARCHITECTURES
bits utilization, [1], [2].
The EPIC architecture doesn’t solve all the problems of the Concurrency is a major step in increasing computational
VLIW architecture. The difficulty with the increased code size performance, especially with today’s technological
(because of the empty slots in the bundles) still remains limitations. Dataflow architecture offers an attractive
unsolved. This has a negative impact on cache performance alternative to the conventional control flow architecture in
and bus bandwidth utilization. Other very important feature of providing concurrency in execution of programs. Execution of
each EPIC processor is the good compiler support, [1]. each dataflow instruction depends only on the availability of
This architecture has only one implementation, as part of its operands, which implies implicit synchronization of
the IA-64 processor architecture of Itanium family processors, parallel activities. There are no constraints in sequencing of
[1]. Intel Itanium architecture processors have been designed dataflow instructions, except for the conventional data
from the ground up to meet the increasing demands for high dependencies in a program.
availability, scalability and performance needed for high-end Data flow architecture differs from control-flow
enterprise and technical computing applications. In its core, architecture, by two basic principles: asynchronous operations
Itanium was designed to address a number of performance and functionality. Dataflow instructions are executed only
bottlenecks in computers, such as memory latency, control when all input operands are available (assuming hardware
flow dependencies and memory address disambiguation. It resources are also available), in contrast to control-flow model
enables the hardware to take advantage of the available which uses program counter for sequential ordering of
Instruction Level Parallelism and to provide the necessary instruction execution. The functionality rule implies that any
resources, while focusing on dynamic runtime optimizations. two enabled instructions can be executed in either order or
Precision Architecture – Reduced Instruction Set Computer concurrently, only if they don’t interfere with each other
(PA-RISC) [7] was originally designed as a 32-bit (don’t have data dependences), which implies parallel
architecture, intended to be easily scalable across a broad processing.
performance range, while providing for straightforward A dataflow program is represented as a directed graph,
migration of applications from existing systems. It was rather where named nodes represent instructions and links
conservative RISC design, but still competitive in terms of represent data dependencies among instructions, [9] [10].
speed, especially for simultaneous multiprocessing and Dataflow graphs can be described as machine language for
floating-point operations. dataflow computers. Data is conveyed from one node to
another in data packets called tokens. This flow of tokens
enables nodes (instructions) which depend on them and fires
them.

TABLE I
COMPARISON OF PRESENTED ARCHITECTURES

Architecture/ ILP Instructions per cycle Instruction format

Characteristics
RISC Yes, pipeline Depends on the pipeline
Fixed length,
depth (usually 4 or 5)
Usually 16,32,64 etc
CISC Yes, similar to superscalar Depends on the Variable length,
operation’s complexity
Complex operations
Superscalar Yes, Multiple pipelines Depends on the pipeline
Fixed length,
depth (usually 8 or 10)
Usually 16,32,64 etc
VLIW Yes, Fixed number of Depends on the number of
Fixed length, Multiple
instructions in the word, execution units (usually 4
instructions in one
Implicit parallelism or 8) word
EPIC Yes, variable, but limited Depends on the number of
Variable length,
number of instructions in execution units Multiple
the word, Explicit (6 to 8) microinstructions in
parallelism the instruction word
Dataflow Yes, variable, but limited Depends on the number of Packet format of
number of instructions (due execution/functional units instructions (PISC)
to operands availability and
hardware resources)

196
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Dataflow architecture can generally be divided into pure results in executing a program in parallel and better overall
dataflow architecture and hybrid dataflow architecture. Pure performance. Also, we need to further investigate the
dataflow firing rule says that an instruction can be executed as possibilities to narrow the gap between memory response time
soon as all input operands are available to it. It gives the and processors working frequencies. However, in order to be
dataflow model asynchronous behavior and self-scheduling of able to go forward and toward developing completely new
instructions. different processor architecture one must make thorough
Pure dataflow architecture is subsequently divided into investigation of existing processor architectures. That was the
static, dynamic and explicit token store architectures, while main reason for leading this investigation, and presenting the
the hybrid architecture utilizes some known control flow results of it within this paper.
mechanisms, [11]. Pure dataflow architecture executes a
program by receiving tokens, each containing data and tag,
processing instructions and sending out newly formed tokens,
REFERENCES
[9]. When a set of matched tokens (tokens with same tag) is
[1] John L. Hennessy, David A. Patterson, “Computer Architecture:
available at the execution unit, processing starts by fetching
A Quantitative Approach”, 2007
the appropriate instruction (with the same tag) from the [2] Nicholas FitzRoy-Dale, “The VLIW and EPIC processor
instruction store. The instruction is executed and the result is architectures”, 2005
generated, containing data and tag of the subsequent [3] William Stallings, “Computer organization and architecture
instructions which depend on it. designing for performance”, 8th edition, Prentice Hall, 2009
Pure dataflow architectures have some serious drawbacks, [4] Sivarama P. Dandamudi, “Guide to RISC processors: for
the major being is bad single thread performance. Other programmers and engineers”, Springer, 2005
problems are: the overhead produced by token matching, as [5] Prof. Vojin G. Oklobdzija, “Reduced Instruction Set
well as implementing efficient unit for matching tokens Computers”, 1999
[6] Zomaya, A.Y.H. , Parallel and Distributed Computing
(resolved to some extent in explicit token store architecture,
Handbook, McGraw-Hill, Ch. 21, Thomas M. Conte,
[12]). ”Superscalar and VLIW Processors”, 1996
Unfortunately drawbacks made it difficult to achieve direct [7] Hewlett – Packard, Precision Architecture :The Processor, HP
implementation of computers based on a pure dataflow model. journal, August 1986
For this reason, possibilities of converging dataflow and [8] Mark Smotherman, “Historical background for EPIC”, February
control-flow models were investigated and broad spectrum of 2011
hybrids (techniques and machines based on them) were [9] J. Silc, B. Robic, T. Ungerer, “Processor architecture: From
developed: threaded dataflow, course-grain dataflow, RISC Dataflow to Superscalar and Beyond”, Springer, 1999.
dataflow, dataflow with complex machine operations, [13], [10] A. L. Davis, R. M. Keller, “Data flow program graphs”, IEEE
Trans. On Computers, February. 1982.
[14].
[11] R. A. Iannucci, “Toward a dataflow/von Neumann hybrid
architecture”, Proc. 15th ISCA, May 1988.
V. CONCLUSION [12] G. M. Papadopoulos, “Implementation of a general-purpose
dataflow multiprocessor”, Tech. Report TR-432, MIT
Laboratory of Computer Science, Cambridge, Aug. 1988.
This stream line of processor architectures is highly [13] J. Silc, B. Robic and T. Ungerer, “Asynchrony in parallel
unlikely to die off. Other architectures are about to emerge, computing: From dataflow to multithreading, Parallel and
some similar to the previous architectures and other Distributed Computing Practices”, 1998.
completely different. The advance of technology requires new [14] R. Buehrer, K. Ekanadham, “Incorporating dataflow ideas into
and better processor architectures and exploitation of the von Neumann processors for parallel execution”, IEEE Trans.
concept of parallel computing. Overcoming the data On Computers, Dec. 198.
dependencies issue has to be done in order to achieve better

197
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The mechanism for flexible symbology in mobile GIS

Miloš Roganović1, Bratislav Predić2, Dragan Stojanović3, Marko Kovačević4

Abstract – This paper presents an innovative approach for of a coordinate geometry and a coordinate reference system. In
implementing visual portrayal of geographic features with point, general, a geometric object is a set of geometric points,
polyline (complex polylines) and polygon geometry for application
in mobile GIS applications. This approach relies on developed represented by their coordinates. Basic geometric objects are
XML styling language for defining custom styles and mapping points, polylines, and polygons [4].
layer subclasses to the specific style. Benefits of this approach are
extending basic styles and applying custom, user defined style to The importance of the visual portrayal of geographic data in
distinguish object that belong to the same layer. Visually separated GIS cannot be neglected. The skill that goes into portraying data
objects can emphasize a specific attribute value that is crucial for
fast identification of objects in time critical situations. Typical is what transforms raw information into an explanatory or
emergency situations require synergic response from various decision-support tool. Fine-grained control of the graphical
response team all accessing different aspects of the same geospatial representation of geographic features is a fundamental
data. Flexible and customizable style definition language is applied
to general vector data allowing separate field services to have
requirement for any professional mapping community. Allowing
custom visualization of shared community spatial data. user to define styling rules for visual portrayal of geographic
features requires the existence of a styling language that the user
Keywords – styling, GIS, XML, visualization. and GIS application can both understand [5].

I. INTRODUCTION The paper is organized as follows: Section II presents

previous work. Section III describes creating custom layer styles
A geographic information system (GIS) is special type of that can be applied to any vector layer. Section IV shows
computer-based information system tailored to store, process, flexible simbology mechanism in mobile GIS application.
and manipulate geospatial data [1]. The ability of GIS to handle Section V summarizes the achieved results.
and process both location and attribute data distinguishes GIS
from other information systems. It also establishes GIS as a
II. RELATED WORK
technology important for a wide variety of applications [2].
Mobile GIS application, as integral part of complete GIS
The fundamental information unit that GIS deals with is solution in the domain of defense and security, is developed on
called a geographic feature. Geographic feature is an abstraction basis of GinisMobile platform [6]. It is capable of visualizing
of a real world phenomenon associated with a location relative geospatial data in the form of raster and vector layers [7]. All
to the Earth [3]. Every feature may have a number of properties. data required for specific field operation is prepared in advance
One or more of the feature's properties may be geometric. using desktop GIS part of the complete solution and deployed to
Geometry provides the means for quantitative description of the mobile devices using GIS server and standardized WMS and
WFS interfaces. Geospatial data being deployed to mobile
spatial characteristics of features, including dimension, position,
device is separated into two groups:
size, shape, and orientation. A geometric object is a combination • Data which is not of primary interest for specific
operation. Data in this group is not expected to change during
1
field operation and it will not be changed by specific field team
Miloš Roganović is with the Faculty of Electronic Engineering, using the mobile device or any other field deployed team. This
Aleksandra Medvedeva 14, 18000 Niš, Serbia, E-mail:
milos.roganovic@elfak.ni.ac.rs
data is referred to as background data and is visualized as static
2
Bratislav Predić is with the Faculty of Electronic Engineering, raster background map.
Aleksandra Medvedeva 14, 18000 Niš, Serbia, E-mail: • Dynamic data which is primary focus of specific field
bratislav.predic@elfak.ni.ac.rs team. Data from this group is organized as vector layers and
3
Dragan Stojanović is with the Faculty of Electronic Engineering, individual features of these layers can be modified by field team
Aleksandra Medvedeva 14, 18000 Niš, Serbia, E-mail: using the mobile device or any other field team involved in the
dragan.stojanovic@elfak.ni.ac.rs operation.
4
Marko Kovačević is with the Faculty of Electronic Engineering,
Aleksandra Medvedeva 14, 18000 Niš, Serbia, E-mail:
markko.marce@gmail.com

198
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Data from the first group mainly contributes to spatial

orientation and landscape/terrain recognition. In the domain of
defense, security and intelligence operations data from the
second group is organized into situational map and this shared
situation map becomes central hub for collaborative field
operations. One example of combined visualization of
background raster map with situational map overlaid is shown in Fig. 3. Selecting operational map elements
Fig. 1.

III. CREATE CUSTOM LAYER STYLE

While desktop applications offer more detail because of the
greater amount of available resources (memory, processing
power, and no restrictions on the battery), mobile application
stand out in a more efficient use in field conditions. With the
price of sensors recently declined significantly, sensors such as
GPS, accelerometers, and gyroscopes are incorporated as
standard in modern mobile devices. Application of mobile
applications that will be discussed in this paper refers to the
Geo-Information Systems. The paper will describe a mechanism
for flexible symbology of geospatial objects. Mobile GIS
application relies on the use of OGC defined standards.
Fig. 1. Combined maps in mobile GIS application
Mobile GIS applications are most frequently used in
All general purpose GIS operations are implemented and
navigation, but it is a narrow set of mobile GIS applications. In
supported. These include numeric coordinate transformations
between various (well known and manually defined) coordinate this paper we are focused on the use and modification of
systems, measurements (length, circumference, area etc.), geospatial data using mobile applications. Data can be in. shp.
zooming and panning map manually. Measurements and dbf format or any specific format, or may be obtained from the
cartometric functions are shown in Fig. 2. service.

Using this method it is possible to load data from OGC

standardized data source memorized in the local file in format
(.shp, .dbf). Also, data can be loaded from the service (WFS). It
is important to emphasize that the objects which are loaded are
vector objects whose location and other attributes can be
changed in real time.

The loaded data contains one or more attributes related to the

geometry of the object itself, and additional attributes that
describes detailed information of the object. Each layer of geo-
spatial data consists of data whose geometry is the same type
(eg. point, line, polygonal). After you load the data layer is
displayed in the default style.
Fig. 2. Basic GIS functions in mobile GIS application
For data that is loaded into the mobile application, later can
All operations are based on map view central point. User be applied created style. Style depends on the geometry type of
drags map and centers crosshairs on situational map elements. layer loaded. In this way, the mobile application styles are
Crosshairs shape depends on focused type of operational map
grouped into three classes: styles for point objects, line objects
element and is shown in Fig. 3.
and polygon objects. For point sources, we can define the
symbol color and shape and size of the spot object. An example

199
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

is given in Fig 4 a). In line objects can define the color and line IV. FLEXIBLE SYMBOLOGY MECHANISM IN MOBILE
type and line endings. An example is given in Fig 4 b). For GIS
polygonal objects, we can define the color and type of linje
which marks the edge of the object, and then we can define the The mechanism is based on XML-defined mapping of
color and type of fill. An example is given in Fig 4 c). attributes of mobile on visual representation of the object. It is
necessary to define a hierarchy or a subset of data for a given
layer. Once we define subsets of data for a given layer, it is
necessary to prepare the appropriate raster or texture for the
visual representation of instances of classes. In this way, all
objects are classified into a special classes of one layer, if you
look at climate change visualization of natural disasters can be
more detailed graphically represented with visualization of
measured values. Disaster management can greatly facilitate
services on the ground to determine the appropriate plan of
action.

As a data source can be any standard defined data source, you

a) b) need to configure mapping of one attribute from source to a
particular that can be used as symbol identifier. Each class in the
hierarchy is defined by selected attribute. This can be done
through a defined XML file that will be used for mapping or in
the application. For specialized display, source defined attribute
that serves to identify the symbols subclasses. Fig. 5 Shows
mapping process, on Fig. 5. a) all source attributes are displayed
unmapped. When user select an attribute, list of possible match
from hierarchy is displayed and user can perform mapping. If
source doesn’t have all attributes from hierarchy, then those
attributes are not considered during display process. Unmapped
attributes from source are added as additional set of attributes
beside hierarchy attributes, although they are not involved in
c) d)
display process.
Fig. 4. Setup style for whole layer based on layer geometry

On Fig. 4.d) is shown one object that belongs to layer

“National parks” of Serbia on which polygon style is applied.

Previously mentioned manner can indicate a particular layer

or vector data to distinguish between two layers of the same
geometry type, but has conceptually different meaning. All this
is enough for use in mobile applications in navigation. If you
need to show or indicate one type of points of interest (POI
Eng.), but what if you want the visually differences of data that
are belonging to the same layer. With that purpose a mechanism
has been developed for flexible symbology in a mobile GIS.
Fig. 5. Mapping an attributes from data source to specific domain
hierarchy

200
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

V. CONCLUSION
To improve the use of mobile applications, a mechanism for
flexible symbology of objects in mobile GIS applications has
created. This mechanism extends basic styles and improves
usability of mobile application. Big improvement has made in
creation custom styles that apply on each object based on
specified attribute value. Mechanism contains tree simple steps
to configure custom mapping. First step is to classify subsets of
objects for specified layer based on object attribute value.
Second step is creation of raster images or textures. Final step is
creating unique mapping. From user perspective, different kind
of services can apply their own simbology in order to visually
represent object from specific domain. This improvement allows
extending visual representation of same geo-spatial data
Fig. 6. Applying different style on one layer objects according to domain in which data are used.

Fig. 6. Shows four objects that belongs to same layer but they VI. ACKNOWLEDGMENTS
are visualized differently, based on previous mapping. Each
Research presented in this paper is funded by Ministry of
object has geometry that indicates location of object, other
Education and Science, Republic of Serbia as part of the
attributes describe objects in details. We configure mapping projects ‘Environmental Protection and Climate Change
between object attribute and symbol identifier. We create list of Monitoring and Adaptation’, Nr. III-43007 and ‘The
textures and include them in mapping. infrastructure for electronically supported learning in Serbia‘,
Nr. III-47003.
User can create object using mobile application and see
preview of the object based on previous mapping, shown on Fig.
REFERENCES
7. [1] Worboys, M., and Duckham, M., GIS: A Computing Perspective,
Second Edition, CRC Press, Boca Raton, FL, 2004.
[2] Chang, K., Introduction to Geographic Information Systems, Third
Edition, McGraw-Hill, New York, NY, 2005.
[3] The OpenGIS Abstract Specification, Topic 5: Features (Version
5.0), document 08-126, Open Geospatial Consortium Inc., January
2009, http://www.opengeospatial.org/standards/as
[4] OGC Reference Model (Version 2.0), document 08-062r4, Open
Geospatial Consortium Inc., November 2008,
http://www.opengeospatial.org/standards/orm
[5] Styled Layer Descriptor profile of the Web Map Service
Implementation Specification (Version 1.1.0), document
05-078r4, Open Geospatial Consortium Inc., June 2007,
http://www.opengeospatial.org/standards/sld
[6] Predic, B., Stojanovic, D., Djordjevic-Kajan, S., „Developing
Context Aware Support in Mobile GIS Framework“, 9th AGILE
Conference on Geographic Information Science, Visegrád,
Hungary, 2006
[7] Rančić, D., Predić, B., Dimitrijević, A. "Optimizations of raster
map visualization in mobile GIS", Proceedings of the 10th
WSEAS International Conference on COMPUTERS,
Vouliagmeni, Athens, Greece, July 13-15, 2006 (pp1191-1195)
ISSN:1790-5117, ISBN: 960-8457-47-5
Fig. 7. Creating new raster image for visual representation of object
with preview

201
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

GinisED tools for spatial analysis of electric power supply

network
Aleksandar Stanimirović1, Leonid Stoimenov2 and Danilo Vulović3
Abstract – GinisED is the geographic information system Pipelines such as water supply, sewage, power supply, heat
specially designed for electric power utility companies. ).GinisED supply, industrial pipelines and communication lines, are
was developed by CG&GIS Lab, Faculty of Electronic essential infrastructures in cities. This tendency has been
Engineering, University of Niš, and is deployed in ED Jugoistok increased in the last decade while rebuilding and substituting
Niš. A very important part of GinisED are tools for spatial
existing principal pipelines to fit in with increased demands of
analysis of electric power supply network, potential or actual
events in electric power supply network and risk factors for a citizens and industrial development. Moreover, newly
certain geographic area. In order to provide support for efficient designed underground pipelines for different purposes have
spatial analysis we have developed a specific data model for been built along with various engineering projects.
representing electric power supply network in GinisED. Our Insufficient, inaccurate and unclear information about the
data model is a combination of topological model (nodes and location and depth of cables and pipelines may cause various
edges for representing geographic information) and typical problems and may even result in tragic accidents [2].
graph data structure (for representing electric network Efficient functioning of utility companies engaged in the
attributes. transmission and distribution of electricity cannot be achieved
Keywords –GIS, power supply network, spatial analysis, data
without proper record keeping and monitoring of the
model transmission and distribution network system [3]. Therefore,
almost any electric power supply company has a need for the
existence of specialized GIS solution that should provide
mechanisms for collecting, storing and manipulating spatial
I. INTRODUCTION data concerning transmission and distribution network system.
This specialized GIS solution provides support for recording
Making decisions based on geography is basic to human company assets, their locations their condition, how they are
thinking because geographic location is an important attribute performing, and how much it costs to provide the service [4].
of all human activities policies, strategies and plans. By GIS tools for utility companies provide a very effective
understanding geography and people's relationship to location, tool for generating maps and statistical reports from a
we can make informed decisions about the way we live on our database. However, GIS functionality far exceeds the
planet. Geographic information systems (GIS) are a special purposes of mapping and generating reports. In addition to the
class of information systems that keep track not only of basic functions related to automated cartography and data
events, activities and things, but also of where these events, base management system, spatial analysis is the vital part of
activities, and things happens or exist [1]. every GIS solution for utility companies. GIS is used to
GIS enables capturing, storing, analyzing, and displaying collect spatial data concerning distribution network, spatial
geographically referenced information. It allows us to view, analysis should be able to answer complex questions
understand, query, interpret, and visualize data in a way that is regarding space.
quickly understood and easily shared. GIS technology can be
used for scientific research, resource management, and II. RELATED WORK
development planning. Because geography location is very
important for every day human activities, GIS solutions
The ability to take the geographic location of objects into
provide support for making decisions and solving many
account during search, retrieval, manipulation and analysis
different problems: environmental monitoring, transportation
lies at the core of a GIS. How well these tasks can be
management, public safety, facility security, disaster
accomplished is determined by the spatial data model. The
management, etc.
starting point for modeling of geographic information is the
Management and visualization of underground utilities
geographic feature. A feature is an abstraction of a real world
have been always of a great concern in many countries.
phenomenon. A geographic feature is a feature associated
with a location relative to the Earth. A digital representation
1
Aleksandar Stanimirović is with the Faculty of Electronic of the real world can be thought of as a set of features [5].
Engineering at University of Niš, Aleksandra Medvedeva 14, 18000 Among many of the commonly used GIS data models,
Niš, Serbia, E-mail: aleksandar.stanimirovic@elfak.ni.ac.rs. simple features model has the most simple data structure.
2
Leonid Stoimenov is with the Faculty of Electronic Engineering Every geographic feature has a unique persistent ID, set of
at University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, thematic attributes (represented as alphanumeric data) and a
E-mail: leonid.stoimenov@elfak.ni.ac.rs. number of geometric attributes. In this model, feature
3
Danilo Vulović is with the Faculty of Electronic Engineering at geometric attributes are limited to simple geometries. Simple
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E-
geometries are defined as 2-dimensional geometries and linear
mail: danilo.vulovic@elfak.ni.ac.rs.

202
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

interpolation is used for curve representation. Basic geometry • Centralized geospatial database for storing data
types are points, lines and polygons [5]. about electric power supply network. Elements of electric
Another model commonly used in GIS is the topological power supply networks are stored as a collection of
model [1]. Topological features are simple features structured geoobjects. Each geoobject is presented by its spatial
using certain topological rules. Introduction of topological component (coordinates or geometry) and thematic
relations between features simplifies data validation, modeling component (e.g. parameters for the conductors and
behavior of connected objects and optimization of different protective devices such as voltage level, conductor type,
spatial analysis. The network model is just a special case of length, construction and device type).
the topological model. The network data structures are used • GinisED Editor is a specialized tool for geographic
for modeling flows. Tree structure is used to model the radial editing of distribution network. It is a desktop application
network, while the graph structure is used to model cyclic developed in accordance with carefully studied needs and
networks. As new elements, topological model introduces requirements of customers. It is used for creation and
nodes and edges. A node is a distinguished point that connects editing of geographic schemes of the network, editing
one or several arcs. An edge is a line composed by a start and parameters of network elements and their connectivity.
an end node. The advantage of this model is the topological • GinisED Web is a WebGIS application that allows
information it is containing: every object includes information quick and easy positioning on a specific geographic area,
about the elements it is related to. search and selection of parts of electric power supply
Database management systems are vital part of every networks. This application implements information in-
modern operational GIS [1]. DBMS solutions provide GIS tegration functionalities and uses data from centralized
with standardized approaches for storing and, more geospatial database.
importantly, accessing and manipulating geographic data • WMS [9], WFS [9] and custom Web Services are
using some standard query language. GIS provides the components that provide GinisED Web and GinisED
necessary tools to load, edit, query, analyze, and display Mobile Server with raster maps and information
geographic data. Innovative work in the GIS field has considering geo-objects. WMS and WFS are components
extended standard relational DBMS solutions with specialized built according to Open Geospatial Consortium
support for storing and managing geographic data. specifications. Custom Web Services are used for data
There have been several attempts to define a standard for integration and enable GinisED Web with searching and
representing and processing geographic data in relational reporting capabilities.
databases. The GIS community, working under supervision of • GinisED Mobile Editor is tool for GPS survey of
ISO and OGC boards, has defined the core geographic types electrical consumers and relevant electrical assets.
and functions to be used in a relational DBMS and accessed GinisED Mobile Server supports mobile GIS applications
using the SQL language [6]. and provides map segments and synchronisation between
mobile database and centralized geospatial database.
III. GINISED – GIS FOR ELECTRIC UTILITY
GinisED is the geographic information system specially
designed for electric power utility companies. It uses the most
modern GIS technologies and methodologies for collecting,
editing, visualization and analysis of spatial electric power
supply network data. GinisED was developed by CG&GIS
Lab, Faculty of Electronic Engineering, University of Niš, and
is deployed in ED Jugoistok Niš (Serbian public electric
power supply company). GinisED helps ED Jugoistok Niš in
everyday operation and maintenance as it provides the
accurate, reliable spatial and non-spatial information to the
utility operational staff, and in turn help them better meet
customer needs [7][8].
GinisED (Figure 1) tools can be divided into three groups
[4][5]:
• Tools for collecting (digitization, map scanning and Fig. 1. GinisED architecture
vectorization using GPS and other specialized equipment)
and editing the spatial electric power supply network data.
• Tools for visualization of spatial electric power IV. MODEL FOR REPRESENTING POWER
supply network data for a certain geographic area. DISTRIBUTION NETWORK
• Tools for spatial analysis of electric power supply
network, potential or actual events in electric power In order to provide support for efficient spatial analysis we
supply network and risk factors for a certain geographic have developed a specific data model for representing electric
area. power supply network in GinisED. Power distribution
Basic components of the GinisED system are (Fig. 1): network usually contains features with line geometry (power

203
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

cables of different types) or point geometry (substations, implements power network elements and contains all thematic
poles, street lights, etc.). The GinisED system provides 2D attributes. Geometry of power distribution network is defined
visualization of all the elements of the power distribution through nodes (point geometry) and edges (line geometry).
network (Fig. 1). The major trace of power cables is mostly Additionally, every node element contains connection matrix
the same, i.e. under the streets, which results in overlapping that defines electrical connections between elements of power
lines. To avoid this overlap, in visualization we offset the distribution network (Fig. 4).
multiple power cables to increase the readability of the map
(schematic view).

Fig. 4. Electrical connections in node

Fig. 2. 2D visualization of power distribution network (schematic
view) GinisED system uses centralized geospatial database for
storing data about electric power supply network. This
In order to represent specifics of power distribution database is implanted using Oracle DBMS (Fig. 5). Each
network we have designed GinisED network model (Fig. 3). network feature on a map is stored as one logical record in
GinisED data model is a combination of topological network EOBJECT table. A topological part of our model is stored in
model (nodes and edges for representing geographic relational database using Oracle Spatial geometry model.
information) and typical graph data structure (for representing Oracle Spatial geometry model has several simple geometry
electric network attributes). The core part of GinisED network types and collections of them. One of the benefits of using
model is based on our GinisFrame object-oriented model for Oracle spatial data types is that a quite extensive number of
representing geographic features. spatial queries can be performed at database level. In most
scenarios spatial relationships of geometries such as “nearest
to”, or “within a specified region” are needed. Using this
mechanism, queries which can be formulated as “is a cable at
a particular distance from a cadastral parcel” or ‘find the
nearest buildings and their distance to a specific cable’” or
“find all underground networks in a given region” can be
performed.

Fig. 3. The GinisED network model

Layer, FeatureLayer and Feature classes are derived from
the original GinisFrame object model. These three classes
define the basic organization of geographic information in
GinisED system. Layer class is a basic organizational
structure of our network model. Different layers can be
organized in a hierarchical structure, thus providing a Fig. 5. GinisED geospatial database
hierarchy and aggregation of geographic features. The
FeatureLayer class defines layers of simple features (Feature
class). This model is extended with NetworkObject class for V. GINISED TOOLS FOR SPATIAL ANALYSIS
network features and NetworkObjectLayer class for
appropriate layers. Network objects, as opposed to simple Tools for spatial analysis are one of the most valuable parts
features do not contain explicitly defined geometry. The of GinisED system. Creating making maps alone and
geometry of the network features is indirectly contained in generating standard statistics reports does not justify the high
topological graph (Fig. 3). cost of building a GIS. The greatest strength of GinisED
EGraph class implements topological graph that defines system is possibility to utilize various types of information in
geometry of power network elements. EObject class

204
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

the spatial context and to generate new information and VI. CONCLUSION
conclusions on the basis of this analysis.
A very important part of GinisED are tools for spatial Development of GinisED system for ED Jugoistok Niš has
analysis of electric power supply network, potential or actual started in 2004. Before GIS was implemented, data
events in electric power supply network and risk factors for a maintenance was erratic and time-consuming. Keeping track
certain geographic area. Spatial analysis in GinisED involves of changes was difficult because hard-copy maps or digital
two types of operations: attribute query (also known as non- CAD drawings were shared among the utility’s different
spatial query) and spatial query. Attribute query requires the departments.
processing of attribute data without referencing spatial For the past six years implemented GIS solution has
information. Spatial query involves selecting features based improved efficiency in overall company operations. GinisED
on location or spatial relationships, which requires processing quickly became an essential part of the ED Jugoistok Niš day-
of spatial information. to-day business. Today, GinisED is considered very important
data resource for the distribution analyses as it contains most
facility information including major network topological
structures. The circuit map display provided by the GinisED
system is the most natural graphic user interface for engineers.
In some applications, such as the trouble call analysis,
geographic maps provide more information to engineers than
the traditional one-line diagrams.

REFERENCES
[1] P. A. Longley, M. F. Goodchild, D. J. Maquire, D. W. Rhind,
Geographic Information Systems and Scince, ISBN 978-
0470721445, Wiley, 3rd ed., 2010.
[2] S. Zlatanova, F. Döner, P. van Oosterom, “Management and
Fig. 6. Spatial query: Find all network elements in specified region visualization of utility networks for local authorities: a 3D
GinisED provides simple point-and-click query capabilities approach” Electronic Government and Electronic Participation,
Joint Proceedings of Ongoing Research and Projects of IFIP
and sophisticated queries, analysis and display functions using
EGOV and ePart, Schriftenreihe Informatik 37, Trauner Verlag,
visual, user-friendly techniques (Fig. 6). GinisED provides pp. 459-474, 2011
timely information and decision making support to GIS users. [3] D. Pickering, J. M. Park, D. H. Bannister, 1993, “Utility
It helps users to solve problems, analyze geographic Mapping and Record Keeping for Infrastructure”, Urban
situations, extract necessary data and generate new Management and Infrastructure - Urban Management
information. Programme, Washington, D.C., Vol. 10, pp. ix-11.
While basic spatial analysis involves some attribute queries [4] J. I. Igbokwe, E. J. Emengini, “GIS in Management of
and spatial queries, complicated analysis typically require a Electricity Distribution Network: A case study of Onitsha-North
series of GIS operations including multiple attribute and L.G.A., Anambra state, Nigeria”,
http://www.gisdevelopment.net/application/utility/power/utility
spatial queries, alteration of original data, and generation of
p0022pf.htm , accessed 12.01.2011
new data sets. In order to support effective spatial analysis [5] OpenGIS Reference Model (Version 0.1.3), document 03-040,
GinisED provide powerful query builder that can combine Open Geospatial Consortium, Wayland, Mass.,
different types of attribute queries and spatial queries. http://portal.opengeospatial.org/files/?artifact_id=3836, March
2003.
[6] OpenGIS Simple Feature Specification for SQL (Revision 1.1),
document 99-049, Open Geospatial Consortium, Wayland,
Mass., http:// portal.opengeospatial.org/files/?artifact_id=829,
March 2003.
[7] A. Stanimirović, D. Stojanović, L. Stoimenov, S. Đorđević–
Kajan, M. Kostić, A. Krstić, “Geographic Information System
for Support of Control and Management of Electric Power
Supply Network”, Proceedings of IX Triennial International
Conference on Systems, Automatic Control and Measurements
SAUM, ISBN-86-85195-49-7, Niš, 2007.
[8] L. Stoimenov, A. Stanimirović, N. Davidović, M. Bogdanović,
A. Krstić, D. Nikolić, “GinisED Enterprise GIS - Framework
for the Utility of the Future”, CIRED 2011, Frankfurt, Germany,
6-9. June, ISSN:2032-9644, www.cired2011.org
[9] Open Geospatial Consortium, WMS and WFS Specifications,
2002, http://www.opengeospatial.org/

Fig. 7. GinisED Query builder

205
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Computer Methods and New Values for

Cut Set Catalan Numbers
Iuliana Dochkova-Todorova1
Abstract – Improved methods to determine the values of the endvertices u and v of s on that part of the boundary of P not
cut set Catalan numbers are presented. These methods require containing the third vertex w of T. Note that the label sum of
computing the number of special solutions of a system of linear the three sides belonging to T equals n-1. If a, b and c are the
inequalities. Moreover, the table of values for cut set Catalan labels of the sides of a triangle T in ∆, where a ≤ b ≤ c, then T
numbers is completed up to n=29.
is said to be of type (a,b).
Keywords – cut set Catalan numbers, dissection of a convex Let Tn be the set of the possible types of triangles:
polygon, triangulation.
Tn = {(a, b) ∈ N 0 × N 0 ; 0 ≤ a ≤ b ≤ (n − 1 − a) / 2}. (1)

We use λa,b for the number of triangles of type (a,b) in ∆. It

I. INTRODUCTION is easy to see that the upper bound λa ,b ≤ (n + 2) /(a + b + 2)
holds for all (a , b) ∈ Tn .
Cut set Catalan numbers were introduced in [1] as a
A triangle of type (a,b) is called a boundary triangle if a=0,
variation of the well-known classical Catalan numbers Cn. The
and an inner triangle if a≥1. The set of types for the inner
Catalan number counts the number of dissections of a regular
triangles is
(n+2)-gon with labelled vertices by n-1 non-intersecting
diagonals into n triangles. The cut set Catalan numbers Sn are I n = {( a, b) ∈ N 0 × N 0 | 1 ≤ a ≤ b ≤ ( n − 1 − a ) / 2} . (2)
defined as follows: Let Pn+2 be a regular convex (n+2)-gon
with labelled vertices. Consider dissections of Pn+2 by n-1 A central triangle in ∆ is a triangle of type (a,b) where
non-intersecting diagonals into n triangles. Two such triangle a+b ≥ n/2–1. A central triangle can be only of certain types
sets T´={T´1, …, T´n} and T˝={T˝1, …, T˝n} are said to be and the number of the central triangles in one dissection can
isomorphic if there exists a bijection φ: T´→ T˝ such that T´i be only one or two.
and T˝i are congruent for i=1, …, n. Then the cut set Catalan Since every diagonal d in ∆ is a side of two triangles, the
numbers Sn is defined to be the number of isomorphism sum of the two labels of d equals n. Thus, the number of
classes of all such dissections of Pn+2. occurrences of the label i in ∆ is equal to the number of
In contrast to the classical Catalan numbers Cn, for Sn no occurrences of the label n-i in ∆ for i=1, …, n-1. These
explicit formula is known up to now. Some methods to equalities and the requirement of the central triangle imply the
determine Sn are presented in [1] and [2]. The new method in following system of linear inequalities.
[4] is partly based on this results. We will now systemize the For (a, b) ∈ I n , i = 0, K (n − 3) / 2 let
methods and analyze the ways to computing Sn. Moreover, the
values of Sn are presented in [4] for n≤29.
The cut set Catalan numbers are registered in the On-Line
 0, if a ≤ i and b > i or if a + b ≤ i − 1,
 1, if a ≥ i + 1,
Encyclopedia of Integer Sequences (OEIS, see [5]) with the  . (3)
sequence number A033961. ζ a ,b (i) = 
 − 1, if a ≤ i , b ≤ i , and i ≤ a + b < n − 1 − i ,
In [1] is proved that Sn equals the number of special
solutions of a system of linear equalities. Then the method is − 2, if a ≤ i, b ≤ i, and a + b ≥ n − 1 − i.
improved in [2] and in [4]. Now we will present here these
computer methods. Then the cut set Catalan number Sn equals the number of
solutions in nonnegative integers of the following system of
( n + 1) / 2 linear inequalities with the ((n − 1) 2 + 3) / 12
II. SPECIAL LINEAR SYSTEM AND SN
variables λa,b ≤ (n + 2) /(a + b + 2) , ( a, b) ∈ I n :
A dissection ∆ of a convex (n+2)-gon P means here a
dissection by n-1 non-intersecting diagonals into n triangles. 0 ≤ 2+ ∑ζ
( a ,b )∈I n
a ,b ( i ) λ a ,b for i = 0,K, (n − 3) / 2 ,
The label lT(s) of a side s belonging to a triangle T in ∆ is (4)
defined as the number of vertices of P between the 0 ≤ 2− ∑
( a ,b )∈I n
λ a ,b − 2 ∑λ
( a ,b )∈I n
a ,b if n is even,
a +b =n / 2 −1 a + b≥ n / 2
1
Iuliana Dochkova-Todorova is with the Faculty of Mathematics
and Informatics at St. Cyril and St. Methodius University of Veliko
0 ≤ 1− ∑λ
( a ,b )∈I n
a ,b if n is odd .
Tarnovo, 83 G. Kozarev str., Veliko Tarnovo 5000, Bulgaria, E-mail: a +b ≥( n −1) / 2
doshkova@uni-vt.bg.

206
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

i ≥ ( n − 1) / 3 follow a ≤ ( n − 1) / 3 , a<i and ζ a ,b (i) ≤ 0 . We

used a computer algebra system to determine all solutions of
the subsystem in nonnegative integers. This step could be
done in a few seconds.
Step 2. Using a simple algorithm, for each solution of the
subsystem we counted the number of possibilities to choose
the remaining variables λa,b such that the inequalities for
i = 0, K , ( n − 1) / 3 − 1 and 0 ≤ λa ,b ≤ (n + 2) /(a + b + 2) are
satisfied. For n=28 and n=29 the calculation were parallelized
and distributed among more computers.
Fig. 1. A dissection of 13-gon
The program determines not only Sn but also all the relevant
Thus, the values of Sn have been determined for n≤29 dissections of a polygon into triangles. The solutions of the
without the need for examination of dissections of a polygon. linear system Eq. (4) give the numbers of the inner triangles
The values for Sn known up to now are given in Table I. and for every dissection the numbers of boundary triangles are
determined from a system of equalities, see [2]. Haw can be
TABLE I composed the dissection corresponding to this set of triangles?
THE VALUES OF SN We start with the central triangle or two central triangles
(possible for n even). For every diagonal we draw another
n 2 3 4 5 6 7 8 9 10 11 12 13 triangle (every diagonal is side in two triangles and has two
Sn 1 1 2 2 4 6 11 17 35 57 115 203 labels). This procedure ends successfully because the relations
between the labels in the system of inequalities.
N 14 15 16 17 18 19 20 This is one of the solutions of a system from Eq. (5): λ0,0=5,
Sn 412 745 1546 2838 5901 11154 23255 λ1,2=λ0,1=2, λ1,4=λ0,4=1,λ1,1=λ1,3=λ2,2=λ2,3=λ2,4=λ3,3=λ0,2=λ0,3=0.
The corresponding dissection is given in Fig. 1.
N 21 22 23 24 25 26
Sn 44263 93169 179214 377441 733151 1547068
IV. CONCLUSION
n 27 28 29
Sn 3020878 6399874 12543862 At the time we work on the systems of inequalities for S30
and S31. Based on a special classification of the dissections
In the system of linear inequalities two inequalities, for i=0 and the recursion formula from [2] we hope to get new values.
and i=1, can be deleted because they follow from the other The cut set Catalan numbers are related to the other
inequalities. For example the system of linear inequalities for research areas: the maximum and minimum number of
n=11 gives the value S11=57: incongruent triangles in a dissection of regular convex
polygon, see [3], the number of dissections by given number
0 ≤ 2 − λ1,1 − λ1, 2 − λ2, 2 + λ3,3 of triangle types, the numbers Sn(i) of cut set Catalan numbers
0 ≤ 2 − λ1, 2 − λ1,3 − λ2, 2 − λ2,3 − λ3,3 with inner triangles etc.
. (5)
0 ≤ 2 − λ1,3 − λ1, 4 − λ2, 2 − λ2,3 − 2λ2, 4 − 2λ3,3
ACKNOWLEDGEMENT
0 ≤ 1 − λ1, 4 − λ2,3 − λ2, 4 − λ3,3
This paper was supported by the project “Modern
III. THE METHODS FOR COMPUTING SN tendencies in the development of software technologies and
algorithms, data processing and adequate training of
The determination of values of Sn is doing with the aid of specialists in these areas”.
computer programs in C. The first program is already
presented in [1] and it gives Sn for n≤19. It counts the number REFERENCES
of solutions of a linear system of equalities. In [2] the method
is improved and we succeeded in evaluating Sn for 20≤n≤25. [1] J. Dochkowa, H. Harborth, I. Mengersen, “Cut set Catalan
This program counts the number of solutions of a linear numbers”, Congr. Numer. 130, pp. 133-139, 1998.
system of inequalities from Eq. (4) in nonnegative integers. A [2] J. Dochkova, I. Mengersen, "Some new results on cut set
equivalent set of the system is constructed and it has a smaller Catalan numbers", Utilitas Math. 64, pp.159-166, 2003.
number of inequalities. At this time the evaluation for S25 took [3] J. Dochkova, I. Mengersen, "Triangle dissections of convex
about 100 hours. By the new values Sn for 26≤n≤29 the polygons", Utilitas Math. 68, pp.255-269, 2005.
[4] J. Dochkova, M. Krone, I. Mengersen, "New values for cut set
algorithm is once again improved [4]. The count the number
Catalan numbers", Utilitas Math. (submittet).
of solutions we proceeded here in two steps. [5] The On-Line Encyclopedia of Integer Sequences, http://oeis.org
Step 1. We first considered the subsystem of a system in
Eq. (4) consisting of the inequalities concerning
i = (n − 1) / 3,K , ( n − 3) / 2 and the last inequality. From

207
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Accelerating Strategies in Evolutionary Algorithms

Vassil G. Guliashki1, Leoneed Kirilov2
Abstract – We present three strategies designed to accelerate of the search space. One such sub-procedure is the local
the convergence during the search process of evolutionary search. It is supposed that each intensification period finishes
algorithms for convex integer optimization problems. The with a found local optimum. To escape from the local
strategies realize a systematic diversification of the search. They optimality a diversification of the search process is necessary
are compared with performance of scatter search and particle
after the intensification period.
swarm optimization.
To be efficient an evolutionary algorithm for search a
Keywords – evolutionary algorithms, convex integer global optimal solution it should quickly perform the
optimization problems, accelerating strategies. diversification of the search. Different ways for diversification
of the search have been developed. For example, during the
diversification phase the individuals could be modified
I. INTRODUCTION independently – like the mutation in GA. But the results are
unexpected in the sense that the modification does not lead
We consider the convex integer programming problem in necessarily to an improvement. A famous example for
the form: diversification is the Tabu list strategy used in the Tabu search
Min F(x) (1) algorithms. Some characteristics of solutions or movements
subject to: gi(x) ≤ 0; i = 1,…,m; (2) (steps in given directions) are stored as forbidden (tabu) for
lj ≤ xj ≤ uj; j = 1,…,n; (3) certain number of iterations. In this manner the cycling and
x ∈ Zn, (4) the trap of local optimality are avoided. Successful
where x is an n-dimensional vector of integer variables xj, j = diversification of the search process is the use of non-convex
1,…,n. By lj and uj are denoted the bounds (lower and upper) combinations of parent solution vectors. In this way
of xj, and F(x) is the multimodal objective function. F(x) may individuals that lie in new regions are systematically
not possess derivatives in an explicit analytical form. The generated (see [7, 9]).
functions gi(x), i = 1,…,m; are convex nonlinear functions and To achieve good convergence speed the successful global
m is the number of nonlinear constraints (2). search methods combine usually two or more metaheuristics
The convex integer problems (see [6, 18]) belong to the in hybrid methods. For example GA are combined with Tabu
class of NP-hard optimization problems. There does not exist Search methods, or with a faster local search procedure, AS –
an exact algorithm, which can solve these problems in time, with local search techniques (see [21]), GA – with clustering
depending polynomially on the problem input data length or procedure (see[4]), SS – with TS or SS – with GA (see [9]).
on the problem size. For this reason many efficient Tabu search can also be coupled with directional search
approximate evolutionary algorithms and metaheuristic approach. Another important way to accelerate the
methods have been created to find out the global optimum of performance of an evolutionary algorithm is to use the
such complex optimization problems (see [8,11,14,17,19,21]). features of the best individuals obtained during the search
To solve problem (1-4) many algorithms which mimic the process and the historically good information they have
natural evolution process of species have been designed in accumulated. This is an elitism – based approach for
order to obtain a global optimum. They could be classified as generating new offspring individuals (see for example [13]).
“evolutionary” or “population based” algorithms (see [14]). A possible strategy is to combine the qualities of a directional
The most familiar and powerful among them are Genetic type method with the good features of evolutionary
Algorithms (GA) (see [11, 15]), Scatter Search (SS) (see [7, algorithms. The directional type steps may accelerate the
9]), Tabu Search (TS) (see [8, 9, 10]), Ant Systems (AS) (see convergence in regular regions of the search space, while the
[1, 2, 3]) and Particle Swarm Optimization (PSO) (see [5, 16, evolutionary algorithms are able to escape the trap of local
17]). The evolutionary algorithms use a population of feasible optima, exploring the whole feasible domain. Accelerating the
solutions (or characteristics of solutions), called individuals, systematic diversification is an open area for further
trial (dispersed) points, ants, particles etc. In this paper is development of search strategies.
used the term individuals. In this paper three strategies for fast systematic
The evolutionary algorithms usually use an improvement diversification of the search process are proposed. The
sub-procedure to intensify the search process in some regions proposed accelerating strategies are described in Section II.
An illustrative example is given in Section III. Some
1
Vassil G. Guliashki is with the Institute of Information and conclusions are drawn in Section IV.
Communication Technologies – BAS, “Acad. G. Bonchev” Str. Bl.
2, 1113 Sofia, Bulgaria, E-mail: vggul@yahoo.com
2
Leoneed Kirilov is with the Institute of Information and II. THE HEURISTIC ACCELERATING STRATEGIES
Communication Technologies – BAS, “Acad. G. Bonchev” Str. Bl.
2, 1113 Sofia, Bulgaria, E-mail: leomk@abv.bg Considering the search process for global optimum there is

208
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

no reason the search to be directed to the region of the best Here k is chosen to cover about 10% to 50% of the
found so far (local optimal or near optimal) solution, because individuals in the population.
in the most cases it will not coincide with the global optimal c) Let xw1 ,…, xwk be the individuals in the current
solution. The same is valid for all known local optimal population having the worst (i.e. the greatest) F-values.
solutions, as well for all explored regions of the feasible Calculate the steps:
domain. The exploration of the whole feasible domain means yi = Pc – xwi , i = 1,…,k; (10)
that there is a guaranteed systematic diversification of the d) Reflect the k worst individuals towards Pc to
search process. A hybrid method performing systematically generate k new individuals (solutions):
diversified search (SDS-method) by means of separating the xnewi = Pc + yi, i = 1,…,k. (11)
feasible domain in sub-regions (cones having a common Round off each xnewi is rounded off to its near integer
vertex) is proposed in [12]. The systematic diversification of point. In case someone new solution is infeasible, i.e. the
the search consists in exploring the cones obtained one by constraints (2)-(3) are violated, restrict the step length:
one. y = θ. y, (12)
Let the feasible domain be denoted by X and let the where θ ∈(0,1).
Tchebicheff center (the point located at the maximal e) In case someone of the so generated individuals is
Euclidean distance from the constraint surfaces) be xtch ∈ X. better than one of the current population, the better individual
We assume that xtch is obtained by means of a method for replaces the worse. If there aren’t generated better individuals
solving convex problems with continuous variables. Then xtch continue by Step 5, otherwise go to a).
is rounded off to the nearest integer point itch. Step 5. Diversification phase
a) Make step β(v(j) – cs) along each ray starting at cs
A. Wave-spreading strategy and passing through the simplex vertices v(j), j = 0,…,n; in
outside direction, so that the new central solutions cs(j), j =
Step 1. Generate a regular simplex with n+1 vertices, 0,…,n; are generated.
using itch as one vertex. The other simplex vertices are b) Around each point cs(j) are generated p uniform
generated in the following manner: distributed solutions’ vectors like in Step 3 and build (n+1)
 itch j + ϕ1 if j≠i new populations Pj.
v(i)j =  i = 1,…,n; j = 1,…,n; (5) c) Perform the Intensification phase for each new
population Pj , j = 0,…,n;.
 itch j + ϕ2 if j=i
d) Make step β(cs – v(j)) along each ray starting at cs in
the opposite of sub-step a) direction, so that the new points
( n + 1) + n − 1 cs(j), j = 0,…,n; are generated. Perform the sub-steps b) and c).
ϕ1 = α.[ ] (6)
n 2 Step 6. Alternate the Diversification and the
(n + 1) − 1 Intensification phase in the same way until reaching the
ϕ2 = α.[ ] (7) boundaries of the feasible region.
n 2
Step 7. Perform simple local search around each found
Let itch be denoted as v(0). Round off each v(j), j = 1,…,n; to locally optimal solution to precisely locate all found optima.
its nearest integer point. There are (n+1) combinations of n REMARKS:
vertices, correspondingly for each facet of the simplex. The initial simplex gets larger and larger in the search
Step 2. Calculate the components of the simplex weight space like a wave raised by a stone in a lake.
center as follows: The parameter β depends on the size of feasible region.
n
For relative small domains the greatest component of β(v(j) –
∑v
( j)
i cs) is 10% of the greatest among the values Qj = uj – lj, for j =
j =0
csi = , i=1,…,n (8) 1,…,n; For larger domains β should be chosen smaller.
n +1 For large feasible domains also the rays passing through cs
Round off each component csi to its nearest integer value. and through each of the (n+1) weight centers of simplex
Step 3. Create an initial population P0 around the weight vertices determining each simplex facet should be explored in
center cs, containing p uniform distributed solution vectors, the way described above.
generated by using deviation of ±δ, where δ is a constant % of
corresponding component (for example δmax = ±1%).
Step 4. Intensification phase B. Slicing strategy
Here is used a reflection like the idea in the simplex
method by Nelder and Mead (see [20]). In this strategy the feasible domain will be separated
a) Order the individuals (solutions) in the current (sliced) in t sub-regions as follows:
Step 1. Compare the values Qj = uj – lj, for j = 1,…,n; and
population in increasing order of their F-values.
find out the maximal value Qj(max) for fixed j = jmax. Let q be
b) Calculate the weight center Pc of first k individuals:
the integer part of Qj(max)/t:
1 k ( j)
Pc = ∑x
k j =1
(9) q=
Q max
j
 (13)
t

209
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

and let lj1 = lj, uj1 = lj1+q–1, lji = uji-1+1; uji = uji-1+q–1; for The corresponding objective function values are:
i=2,…,t–1, and ljt = ujt-1+1; ujt = uj, where j = jmax. F(x(1)) = 3764084, F(x(2)) = 3742064, F(x(3)) = 3720244,
Step 2. Divide the constraint system (3) into t constraint F(x(4)) = 3816053, F(x(5)) = 3793853, F(x(6)) = 3771853,
sub-systems: F(x(7)) = 3750053, F(x(8)) = 3828244, F(x(9)) = 3806224,
lj ≤ xj ≤ uj; j = 1,…,n; j ≠ jmax; F(x(10)) = 3784404;.
lji ≤ xji ≤ uji; j = jmax; i = 1,…,t. (14) For k = 10% we choose the best individual: x(3). The worst
Each sub-region is defined by the constraint systems (2), (4) individuals are x(8), x(4) and x(9). The worst individuals are
and by one constraint sub-system from (14). reflected towards x(3). Three new better individuals are
Step 3. Perform diversification of the search process by generated and they replace the worst individuals x(8), x(4) and
going from one sub-region to another, generating the initial x(9). Proceeding in this way until no better individuals are
population at random with uniform distribution around the generated, and then performing a simple local search, the
Tchebicheff center of the current sub-region. Then perform procedure finds out the locally optimal solution x(3*) = (6100,
the Intensification phase described in Wave – spreading 3400) with objective function value F(x(3*)) = 12.
strategy (Step 4.) in each sub-region. The generated simplex has the following rounded off
vertices: v(0) = (5000, 5000), v(1) = (5003, 5010) and v(2) =
(5010, 5003). We will consider the performance of this
C. Hybrid strategy
strategy along one exploring ray, say (v(0 – cs) = (–4, –4).
Proceeding with β = 250 at v(0) the search procedure creates
This strategy consists in slicing the feasible domain in the
consecutively 5 initial populations around the calculated
way described in Slicing strategy. After that the search
central solutions cs(1) = (4000, 4000), cs(2) = (3000, 3000), cs(3)
procedure performs a Wave – spreading strategy in each sub-
= (2000, 2000), cs(4) = (1000, 1000), cs(5) = (0, 0). The last
region.
population reaches the boundaries of the feasible domain, so
that this direction is explored. The third generated population
III. ILLUSTRATIVE EXAMPLE around cs(3) = (2000, 2000) comes in the sub-area A1, so that
the intensification phase finds out the optimum F(x(1*)). The
same is repeated with the fourth and fifth generated
Let us consider the following two-dimensional example.
population.
Five sub-areas in the feasible domain are defined:
Then the procedure explores the opposite direction, creating
A1 = {0 ≤ x1, 0 ≤ x2, 21x1 + 20x2 – 84000 ≤ 0}
again five initial populations around the calculated central
A2 = {0 ≤ x1, x2 ≤ 10000, 0 < 21x1 + 20x2 – 84000, solutions cs(6) = (6000, 6000), cs(7) = (7000, 7000), cs(8) =
7x1 – 5x2 + 15000 ≤ 0} (8000, 8000), cs(9) = (9000, 9000), cs(10) = (10000, 10000).
A3 = {x1 ≤ 7200, 0 ≤ x2, x2 ≤ 10000, The last three populations come in the sub-area A5, so that the
0 < 21x1 + 20x2 – 84000, 0 < 7x1 – 5x2 + 15000} intensification phase finds out the optimum F(x(5*)).
A4 = {7200 < x1, x1 ≤ 10000, x2 ≤ 5900, 0 ≤ x2} Going on along the other two exploring rays in both
A5 = {7200 < x1, x1 ≤ 10000, 5900 < x2, x2 ≤ 10000} possible directions the search procedure finds out also the
The optimization problem is: optima F(x(4*)) and F(x(2*)).
Min F(x) = 10 + (x1 – 2500)2 + (x2 – 1000)2 if (x1, x2)∈ A1; During the exploration of whole feasible domain 31
 7 + (x1 – 1500)2 + (x2 – 7000)2 if (x1, x2)∈ A2; intensification phases are performed.
12 + (x1 – 6100)2 + (x2 – 3400)2 if (x1, x2)∈ A3; Slicing strategy
11 + (x1 – 9800)2 + (x2 – 2100)2 if (x1, x2)∈ A4; We choose the component x1 as slicing component. The
 3 + (x1 – 8100)2 + (x2 – 9700)2 if (x1, x2)∈ A5; created sub-areas are: B1: 0 ≤ x1 < 1000; 0 ≤ x2 < 10000;
subject to: 0 ≤ x1 ≤ 10000; B2: 1000 ≤ x1 < 2000; 0 ≤ x2 < 10000; …
0 ≤ x2 ≤ 10000; B10: 9000 ≤ x1 ≤ 10000; 0 ≤ x2 < 10000;
This problem has five local optima – one per each sub-area: In each sub-area is generated an initial population randomly
x(1*)= (2500, 1000), x(2*)= (1500, 7000), x(3*)= (6100, 3400), with uniform distribution. Performing the above described
x(4*)= (9800, 2100), x(5*)= (8100, 9700); intensification phase in each sub-area the search procedure
The corresponding objective function values are: finds out all locally optimal solutions. To explore the whole
F(x(1*))=10; F(x(2*)) = 7; F(x(3*)) = 12; F(x(4*)) = 11; feasible domain 10 intensification phases are performed.
F(x(5*)) = 3. Hence the global optimal solution is x(5*). Some of them are more time-consuming than the
Wave-spreading strategy intensification phases performed by Wave-spreading strategy.
Starting at the Tchebicheff center itch = (5000, 5000) the Hybrid strategy
simplex with vertices (5000, 5000), (5002.588, 5009.659) and This strategy also finds out all possible local optima. Here
(5009.659, 5002.588) is generated. The weight center of the the value of parameter β remains the same like in the Wave-
simplex is cs = (5004, 5004). spreading strategy, because the component x2 keeps its
The population P(0) includes 10 points (individuals): variation interval unchanged. This leads to great steps along
x(1) = (4994, 4994), x(2) = (5004, 4994), x(3) = (5014, 4994), the exploring rays and in some directions already the first
x(4) = (4985, 5004), x(5) = (4005, 5004), x(6) = (5005, 5004), generated initial population is infeasible. For this reason β
x(7) = (5015, 5004), x(8) = (4994, 5014), x(9) = (5004, 5014), should be reduced in half and this is repeated until the
x(10) = (5014, 5014);. generating a feasible population becomes possible.

210
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Comparisons with other algorithms [3] Colorni A., M. Dorigo and V. Maniezzo, „An Investigation of
The described strategies are compared with scatter search Some Properties of an Ant Algorithm“, in Parallel Problem
and with particle swarm optimization. Starting with the initial Solving from Nature 2, R, Männer and B. Mandrieck eds.,
population P(0) these two algorithms are able to find out only North-Holland: Amsterdam, 1992, pp. 509-520.
[4] Damavandi N. and Safavi-Naeini S., “A hybrid evolutionary
one locally optimal solution, and this is x(3*). Better
programming method for circuit optimization”, In Proc. IEEE
performance is achieved in case the initial population is Transactions on circuits and systems – I: Regular papers, vol.
enough dispersed. The scatter search has better chances to 52, No. 5, May 2005, pp. 902-910.
find out the global optimal solution in case the limit of [5] Eberhart R. C. and J. Kennedy, “A new optimizer using particle
iterations is large. In this case it needs more than 1000 swarm theory”, Proceedings of the Sixth International
generations. Symposium on Micromachine and Human Science, Nagoya,
The presented problem may be solved by genetic algorithm, Japan, pp. 39-43, 1995.
using niches. In this way all optima can be found but this [6] Garey M. R. and Johnson D. S., “Computers and Intractability:
performance will be essentially more time consuming than the A Guide to the Theory of NP-Completeness”, W. H. Freeman,
San Francisco 1979.
proposed solution procedures. [7] Glover F., “Heuristics for integer programming using surrogate
constraints”, Decision Sciences 8, 1977, pp. 156-166.
IV. CONCLUSIONS [8] Glover F., “Future path for integer programming and links to
artificial intelligence”, Computers & Operations Research 13,
1986, pp. 533-549.
The presented strategies for fast systematic diversification [9] Glover F., “Tabu Search for Nonlinear and Parametric
of the search have the following advantages: Optimization (with Links to Genetic Algorithms)”, Discrete
• They systematically diversify the search process, Applied Mathematics, “Viewpoints on Optimization”, 1991.
avoiding in this manner the trap of local minima. [10] Glover F., “Tabu search and adaptive memory programming –
• They explore roughly the whole feasible domain and advances, applications and challenges”, in Advances in
have good chances to find out the global optimal Metaheuristics, Optimization and Stochastic Modeling
Technologies, Barr, Helgason and Kennington eds.-, Kluwer:
solution.
Boston, MA., 1997, pp. 1-75.
• The applying of local search technique at the end of the [11] Goldberg D. E. Genetic Algorithms in Search, Optimization and
search process guarantees the good quality of the Machine Learning, Addison Wesley, Reading, Mass, 1989.
obtained solution. [12] Guliashki, V., (2007) “A Hybrid Population Based Method
• The proposed strategies have a better convergence to the Solving Convex Integer Optimization Problems”, In:
global optimum in comparison to other global search Proceedings of ICEST2007, (Editor Prof. Dr. Mitrovski, C.),
algorithms, in which the search process does not perform June 24 – 27, 2007, Ohrid, Macedonia, vol. I, pp. 249-252.
a systematic diversification. [13] Guliashki V., C. Korsemov, H. Toshev, (2010) “Elitism Based
Evolutionary Algorithm for Discrete Optimization Problems”,
• They are simpler and don’t require large computer In: Proceedings of ICEST2010, (Editor Prof. Dr. Mitrovski, C.),
memory and complex memory organization in June 23 – 26, 2010, Ohrid, Macedonia, vol. I, ISBN: 978-9989-
comparison to other global search strategies like Tabu 786-57-0, pp. 281-284.
search. [14] Hertz A., Kobler D., „A Framework for Description of
• They use populations with relatively small size. This Population Based Methods“, Tutorials and Research Reviews,
makes them efficient in solving large dimensional 16-th European Conference on Operational Research Brussels,
problems. Belgium, 1998, pp. 48-59.
[15] Holland J. H., „Adaptation in Natural and Artificial Systems”,
• The proposed strategies are easy for computer
The University of Michigan Press: Ann Arbor, MI, 1975, MIT
programming implementation. Press, 1992.
[16] Kennedy J. and R. C. Eberhart, “Particle Swarm Optimization”,
ACKNOWLEDGEMENT Proceedings of IEEE International Conference on Neural
Networks, Piscataway, N. J., 1995, pp. 1942-1948.
[17] Krusienski D. J. and Jenkins W. K. “Design and performance of
The authors gratefully acknowledge the support of adaptive systems, based on structured stochastic optimization
Bulgarian National Science Fund, Grant No DTK02/71 “Web- strategies”, IEEE Circuits and Systems, Vol.5, No.1, first
Based Interactive System, Supporting the Building Models quarter 2005, pp.8-20.
and Solving Optimization and Decision Making Problems”. [18] Nemhauser G. L. and Wolsey L. A. “Integer and Combinatorial
Optimization”, Wiley, New York. 1988.
[19] Pirlot M., “Heuristic Search Methods”, Oper. Res. Designing
REFERENCES Practical Solutions; Tutorial and Research Review Papers, Euro
XIII/OR 36, The joint EURO/Oper. Res. Society Conference;
[1] Colorni A., M. Dorigo and V. Maniezzo, “Positive Feedback as Univ. of Strathclyde, Glasgow, 19-22, VII. 1994. pp. 180-201.
a Search Strategy”, Department of Electronics, Politechnico di [20] Reklaitis G.V., Ravindran A., Ragsdell K. M., Engineering Op-
Milano, Italy, Working paper 91-16, 1991. timization. Methods and Applications, John Wiley & Sons, 1983.
[2] Colorni A., M. Dorigo and V. Maniezzo, “ Distributed [21] Taillard E. D., Gambardella L. M., Gendreau M. and J.-Y.
Optimization by Ant Colonies”, in Proceedings of the First Potvin, “Adaptive Memory Programming: A Unified View of
European Conference on Artificial Life (ECAL-91), F. J. Varela Metaheuristics”, Tutorials and Research Reviews, 16-th
and P. Bourgine eds., The MIT Press: Cambridge, MA, 1991, European Conference on Operational Research Brussels,
pp. 134-142. Belgium, 1998, pp. 30-38.

211
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

2D Weather product visualization using Marching

Squares algorithm
Igor Antolović1, Dejan Rančić2, Vladan Mihajlović3, Dragan Mihić4, Marija Đorđević5
Abstract – In this paper we present a efficient way for 2d visualization algorithms offering climatology experts
weather product visualization by using an approach which powerful tools for weather/climate tracking and analysis.
is based on the Marching Squares algorithm. We examine This paper focuses explicitly on visualization of scalar data
the possibilities of extracting both isolines and color filled using the Marching Squares algorithm considering it can be
isoareas as well as improvements in order to avoid the directly applied on 2D scalar weather product visualization.
default broken polyline Marching Squares result and force Although the most common version of this algorithm is the
more smooth visualization. By using different sampling classic linear segment approach, we will further cover
resolutions and parameters we test and evaluate this possible besier curve advancements in order to generate
approach on real weather temperature products. smooth isolines as well as the extended Marching Squares
version suited for extraction of color filled isoareas.
Keywords – Isoline, Marching squares, besier curves Considering that the isoline extraction and visualization
concept represents a scalar field visualization problem in the
I. INTRODUCTION second part of this paper we give a short review of available
techniques for scalar field visualization. The third part gives a
detailed description of the Marching Squares algorithm
Modern climatology depends heavily on weather tracking
followed by the fourth part which presents practical
GIS (Geographic Information System) [1] applications and
implementation results in combination with mean temperature
their ability to process, analyze and visualize weather data.
This kind of systems range from simple applications for test data. Finally the fifth part consists of conclusions and
processing of climate data to more complex weather notes on future work.
forecasting systems [2]. The nature of weather data can be
vector (e.g. wind) or scalar (temperature, pressure, II. 2D SCALAR FIELD VISUALIZATION CONCEPTS
precipitation etc.) thus emphasizing the need for various
visualization algorithms aimed towards both 2D and 3D data A discrete 2D scalar field is essentially a (mxn) grid where
as well as vector and scalar data types. HAISIS (Integrated every grid cell takes a value from a real set of numbers:
Hail Suppresion Information System) [3,4,5] is one such
system. It is developed at the Faculty of Electronic S : (x 0 ,.., x m ) × (y 0 ,.., y n ) → R (1)
Engineering in Niš, at the Computer Graphic and Geographic
Information Laboratory (CG&GIS Lab) to satisfy
There are basically two types of 2D scalar data
requirements of Republic Hidrometeorological Service of
visualization techniques: color mapping and isoline extraction
Serbia (RHMSS). It is operating more that 10 years and it
as shown on Fig.1.
supports LIC (Line Integral Convolution) [6] methods for
vector data visualization, 2D isoline tracing, 3D isosurface
extraction, 3D wind visualization and various other

1
Igor Antolović is with the Faculty of Electronic Engineering,
Aleksandra Medvedeva 14, 18000 Nis, Serbia,
E-mail: igor.antolovic@elfak.ni.ac.rs.
2
Dejan Rančić is with the Faculty of Electronic Engineering,
Aleksandra Medvedeva 14, 18000 Nis, Serbia,
E-mail: dejan.rancic@elfak.ni.ac.rs.
3
Vladan Mihajlović is with the Faculty of Electronic Engineering, Fig.1. Color mapping (left) and isoline extraction (right)
Aleksandra Medvedeva 14, 18000 Nis, Serbia,
E-mail: vladan.mihajlovic@elfak.ni.ac.rs. Color mapping approach can be most easily implemented
4
Dragan Mihić is with the Republic Hydrometeorological Service and uses a quite straightforward approach. The basic idea
of Serbia and South East European Virtual Climate Change Center,
E-mail: dragan.mihic@hidmet.gov.rs
behind color mapping is to traverse through all scalar field
5
Marija Đorđević is with the Republic Hydrometeorological values and map those values into colors. Additionally color
Service of Serbia and South East European Virtual Climate Change values can be interpolated in order to get a smoother output.
Center, E-mail: mapuja@gmail.com Usually instead of RGB (Red Green Blue) color model, HSB
(Hue Saturation Brightness) is used considering that it defines
more practical color scheme. Also an interesting

212
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

recommendations on using colors with climate data can be • Use the calculated index in order to extract
found in [7]. information of which edges are intersected by the
On the other hand when isoline extraction is considered it is isoline from the precalculated edge table.
necessary to start from its very definition. For a given isovalue • By using linear interpolation determine the exact
(c) an isoline consists of a set of points defined by: points of intersection of the isoline and the cell
edges.
I = {(x, y) | S(x, y) = c} (2) • Connect the intersection points in order to get the
current isoline segment.
Meaning that isolines consists of all points that share a
constant isovalue (c). On the other hand when it comes to
practical isoline extraction there are two basic methods:
• Contour tracing
• Isoline segment extraction using Divide and Concur
approach

Contour tracing is the most intuitive isoline extraction

method and it is based on the idea of tracing values on a scalar
field grid from a single seed point until a boundary is reached
or a contour loop is detected. Recent work on this can be
found in [8]. By using this method a continuous poly-line is
produced which can be later additionally interpolated and
smoothed. The common problem in this approach can be
found in the process of seed point determination. Also in the
case of multiple contour tracking, multiple track markers must
be used. As previously stated, one practical application of this Fig.2. Marching Squares 16 possible cases
method can be found also in the HASIS 3DI system where
this approach is used to extract dBz contours for given Upon processing all cells and drawing all isoline segments
isolevels. a full isoline set will be created as shown on Fig.3.
The Divide and Concur approach relies on the fact that the
problem of isoline extraction can be simplified by extracting
isoline segments in single scalar field cells and afterwards join
those segments into continuous isolines [9]. This method is
applicable for 3D as well 2D scalar fields. The 3D case of this
method is well known by the name of Marching Cubes [10]
and it is used for fast extraction of isosurfaces. In this paper
we will further explore and describe it’s simplified 2D version
also known as the Marching Squares algorithm.

Fig.3. Example of isoline extraction sheme for isovalue (7)

III. MARCHING SQUARES
This approach is very fast but it has two main
The main idea behind the Marching Squares algorithm is to
disadvantages:
simplify the isoline extraction and visualization procedure by
• The quality of the generated isolines depends on the
processing scalar field cells one at a time.
resolution of the grid. Since the processing of a
It can be easily concluded that the values in the cell corners
single cell produces line segments it is obvious that
can be above or below the given iso-value. Considering this
on small resolution grid isolines will be broken
fact a set of 16 possible cases can be isolated as shown on Fig.
polylines.
2. Furthermore a simple scheme can be involved in order to
index these cases where 0 means the corner value is below the • The second disadvantage is that there is no
required isovalue and 1 means the corner value is above the information about a continuous isoline.
required isovalue.
This coding scheme can be uses to precalculate index tables It is obvious that for fixed edge intersection points there can
in order to drastically speedup the process of identifying the be lot of solutions for the isoline segments but it is necessary
current cell case. to use the approach which provides the smoothest output.
The steps of processing the current cell can be described as On order to increase the isoline quality a common method
follows: is to take advantage of the gradient property of the scalar
field. Basically the gradient is a vector which is pointed
• Compare the current cell corners against the required
towards the greatest differential change in the given point of a
isovalue and determine the cell index.
scalar field.

213
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Let the 2D scalar field be defined as: By processing all cells in this way and generating besier
isoline segments, a drastic increase in quality can be noticed
S = f ( x, y ) (3) as the isolines show smooth flow through the cells.
Finally a simple polygonal scheme can be derived from the
The gradient can be calculated as follows: original Marching Squares set as shown on Fig.5. This new
set can be used to generate color filled isoareas. By combining
∂S ∂S multiple layers of isoareas a multicolor isoarea can be
∇S = i + j (4) generated as shown on Fig.6. The only problem with this
∂x ∂y approach is that it is order dependent. It is obvious that the
In the practical case where a discrete scalar field is used the isoareas must be sorted in increasing isovalues going from
gradient components can be calculated by using central bottom to top.
differences as follows:
S ( x + 1, y ) − S ( x − 1, y )
g ( x, y ) x = (5)
2∆x
S ( x, y + 1) − S ( x, y − 1)
g ( x, y ) y = (6)
2∆y

By using these equations we can easily calculate gradient

vectors of the cell corners and furthermore calculate gradient
vectors on isoline and grid intersection points by using linear
interpolation. The most interesting property of the gradient
vector is that it is perpendicular to the isoline passing through
a given point, meaning that the vector normal to the gradient
vector acts as a tangent vector of the isoline. This information
can be used to construct smooth isoline segments instead of
just using straight lines. In order to accomplish this, besier
curves can be used considering the flexible relation between Fig.5. Polygonal Marching Squares scheme
besier control points and besier curve normals.

The process of smooth isoline segment (Fig. 4) construction

can be described in several steps as follows:
• Calculate gradient vectors in cell corner points,
• Calculate gradient vectors on isoline intersection
points by interpolating between corner gradient
vectors,
• Calculate normal vectors of previously calculated
isoline gradient vectors,
• Find normal vector intersection M,
• Calculate besier control points as half distances
between isoline intersection points and normal Fig.6. Multilayered polygonal isolevel approach
intersection point M.
IV. RESULTS
For testing purposes as an input data a average temperature
dataset was used which was sampled on a 64x64 grid. In order
to visually compare the quality of the generated isolines both
classic linear and advanced besier approach was used. The
results are shown on Fig. 7. What can easily be noticed is that
for higher sampling resolutions as well small cell sizes the
difference between the advanced smoothing and classic linear
approach becomes indistinguishable. On the other hand for
higher zoom levels the broken polyline appearance of the
classic linear method becomes more obvious as shown on Fig.
7.c). Beside the classic isoline extraction, the multilayered
isoarea approach was also implemented so depending on the
need either of this visualization techniques can be used. For
example in lots of cases a combination of isolines and terrain
Fig.4. Besier curve construction examples map is can be used as shown on Fig.7.b)

214
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

a) b) c)
Fig.7. a) Smooth isoareas with normals b) Non-smooth isolines on map c) Comparison of smooth and non-smooth details

adaptation and mitigation" (43007) financed by the Ministry

of Education and Science of the Republic of Serbia within the
V. CONCLUSION framework of integrated and interdisciplinary research for the
period 2011-2014.
Efficient visualization of climate data products is a basic
requirement in all weather tracking and analysis GIS
applications. Taking into account the nature of this data, two REFERENCES
groups of climate variables can be distinguished. The first
group consists of scalar climate variables (temperature, [1] D. Rančić, A. Dimitrijević, V. Mihajlović, "GIS and Virtual
pressure, precipitation etc.) and the second group consists of Reality Systems Integration", ICEST 2004, Bitola, Macedonia,
vector climate variables (e.g. wind). pp. 313-316, 2004.
In this paper we considered visualization techniques for the [2] L. A. Treinish, “Visual data fusion for applications of high-
resolution numerical weather prediction”, IEEE Proceedings of
first group of variables. This kind of visualization has already the conference on Visualization '00, 2000, pp. 477 - 480.
been identified as a scalar field visualization problem which is [3] V. Mihajlović, S. Đorđević-Kajan, D. Rančić, B. Predić, I.
most commonly solved by using isoline extraction approach. Antolović, P. Eferica, Z. Babić, “Architecture of HASIS-3D
Among two basic isoline extraction methods (by contour System Designed for Hail Suppression Purposes”, Proceedings
tracing and by using the Marching Squares divide and concur of ICEST 07, Ohrid, jun 2007.
method ) we choose the second method considering that it [4] Đorđević-Kajan, S., Milovanović, B., Rančić, D., Kostić, A.,
relies on precalculated tables thus providing great speed as Stanić, Z., Stoimenov, L., “Hail Suppression Information
well as multiple isoline extraction. System of Radar Center”, GIS/LIS'96, Budapest Hungarz, June
As a proof of concept we implemented the Marching 1996, pp.102-111.
[5] Rančić, D., Smiljanić, M., Đorđević-Kajan, S., Kostić, A.,
Squares Algorithm and additionally considered isoline Eferica, P., Vuković, P., Vučinić, Z., “Radar Data Processing
segment smoothing by using besier curves. Finally we for Cloud Seeding in Hail Suppression Information System”,
compared the results of classic linear Marching Squares RADME 98
approach with besier curve based results. As it is shown the [6] M. Kovačević, V. Mihajlović, I. Antolović, D. Rančić, Z. Babić:
classic linear approach suffers from a broken polyline look LIC based Visuelization of Air Flow in Clouds, Conference on
compared to the smooth besier based approach. Furthermore Computer Science and Information Technologies – YUINFO
we compared isoareas and isoline visual results as well 2010, Kopaonik, Serbia, 3 – 6 March 2010. (In Serbian).
variants of isoline on map backgrounds. Upon evaluating this [7] American Meteorological Society, Guidelines for using Color
results we can conclude that the Marching Squares algorithm to Depict Meteorological Information: IIPS Subcommittee for
Color Guidelines. 1993
give a rather elegant, fast, and visually satisfying solution to [8] Fu Chang , Chun-Jen Chen , Chi-Jen Lu, A linear-time
the isoline extraction problem. What is also important to component-labeling algorithm using contour tracing technique,
emphasize is that the final quality of the generated isolines Computer Vision and Image Understanding, v.93 n.2, p.206-
depends directly no the sampling grid resolution thus the 220, February 2004
future work on this matter will include considerations of [9] G. Cottafava, G. Le Moli, "Automatic Contour Plotting",
techniques for isoline extraction on lower sampling grid Comm. ACM 12, 7 (July 1969), pp. 386-391.
resolutions. [10] Lorensen W, Cline H. Marching cubes: a high resolution 3D
surface construction algorithm. Computer Graphics 1987;
21(4):163–9
ACKNOWLEDGEMENT

This paper was realized as a part of the project "Studying

climate change and its influence on the environment: impacts,

215
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Efficient Implementation of
BDD Packages on the GPU Platform
Miloš M. Radmanović1 and Dušan B. Gajić2

Abstract – Efficient construction and manipulation of Binary Parallel computing can be used to efficiently solve large-
Decision Diagrams (BDDs) is an important component of scale problems, either by distributing computational loads
Computer Aided Design (CAD) tasks. One solution to improve the among processors or by utilizing the large memory in parallel
performance of BDD packages is to perform some of the BDD networked workstations. Parallel processing of BDDs can be
operations in parallel using the Graphics Processing Unit (GPU).
The recent evolution of GPU frameworks for programming general
used both to reduce the BDD algorithm running time and to
purpose computations, such as the OpenCL and Nvidia CUDA, has extend the memory limitations which exist in the traditional
offered GPUs as a powerful and attractive choice for developing single-processor sequential computing.
high-performance numerical applications. In this work, we propose In order to increase the performance of BDD packages, the
an efficient implementation of a BDD package which distributes concept of parallelism has been introduced to the BDD
computations over central processing units (CPUs) and GPUs. The representations and algorithms in several papers. In [9], a
proposed implementation exploits various sources of parallelism parallel algorithm for the construction of BDDs is described.
that can be found in BDD packages. The experimental results The algorithm is motivated by the fact that the construction of
confirm that the application of the proposed solution leads to a BDD, for certain large or particularly complex Boolean
significant computational speedups in BDD packages.
functions, can be a very time-consuming task. In order to
Keywords – Binary decision diagrams, BDD package, parallel overcome limitations of computational resources, research in
implementation, graphics processing unit, GPU computing. [10] presents an approach which distributes the BDD data
structure across multiple networked workstations. Further,
several techniques are introduced which allow parallelization
I. INTRODUCTION of depth-first search algorithms on a BDD. Reference [11]
presents a parallel algorithm for BDD construction targeted at
Binary Decision Diagrams (BDDs) are the dominant data shared memory multiprocessors and distributed shared
structure for representing Boolean functions in CAD memory systems. The results obtained using a shared memory
applications. The application of BDDs is further extended multiprocessors system show speedups of over 2×, with four
with their use in various areas of computer science and processors, and up to 4×, with eight processors. Alongside the
engineering. In practice, the success of BDD representations research on parallel BDD construction, various BDD-
depends on the abillity to efficiently manipulate large BDDs. algorithm parallel implementations were developed for
Therefore, considerable research has been conducted in order networks of workstations [12-14]. In [15], some key
to develop more efficient implementations of BDD algorithms algorithms for performing BDD operations are first described
[1-5]. and, afterwards, an approach to their parallelization is
BDD algorithms are usually built on top of BDD packages. described, with a goal to achieve efficient execution of BDD
Many BDD package implementations have been developed in packages on multicore CPUs. The technique of general
a variety of programming languages and most of them are purpose computing on the GPU (GPGPU) enables parallel
freely available as public domain on the Internet. The choice processing of non-graphics algorithms using graphics
of a BDD package for a certain application is typically guided hardware. Only recently, the possibility of using GPUs to
by the following package characteristics: functionality, solve complex problems in logic design has been explored by
software interface, robustness, reliability, portability, support, researchers, for example in [16-20].
and performance. In most cases, the performance of a BDD Motivated by the exisiting research on efficient execution
package is of major concern. Parameters which influence the of parallel BDD operations on multicore CPUs and possibility
performance of a BDD package include the choice of the of using GPUs, in this paper we propose an efficient
programming language and the software and hardware implementation of a BDD package using the GPU platform.
platforms, BDD node structure, type of garbage collection, The proposed implementation exploits the various sources of
unique and operation hash table strategies [6-8]. parallelism that exist in BDD packages. We address several
topics considering parallel computations in BDD packages
1
Miloš M. Radmanović is with the Faculty of Electronic and present their mapping to the GPU architecture. The
Engineering, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E-mail: experimental results confirm that the application of the
milos.radmanovic@gmail.com proposed implementation of a parallelized BDD package leads
2
Dušan B. Gajić is with the Faculty of Electronic Engineering, to significant computational speedups over traditional C/C++
Aleksandra Medvedeva 14, 18000 Niš, Serbia, E-mail: implementations processed on CPUs.
dusan.b.gajic@gmail.com

216
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The paper is organized as follows: Section 2 shortly algorithms is typically based on the BDD node data structure,
introduces the BDD representation of Boolean functions. unique and operation tables, and depth-first BDD traversal.
Section 3 describes the structure of a BDD package and the The decisions made in defining the BDD node data
basic BDD algorithms. Section 4 presents the GPU as a structure have impact on memory space requirements for
computing platform. Section 5 discusses the operations in the storing node objects. There are many choices for defining a
BDD package for which we introduce the GPU processing. BDD node object, but every node usually contains: an id, then
Section 6 shows experimental results obtained with the cofactor, else cofactor, next pointer, and reference counter
proposed implementation. Finally, Section 7 offers some [22]. The BDD construction is based on applying the traversal
concluding remarks and directions for future work. in a depth-first manner.
The maintance of a BDD representation is improved by
storing BDD nodes in a dictionary, called the unique table.
II. BINARY DECISION DIAGRAMS The unique table maps a unique triple of (v, g, h) for a BDD
node, where v is the variable identifier, g is the node
BDDs consist of non-terminal (decision) nodes, 0-edges connected to the "1" edge, and h is the node connected to the
and 1-edges attached to all non-terminal nodes, a ‘0’ terminal "0" edge. The unique table is a hash table with the hash
node, and a ‘1’ terminal node, as shown in Fig.1. The non- collisions resolved by chaining. A hash function is applied to
terminal node with no upper nodes is called a root node. As it the triple to obtain the index in the unique table of the start of
can be seen from Fig.1, a variable is related to every non- a collision chain of nodes. Comparing the unique triple
terminal node, such that every path from the root node to one against the nodes in the collision chain addresses the look up.
of the terminal nodes respects the same variable ordering. The efficient implementation of almost all recursive BDD
A Boolean function can be converted into an equivalent manipulation algorithms is made possible by the operation
function by performing Shannon expansion based on the fixed table. This table is also implemented as a hash table with the
variable ordering. This new function can be represented by a collisions resolved by chaining. The collision lists can be kept
binary tree. The corresponding BDD is constructed from this sorted to reduce the number of memory accesses required on
binary tree by applying the two reduction rules (redundant average for the lookup. Table sizes which are prime numbers
node elimination and equivalent sub-graphs sharing). The require an expensive modulo operation. Table sizes that are a
Boolean operations such as the logical AND, logical OR, etc., power of 2 are often better handled by memory management.
can be achieved by using BDD manipulations, which have an As the variable ordering can have significant impact on the
average time complexity propositional to the size of BDDs. It size of a BDD, dynamic variable reordering component is a
is well known that the size of the BDD for a given Boolean fundamental part of all modern BDD packages. Dynamic
function depends on the variable order for the function. The variable reordering algorithms are generally based on the
strength of BDDs is that they can represent Boolean function shifting algorithm [23]. The BDD variable order changes by
data with high level of redundancy in a compressed form. exchanging nodes in one level with nodes in the neighbouring
levels. Dynamic variable ordering should best be invoked as
an asynchronous process that can be activated at any time
during the BDD manipulation. Dynamic variable ordering is a
complex problem since finding an optimal ordering is NP-
hard. Futher, small changes in the BDD ordering may have
significant impact on both the space and time requirements.
BDD computations are memory intensive, especially when
large BDDs are involved. They not only require a lot of
memory, but also frequent accesses to many small data
structures. Furthermore, many intermediate BDD results are
created to arrive at a resulting BDD. These computations may
Fig. 1. BDD representation of the function defined by the truth table have poor memory handling, as there is not a solution to
F = [00101111]T. ensure that the accessed BDD nodes are close in memory. It is
important to have a garbage collector component [24] to
III. BDD PACKAGES automatically remove BDD nodes that are no longer useful. In
modern BDD packages, garbage collector component is based
BDD packages are deployed in many software tools, on reference counting and the recycling of nodes for later
particularly in the area of logic design, and they typically deal reuse. Garbage collection is activated when the percentage of
with the following common implementation features [1]. A the unusable BDD nodes reaches a threshold. Unusable BDD
BDD package has three main components [21]: nodes are nodes with zero reference counts. Some of unusable
 The BDD algorithm component, BDD nodes may become usable again (recycled) if they are
obtained as results of new subproblems. Thus, in the case
 Dynamic variable reordering component,
when BDD nodes change state between „usable“ and
 Garbage collection component.
„unusable“ frequently, garbage collection can reduce the
The BDD algorithm component builds the result BDDs for benefit of the operation tables and decrease the overall
various Boolean operations. The implementation of these performance of a BDD package.

217
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

IV. THE GPU ARCHITECTURE AND GPGPU operation is a crucial component of the Apply procedure
which is central to the BDD construction and manipulation
Processor frequency progress, which followed the Moore’s [1]. The lookup_insert operation returns the key, if it already
law for more than four decades, reached a limit in 2003, exists in the hash table, or, otherwise, inserts the key.
mostly due to the inability to further solve the problems of Reference [15] shows how this operation can be safely
heat dissipation and energy consumption. Since then, there are executed in parallel on multicore processors. The
two approaches in the development of computer architectures. lookup_insert operation within the BDD algorithm component
The multicore approach, typical for CPUs, seeks to maintain in our BDD package is, therefore, implemented as an OpenCL
the execution speed of sequential programs while moving into kernel which performs the same function over different keys.
multiple cores. In contrast, the manycore approach, found in Since GPUs use hardware multithreading [25, 26, 27], this
GPUs, focuses more on the execution throughput of parallel automatically allows simultaneous execution of as many
applications. This resulted in a rapid evolution of GPU lookup_insert operations as there are active GPU threads.
architectures. The GPU evolution started from fixed-function The effectiveness of garbage collection component can
hardware specialized for rendering computer graphics, which have significant impact on both space and time requirements
first appeared in 1999, and developed into a massively of a BDD package. When garbage collector removes unusable
parallel, scalable, and fully programmable platform BDD nodes, the unique and the operation table entries that
characterized by exquisite memory bandwidth and reference these nodes must also be removed to eliminate
computational power. Due to this, many of the general- unusable references. If garbage collection is not invoked
purpose applications which were processed on CPUs are now frequently enough, the memory usage can be greatly
re-implemented in order to efficiently harness the GPU increased. An OpenCL kernel for garbage collection is
resources. For more details on recent changes that made developed so that each GPU thread removes an entry from the
GPGPU possible, see [25, 26, 27, 28]. hash tables. Since thousands of GPU threads can be active at
The GPU parallel processing model is based on a large the same time, this leads to a massively-parallel GPU garbage
number of processor cores which can directly address into a collection. The transfer of the garbage collection task to the
global GPU memory. The GPU architecture follows the single GPU, also allows the CPU to be free to perform other tasks
program, multiple data (SPMD) paradigm [26, 27], features a for which it may be more suitable.
multi-level memory hierarchy and has simple branching
circuits. In SPMD computing, a large number of threads VI. EXPERIMENTAL RESULTS
execute in parallel the same function, called a kernel, over
different data. In this section, we compare the performance of our GPU
Application Programming Interfaces (APIs) most often accelerated BDD package implementation, which incorporates
used for the development of GPGPU programs are Nvidia’s the before-mentioned OpenCL kernels, and a single-threaded
CUDA and Open Computing Language - OpenCL. CUDA is a C/C++ implementation of the BDD package on the CPU. For
vendor-specific development framework and only supports the comparison, we use a set of well-known standard
execution on Nvidia’s GPU hardware. Therefore, we give benchmarks. Table I presents a view on the performance of
advantage to OpenCL which is hardware agnostic. Further, the BDD package computations performed in the basic BDD
the OpenCL C programming language, included in the construction algorithm on the CPU and the GPU.
framework, allows development of programs that are both
TABLE I
accelerated and portable across a wide set of devices (CPUs,
COMPARISON OF THE BDD CONSTRUCTION TIMES FOR THE BDD
GPUs, Field Programmable Gate Arrays (FPGAs), Digital
PACKAGE ON THE CPU AND THE GPU
Signal Processors (DSPs), Cell processors, embedded
processors) [28]. computation time [s]
Benchmark in / out / cubes
CPU GPU
V. GPU ACCELERATION IN THE BDD PACKAGE alu4 14 / 8 / 1028 0.15 0.08
apex1 45 / 45 / 206 5.18 0.81
Motivated by the existing work on the parallelization of apex2 39 / 3 / 1035 3.31 0.62
components in BDD packages, described in Section 1, we apex5 117 / 88 / 1227 0.30 0.17
explored various sources of parallelism that exists within the cordic 23 / 2 /1206 0.06 0.04
algorithms included in BDD packages in order to develop an cps 24 / 109 / 654 0.15 0.09
efficient model of parallel BDD operations on GPUs. The misex2 25 / 18 /29 0.05 0.03
components of the BDD package that take advantage of the misex3 14 / 14 / 1848 0.03 0.02
GPU processing in our present solution are the BDD table3 14 / 14 / 135 0.02 0.02
algorithm and the garbage collection components. table5 17 / 15 / 158 0.01 0.01
The effectiveness of caching within unique and operation
tables of the BDD algorithm component strongly influences The test platform features an Intel i7-920 quad-core
the number of subproblems generated in the BDD algorithm processor, operating at 2.66 GHz, and has 4 GBs of DDR3-
task execution. Thus, the hash tables (unique and operation 2000 RAM. GPU that is used is an Nvidia GeForce GTX
tables) in a BDD package need to support concurrent 560Ti with 1GB of GDDR5 RAM, composed of 384
execution of the hash operation lookup_insert(key). This

218
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

streaming processors. The OpenCL kernels are developed [9] S. Kimura, E.M. Clarke, "A parallel algorithm for constructing
using the AMD Accelerated Parallel Programming SDK 2.6. binary decision diagrams", in IEEE Intl. Conf. on Computer
The size of the unique and the operation tables is limited to Design: VLSI in Computers and Processors, 1990, 220-223.
[10] T. Stornetta, F. Brewer, “Implementation of an efficient parallel
8191 entries. The garbage collection is activated if the hash
BDD package”, in Proc. of Design Automation Conf., Las
tables exceed the 80%-full marker. All benchmarks are used Vegas, USA, 1996, 641-644.
in the Espresso-mv or pla format [29] and the computation [11] B. Yang, D. R. O'Hallaron, “Parallel breadth-first BDD
times are reported in seconds. construction”, 9th ACM SIGPLAN Symposium on Principles and
As it can be seen from Table I, the addition of the GPU Practice of Parallel Programming, 1997, 145-156.
acceleration to the BDD package brings clear performance [12] D. Caban, D. Milford, "A parallel BDD engine for logic
benefits over the CPU-only solution. The speedup, in terms of verification", in Proc. 5th Annual IEEE Int. ASIC Conf. and
the BDD construction algorithm computation time, for most Exhibit, 1992, 499-502.
of the cases is substantial and varies from 6.4× to 1.5×. [13] R.K. Ranjan, J.V. Sanghavi, R.K. Brayton, A. Sangiovanni-
Vincentelli, "Binary decision diagrams on network of
However, it should be noted that the speedup may not be
workstations," IEEE Int. Conf. on Computer Design: VLSI in
achieved in some cases, e.g., in the case of the benchmark Computers and Processors, 1996, 358-364.
table3, because the construction of the BDD in this case is not [14] K. Milvang-Jensen, J. Alan, "BDDNOW: A parallel BDD
enough computationally-intensive to benefit from the package", in Proc. 2nd Int. Conf. on Formal Methods in
introduction of the GPU. Computer-Aided Design, 1998, 501-507.
[15] H. Yuxiong, “Multicore-enabling a binary decision diagram
algorithm”, Intel Software Network, 2009,
VII. CONCLUSION http://software.intel.com/en-us/articles/multicore-enabling-a-
binary-decision-diagram-algorithm.
This paper proposes an implementation of a BDD package [16] K. Gulati, S. Khatri. “Towards acceleration of fault simulation
which uses the GPU hardware for the acceleration of certain using graphics processing units”, in Proc. 45th ACM/IEEE
data-parallel operations. The proposed implementation Design Automation Conference, 2008, 822-827.
exploits several sources of parallelism that exist within BDD [17] V. Bertacco, D. Chatterjee, “High performance gate-level
simulation with GP-GPU computing”, Int. Symp. on VLSI
packages. In particular, we discuss the parallel OpenCL
Design, Automation and Test, 2011, 1-3.
implementation of the lookup_insert hash operation, which is [18] D. Chatterjee, V. Bertacco, "EQUIPE: parallel equivalence
of central importance to the BDD algorithm component, and a checking with GP-GPUs", IEEE Int. Conference on Computer
GPU-accelerated garbage collection component. The Design, 2010, 486-493.
experimental results confirm that the application of the [19] D. Gajić, R. Stanković, "GPU accelerated computation of fast
proposed implementation, which distributes the BDD package spectral transforms", Facta Universitatis - Series: Electronics
operations over the CPU and the GPU, leads to significant and Energetics, vol. 24, no. 3, University of Niš, Serbia, 2011,
computational speedups. The results presented in the paper 483-499.
may also be helpful in the general study on improvement of [20] D. Gajić, R. Stanković, M. Radmanović, “Implementation of
dyadic correlation and autocorrelation on graphics processors”,
BDD packages. Since these first research results look
Int. J. Reasoning-based Intelligent Systems (IJRIS), vol. 4, nos.
promising, further work on this topic will be devoted to the 1/2, Inderscience, Geneva, Switzerland, 2012, 82-90.
extension of the GPU acceleration method to the [21] B. Yang, "Optimizing model checking based on BDD
implementation of other operations that are common in the characterization", PhD Dissertation, Carnegie Mellon University
components of BDD packages. Pittsburgh, USA, 1999.
[22] B. Yang et al., “A study of BDD performance in model
checking”, in Proc. Formal Methods in CAD, 1998, 255-289.
REFERENCES [23] R. Rudell, “Dynamic variable ordering for ordered binary
decision diagrams”, in Proc. Int. Conf. on Computer-Aided
[1] K. Brace, R. Rudell, R. Bryant, "Efficient implementation of a Design, 1993, 139–144.
BDD package", in Proc. Design Automation Conf., 1990, 40-45. [24] N. Klarlund, T. Rauhe, "BDD algorithms and cache misses",
[2] J. Sangavi, R. Ranjan, R. Bryton, A. Sangiovanni-Vincentelli, BRICS Report Series RS-96-5, Department of Computer
"High performance BDD package based on exploiting memory Science, University of Aarhus, 1996.
hierarchy", in Proc. of the Design Automation Conf., 1996. [25] T. Aamodt, “Architecting graphics processors for non-graphics
[3] D. Long, “The design of cache-friendly BDD library“, in Proc. compute acceleration’, in Proc. IEEE PACRIM Conf., 2009,
1998 IEEE/ACM Intl. Conf. on CAD, 1998, 639 - 645. 963-968.
[4] G. Janssen, “Design of pointerless BDD package“, 10th Int. [26] S. Ryoo, et al., “Optimization principles and application
Workshop on Logic&Synthesis, Lake Tahoe, USA, 2001. performance evaluation of a multithreaded GPU using CUDA”,
[5] R. Ebendt, F. Gorschwin, R. Drechsler, Advanced BDD in Proc. 13th ACM SIGPLAN Symp. on Principles and Practice
Minimization, Springer, New York, 2005. of Parallel Programming, 2008, 73–82.
[6] G. Janssen, “A consumer report on BDD packages“, in Proc. [27] J. Owens, M. Houston, D. Luebke, S. Green, J. Stone, and J.
16th Symp. Integrated Circuits and Systems Design, 2003, 217- Phillips, “GPU computing”, Proc. of the IEEE, vol. 96, no. 5,
223. 2008, 279–299.
[7] F. Somenzi, “Efficient manipulation of decision diagrams,” Int. [28] B. Gaster et al., Heterogeneous Computing with OpenCL,
J. Software Tools for Technology Transfer (STTT), vol. 3, no. 2, Elsevier, 2011.
2001, 171-181. [29] R. Rudell, Espresso Misc. Reference Manual Pages,
[8] M. Sentovich, "A brief study of BDD package performance", in http://embedded.eecs.berkeley.edu/pubs/downloads/espresso/ind
Proc. of the Formal Methods on CAD, 1996, 389-403. ex.html, 1993.

219
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Architecture of Distributed Multiplatform GIS for

Meteorological Data Analysis and Visualization
Marko Kovačević1, Aleksandar Milosavljević1, Vladan Mihajlović1 and Dejan Rančić1
Abstract – This paper presents a distributed multiplatform sharing geospatial data and functions can be done by
architecture of geographic information system (GIS) capable of extending existing traditional desktop GISs with a set of “web
providing flexible and efficient geospatial and meteorological interfaces”. This kind of extension forms a GIS Server
data analysis and visualization to a broad spectrum of user component. Besides GIS Server, proposed architecture
profiles. Traditionally, geographic information systems were
realized as monolithic and platform-dependent applications and
consists of Desktop GIS, Web GIS and Mobile GIS
used by rather small group of GIS professionals. With the subsystems. Each of the identified subsystems has its different
development of computer hardware, Internet and availability of role and intended user groups in the system as a whole. Great
geospatial data in recent years, GIS evolved and adapted to new diversity of meteorological data applications (weather
environments that emerged. Grater availability has lead to forecast, hail suppression, climate change monitoring,
greater importance and use of GIS in many different domains. In aviation, agriculture, military etc.) implies variety of different
this paper we will focus on architecture specification that allows GIS user needs and profiles. This paper will focus on
distribution of geospatial data and functions to different identifying subsystem roles and possible use case scenarios
hardware and software platforms. The proposed architecture that can generally be applied for all meteorological data
consists of four subsystems: Desktop GIS, GIS Server, Web GIS
applications.
and Mobile GIS. Each of the identified subsystems has its
different role and intended user groups in the system as a whole. In order to build a distributed GIS that is open for
Great diversity of meteorological data applications (weather connecting to a variety of different geospatial data sources,
forecast, hail suppression, climate change monitoring, aviation, commonly acceptable standards for implementation of these
agriculture, military etc.) implies variety of different GIS user web interfaces are needed [5]. Currently, Open Geospatial
needs and profiles. This paper will focus on identifying Consortium, as an international industrial consortium with an
subsystem roles and possible use case scenarios that can aim of developing publicly available standards in the field of
generally be applied for all meteorological data applications. GIS, has several implementation specifications that
standardize this field. Our functional needs for web interfaces
Keywords – GIS, Multiplatform Architecture, Meteorology.
were met by three of these specifications. The first is Web
Map Service (WMS) Implementation Specification [6] that
describes a web service interface for custom maps retrieval.
I. INTRODUCTION The second is Web Feature Service (WFS) Implementation
Specification [7] which describes a web service interface for
A geographic information system (GIS) is a special type of manipulating, querying, and retrieval of geospatial entities
computer-based information system, tailored to store, process, using both spatial and non-spatial criteria. The third is Web
and manipulate geospatial data [1]. The ability of GIS to Coverage Service (WCS) Implementation Specification [8]
handle and process both location and attribute data, which describes a web interface for retrieval of geospatial data
distinguishes GIS from other information systems. It also as "coverages" – digital geospatial information representing
establishes GIS as a technology that is important for a wide space-varying phenomena. This paper focuses on a global GIS
variety of applications [2]. Traditionally, geographic architecture specification that is based on previously
information systems were built as monolithic and platform- mentioned specifications and that allows distribution of
dependent applications [3], but, with the development of geospatial data and functions to different hardware and
computer hardware and availability of geospatial data in software platforms.
recent years, GIS evolved and adapted to new environments The paper is organized as follows: in the second part, we
that emerged [4]. Greater availability has lead to greater define subsystem-level architecture with emphasis on
importance and use of GIS in many different areas. What was communication between them. The third part is dedicated to
reserved for a small group of professionals in the past now Desktop GIS and GIS Server subsystems. In the fourth part
became available worldwide, using Internet, on personal client side subsystems are discussed in more detail. The fifth
computers and mobile devices. part presents a case study concerning subsystems’ use in
Modern distributed GIS architectures rely on a client/server meteorological data applications. Finally, in the conclusion,
model where clients provide user access to geospatial data, the achieved results are summarized.
while one or more servers provide their sharing. The task of
II. SUBSYSTEM-LEVEL ARCHITECTURE
1
Marko Kovačević, Aleksandar Milosavljević, Vladan Mihajlović
and Dejan Rančić are with the Faculty of Electronic Engineering at The architecture of proposed GIS solution consists of four
University of Niš, Aleksandra Medvedeva 14, Niš 18000, Republic subsystems: Desktop GIS, GIS Server, Web GIS, and Mobile
of Serbia, E-mail: marko.kovacevic@elfak.ni.ac.rs,
alexm@elfak.ni.ac.rs, wlada@elfak.ni.ac.rs, ranca@elfak.ni.ac.rs.

220
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

GIS. Like all modern distributed GIS architectures, proposed system is GIS framework – a set of C++ class libraries
solution architecture relies on a client/server model. developed in Laboratory for Computer Graphics and GIS
Communication between server side subsystems is depicted (CG&GIS Lab), on Faculty of Electronic Engineering in Niš.
by the component UML diagram shown in Fig. 1. The GIS framework implements common data model and most of
responsibility of Desktop GIS component, in the server side of the functionalities. Desktop GIS application and GIS Server
the system, is creation and preparation of a content that can be system service can be seen as a kind of an interface toward
distributed via GIS Server. GIS Server implements OGC them. The main advantage of sharing the same core between
WMS, WFS and WCS services in order to distribute these two subsystems is in maintainability. When some new
geospatial data and functions to its clients. For preparation of feature (or change) is added to the framework (e.g. support for
geospatial content, Desktop GIS relies on use of a project file. new data format) it automatically becomes present in both
The project file is an XML document that contains Desktop GIS application and GIS Server. A simplified
specification of all coordinate systems, geodata services, overview of the implementation architecture of GIS
coverage data, layers, and styles that are used for geospatial framework along with Desktop GIS and GIS Server is shown
and coverage data organizing and presentation. While in Fig. 3. There are two packages (Visual C++ projects)
Desktop GIS allows creation and manipulation of GIS responsible for production of executable components of
projects, GIS Server only use previously prepared projects for Desktop GIS application and GIS Server system service,
geospatial content distribution. while the other packages (i.e. projects) represents previous
Component UML diagram in Fig. 2 shows client side introduced GIS framework.
subsystems of proposed GIS solution. Mobile GIS and Web
GIS rely on web interfaces provided by the GIS Server
(WMS, WFS and WCS) in order to provide their own <<EXE>> <<EXE>>
Desktop GIS application
functionalities. Desktop GIS also has the ability of utilizing GIS Server system service

web interfaces provided by the GIS Server. GIS Server is

intentionally left out in Fig. 2 because, from the client
standpoint, it can be any GIS server (or servers) that
<<DLL>> <<LIB>>
implements OGC standardized WMS, WFS and WCS. GIS XML

Desktop GIS

<<LIB>> <<LIB>>
<<artifact>>
Proj4 GDAL
Geospatial Data
Project File

GIS Server <<LIB>> <<LIB>>

GEOS ADO

Fig. 3. Simplified implementation architecture of Desktop GIS and

WMS WFS WCS GIS Server

Fig. 1. Server side subsystems The GIS framework is implemented with support of four
open source libraries: Proj4, GDAL, GEOS, and ADO. Proj4
[9] is a library that implements mechanism for definition of
WMS WFS
various coordinate reference systems (i.e. projections) and
WCS
reprojection of coordinates between them. GDAL [10] is a
library for raster geospatial data formats management, while
Desktop GIS Mobile GIS GEOS [11] library enables support for spatial querying and
Web GIS
geoprocessing. Finally, ADO [12] library contains a set of
classes that simplifies database accessing and management.
The majority of the framework implementation (e.g. data
model) is contained within GIS dynamic library classes, while
Fig. 2. Client side subsystems XML is only helper library for accessing, managing, and
creating XML documents. Since Desktop GIS and GIS Server
share the same implementation framework, both of them are
III. DESKTOP GIS AND GIS SERVER tied to the same platform. In our development we relied on
Microsoft Visual C++ 2008 and Microsoft Foundation Classes
The described relation between Desktop GIS and GIS (MFC). As a result, current Desktop GIS and GIS Server
Server implies tightly coupled implementation of these two platform is limited to Microsoft Windows and Microsoft
components. Core of Desktop GIS application and GIS Server Windows Server series of operating systems.

221
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

IV. CLIENT SIDE SUBSYSTEMS meteorological data application requires different user profiles
that differ in:
As we already stated, client side of the proposed GIS • Education level,
solution includes Web GIS, Mobile GIS and Desktop GIS. • Responsibility for decision making,
Web GIS and Mobile GIS completely depend on services and • Working environments,
geospatial data supplied by GIS Server. Desktop GIS has the • Experience with GIS tools, etc.
capability of utilizing web interfaces provided by the GIS For all numbered reasons, using single platform solution
Server, but also has other responsibilities and functionalities, (one GIS application) can be inappropriate for wider user
as described in section 3. audience. In order to analyze the use of proposed GIS
Web GIS’s primarily use is geospatial data visualization subsystems, we should identify several typical GIS user
and querying. GIS Server’s WMS service is used for retrieval profiles of meteorological data application in general:
of custom maps in form of raster images. WFS service enables • Meteorological data analyzers
query execution and results retrieval in form of GML [13] • Decision makers
encoded geographic features. WCS service enables coverage • Simple viewers
data retrieval for defined spatial and temporal constraints. The • Users involved in “on field” operational tasks, and
implementation of Web GIS heavily relies on OpenLayers • Applications administrators.
[14], open source JavaScript library developed by the Open Meteorological data analyzers are the most complex group
Source Geospatial Foundation. The implementation of Web of users that includes experts in particular meteorological (or
GIS is based on several web technologies such as: PHP, some related) domain which are also specialized for GIS tools
JavaScript, AJAX, and .NET. use. For that reason, powerful Desktop GIS applications are
Mobile GIS has similar background architecture, i.e. it also what this spectrum of users needs. Furthermore, these users
use WMS, WFS and WCS services of GIS Server, but the often need specialized tools for geospatial and meteorological
implementation and intended use is quite different. As it is data analysis. Our solution includes development and delivery
intended to run on hand-held devices, as “on field” tool for of such specialized tools within Desktop GIS. These tools are
data acquisition and situation awareness, the platform of developed as plugins – DLLs that conform to a predefined
choice was Microsoft Windows Mobile. Mobile GIS interface used for interaction with the application and
subsystem is implemented using C# programming language underlying geospatial data. When present, a plugin is
and .NET Compact Framework. Unlike Web GIS that drives automatically loaded into Desktop GIS application and can be
huge support form OpenLayers library, Mobile GIS’s accessed from appropriate menu item. A sample screen shoot
communication with WMS, WFS and WCS services is of Desktop GIS application is shown in Fig. 4.
implemented from scratch. Since Mobile GIS, besides map
visualization and coverage data querying, also deals with
geospatial data acquisition, GIS Server’s WFS service has to
be extended with transactional part of the specification.
Transactional WFS, in addition to querying, allows
manipulation of geographic features using insert, update and
delete operations.
As previously stated, Desktop GIS is able to use web
interfaces provided by GIS Server: WMS, Transactional WFS
and WCS. Classes that implement communication between
Desktop GIS and above listed web interfaces are contained
within GIS library (Fig. 3). Since Desktop GIS and GIS
Server share the same GIS framework, GIS Server can also
act as a client of some other GIS server, thus enabling the
creation of cascade connection between servers. Being the
most powerful platform of the three, Desktop GIS provides Fig. 4. Screen shot of Desktop GIS application
additional functionalities which are not implemented in Web Decision makers are users that have authority to make
GIS and Mobile GIS: various tools for WCS coverage data decisions (e.g. hail suppression activity initiation, aviation
analysis and flexible style management for visual flight plan regulation, taking appropriate actions in
representation of WCS coverage data, WMS raster images and agriculture, military etc.). Although usually experts in
WFS geographic features. particular domain, decision makers are characterized as casual
GIS users. Their primarily tasks are not GIS related, i.e. they
V. CASE STUDY (can) only use GIS as another information source for decision
making. More often, decision making must by fast, so this
Using GIS in meteorological domain involves variety of category of users must be quickly presented with the exact as
different user needs and profiles. Meteorological data possible information they need. By our opinion, simplified
applications are vastly diverse: weather forecast, hail applications with easy-to-use user interfaces, along with
suppression, climate change monitoring, aviation, agriculture, adequate geospatial and meteorological data content are what
military etc. From the GIS user perspective, each decision makers need. Client side subsystems, primarily Web

222
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

GIS and Mobile GIS applications, are what we see as the most Server is only part of the responsibilities and functionalities of
appropriate solution for this category. A sample screen shots Desktop GIS. Primary use of Web GIS is geospatial data
of Web GIS and Mobile GIS are shown in Fig. 5. visualization and querying, and it is intended for broader
The third group, simple viewers, includes broader audience spectrum of less specialized users. Mobile GIS, as it runs on
of generally non-specialized people interested in some aspect hand-held devices, is basically intended for “on field” data
of particular meteorological domain (e.g. weather forecast). acquisition and situation awareness. As the most powerful
Web GIS is the most suitable tool for this user group. client side subsystem that provides various specialized
Many meteorological applications include “on field” analysis tools along with basic GIS functionalities, Desktop
operational tasks. In terms of GIS functions we recognized GIS is primarily indented for specialized users.
two major categories: simple meteorological data collecting
and support for enhanced situation awareness. On field
operations, regarding the environment, require specialized
ACKNOWLEDGEMENT
computing hardware and the running software. Mobile GIS, as
the proposed tools of choice, basically satisfies listed Research presented in this paper is funded by Ministry of
requirements for this user group. Education and Science, Republic of Serbia as part of the
Finally, the last identified category of GIS users is involved project “Environmental Protection and Climate Change
in applications administration. The personnel appointed to Monitoring and Adaptation”, Nr. III-43007.
these tasks must be well educated and especially trained to
support maintenance of the proposed GIS system. Although REFERENCES
they directly do not use any of the presented subsystems they
must have good knowledge of all of them. [1] Worboys, M., Duckham, M., GIS: A Computing Perspective.
Second Edition. CRC Press, Boca Raton, FL, 2004.
[2] Chang, K., Introduction to Geographic Information Systems,
Third Edition. McGraw-Hill, New York, NY, 2005.
[3] Wong, S. H., Swartz, S. L., Sarkar, D., “A Middleware
Architecture for Open and Interoperable GISs”, IEEE
MultiMedia, 9(2)(2002) pp. 62-76.
[4] Shekhar, S., Vatsavai, R. R., Sahay, N., Lime, S., “WMS and
GML based Interoperable Web Mapping System”, Proceedings
of the 9th ACM international symposium on Advances in
geographic information systems, Atlanta, Georgia, USA, 2001,
pp. 106-111.
[5] Milosavljević, A., Đorđević-Kajan, S., Stoimenov, L., “An
Application Framework for Rapid Development of Web based
Fig. 5. Screen shot of Web GIS (left) and Mobile GIS (right) GIS: GinisWeb”, in: J.T. Sample, K. Shaw, S. Tu, M.
Abdelguerfim eds. Geospatial Services and Applications for the
Internet, Springer, New York, 2008, pp. 49–72.
[6] La Beaujardière, J. ed., Web Map Service Implementation
VI. CONCLUSION Specification (Ver. 1.1.1), document 01-068r3, Open Geospatial
Consortium Inc., 2002,
Great diversity of meteorological data applications implies http://www.opengeospatial.org/standards/wms.
large spectrum of potential GIS users. In this paper we [7] Vretanos, P. A., ed., Web Feature Service Implementation
focused on two main issues. Firstly, we presented distributed Specification (Ver. 1.0.0), document 02-058, Open Geospatial
multiplatform architecture of GIS applicable in Consortium Inc., 2002,
meteorological applications that consists of four subsystems: http://www.opengeospatial.org/standards/wfs.
Desktop GIS, GIS Server, Web GIS and Mobile GIS. Second [8] Whiteside, A., Evans, J. D., ed., Web Coverage Service (WCS),
Implementation Specification (Ver. 1.1.0.), document 06-083r8,
issue was identifying subsystem roles and possible use case Open Geospatial Consortium Inc., 2006,
scenarios that can generally be applied for all meteorological http://www.opengeospatial.org/standards/wcs.
data applications. [9] PROJ.4 – Cartographic Projections Library,
Proposed GIS solution architecture relies on a client/server http://proj.osgeo.org.
model. Server side subsystems are GIS Server and Desktop [10] GDAL – Geospatial Data Abstraction Library,
GIS. In this context, the main role of Desktop GIS application http://www.gdal.org.
is content preparation in terms of data acquisition and [11] GEOS – Geometry Engine, Open Source,
processing. In order to distribute prepared data, we rely on http://trac.osgeo.org/geos.
well-known concept of project file. Once the project file and [12] Antollini, C., "A set of ADO classes - version 2.20", The Code
Project, 6 Sep 2005,
corresponding geospatial data are prepared, they can be shared http://www.codeproject.com/KB/database/caaadoclass1.aspx.
using WMS, WFS and WCS services of GIS Server [13] Cox, S., Cuthbert, A., Lake, R., Martell, R. eds., OpenGIS
subsystem. Client side subsystems are Web GIS, Mobile GIS Geography Markup Language Implementation Specification
and Desktop GIS. Web GIS and Mobile GIS completely (Ver. 2.1.2), document 02-069, Open Geospatial Consortium,
depend on services and geospatial data supplied by GIS 2002, http://www.opengeospatial.org/standards/gml.
Server, while utilization of web interfaces provided by GIS [14] OpenLayers – Free Maps for the Web,
http://www.openlayers.org.

223
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Tracking Analogue to Digital Converter Modelling using

VHDL-AMS
Marieta Kovacheva1 and Ivailo Pandiev2
Abstract – This paper describes a VHDL-AMS implementation ADC12. Its resolution can be set as 8 -, 10 - or 12-bit. In the
of a behavioural model for tracking analogue to digital converter PSpice based model the voltage at in mode and ref mode with
(ADC). For creating the model, simplification and build-up respect to ground mode is sampled starting at the rising edge
techniques known from modelling analogue integrated circuits of the convert signal and ending when the status goes high.
have been adapted. This generalized model is independent from
actual technical realizations and is based upon compromises
These models with a suitable choice of parameters and
regarding the representation of exact circuit structure in the elements can be used of a mixed-signal circuit simulations,
model. The proposed tracking ADC model accurately predicts but not confirm to the architectures of a broad class of the
the circuit behaviour for dc and transient responses. Its beha- monolithic ADCs. Transformation from OrCAD PSpice A/D
viour is created consequently using tracking ADC basic struc- to System Vision libraries can be done, but it’s quite
ture. The modelling of the ADC behaviour is implemented and complicated, requires a lot of resources and additional
confirms to the format of the simulation program System Vision processing. In Matlab Simulink exists one ADC model, called
5.5 (from Mentor Graphics). The simulation results show accu- idealized ADC quantizer with the following changeable
rate agreement with the theoretical predictions. parameters: (1) number of converted bits, (2) minimum and
Keywords – Mixed-signal circuits, Analogue to digital conver-
maximum output voltage and (3) output data type. This model
ters, Counting method, Tracking ADCs, VHDL-AMS, Mixed- is generalized structure and cannot be used for several specific
signal simulation. applications. Also an investigation in the published articles
was made and the conclusions are following – found one
article describing a VHDL implementation of a behavioural
model for pipeline ADCs by using VHDL description to
I. INTRODUCTION facilitate the synthesis of the digital part [8], another one for
simulation tools in designing sigma delta ADC and its
An analogue to digital converter (ADC) is a device that efficiency by using both Simulink and VHDL-AMS [9] and
converts a continuous quantity to a discrete time digital the last one presents behavioural VHDL-AMS model for
representation. Typically, an ADC is an electronic device that monolithic Half-Flash (Two Step) 10-bit ADC [10]. The goal
converts an input analogue voltage or current to a digital of this paper is to develop a behavioural VHDL-AMS model
number proportional to the magnitude of the voltage or the that accurately simulates the basic characteristics of tracking
current. The digital output may use different coding schemes. ADCs.
Basically there are three different conversion methods: (1) TABLE 1 COMPARISON OF CONVERSION METHODS
parallel method (convert a word at a time), (2) weighting Number Number of
method (convert a digit at a time) and (3) counting method ADC types of reference Characteristics
(convert a level at a time). In Table 1, is shown a comparison clocks voltages
between three basic methods of conversion [1-4]. The benefits very fast; needs many
of the tracking converters are that (1) their output is parts ( 2 n − 1 compa-
continuously available and (2) their accuracy is very high. The Parallel rators); lower resolu-
main drawback of tracking ADC is their slow step response, (Flash) 1 2n − 1 tion, expensive; large
however, this is compensated with their possibility of large method power consumption
resolution (12-,14-,16- bits). capable of high speed;
After analyze of the existing model libraries in OrCAD medium accuracy;
PSpice A/D [5], SystemVision (from Mentor Graphics) [6] Successive- good tradeoff between
and Matlab Simulink [7] some conclusions are made. In approximation n n speed and cost; speed
method limited ~ 5Msps;
System Vision libraries an ADC behavioural model can be
require S/H circuit
found, but it works by using successive-approximation input signal is avera-
method and its resolution is fixed to 10 bits, others types of ged; greater noise
converters are not included. In OrCAD PSpice A/D libraries Counting immunity; high
exists simulation model of ADC – called ADC8, ADC10 or method max 2 n 1 accuracy; slow; require
S/H circuit

1
Marieta Kovacheva is with the Faculty of Electronics at II. PRINCIPLE OF TRACKING CONVERTERS
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000,
Bulgaria, E-mail: m_kovacheva@tu-sofia.bg. OPERATION
2
Ivailo Pandiev is with the Faculty of Electronics at Technical The counting method requires the least circuit complexity,
University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, Bulgaria.E-
but the conversion time is considerably longer than with the
mail: ipandiev@tu-sofia.bg.

224
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

other methods – generally between 1ms and 1s. This is a great III. MODELLING OF TRACKING ADC WITH VHDL-
plus for slowly changing signals. The counting method can be AMS
implemented in various ways; one of them is the tracking
conversion. The basic structure of tracking ADC is given on
The created behavioural model of tracking ADC is
Fig. 1.
K2
developed by using a style combining structural and
− 0,5U LSB
mathematical description. The structural description is the
G3
+ netlist of the model and the behavioural description consists of
Uin Σ ≥1 eoc
−
CLK simultaneous statements to describe the continuous behaviour.
+ 0,5U LSB The behaviour of the proposed ADC is strictly described using
K1
G1 & & G2 basic structure given on Fig. 1.
dout dout
U _ out
DAC
up/down C+
counter C−
A. A behavioural language: VHDL-AMS
U _ ref dout

VHDL-AMS is a comparatively new standard 1076.1 of

Fig. 1. Tracking ADC basic structure. VHDL that supports hierarchical description and simulation of
The operation of tracking ADC is the following. We analogue and mixed-signal applications with conservative and
subtract compensating voltage v_out from input voltage vin. non-conservative equations. On the mixed-signal side a
The voltage v_out is the DAC output voltage. The result of variety of abstraction levels is supported. The VHDL-AMS
subtracting the two voltages is passed to two analogue modelling is not restricted to mixed-signal applications but
comparators K1 and K 2 . It is compared with reference also supports thermal and mechatronic systems.
voltages vctrl_pos and vctrl_neg with values ±0,5U LSB
B. A tracking ADC behavioural VHDL-AMS model
( U LSB is the voltage unit for the least significant bit; that is the
voltage for resulting number dout = 1 ). If the voltage The proposed behavioural model of a tracking ADC is
differrence is bigger than +0,5U LSB , the comparator K1 developed following the design method based on a Top-Down
jumps to '1' ( K 2 output is '0' ) and enables the logic gate G1 . analysis approach and applying simplification and build-up
This gate passes the impulses from clock generator to the technique, known from modelling analogue integrated
summing input of the up/down counter. The counter increases circuits. The process of model building and testing can be
its value and DAC output voltage tracks the input voltage. If broken down into three main steps: 1) structure the model; 2)
the voltage difference is smaller than −0,5U LSB , the build the model; 3) validate the model.
The structure of the tracking ADC model is built using the
comparator K 2 jumps to '1' ( K1 output is '0' ) and enables results obtained by analyses of the simulation models for
the logic gate G 2 . This gate passes the impulses from clock ADCs in [5, 6] and by using the structures presented on Fig. 1.
generator to the subtracting input of the up/down counter. The The model includes two generic parameters with numerical
counter decreases its value. The up and down changes in the values: v _ ref = 5,12V – reference voltage and Nbits = 12 –
digital code of the DAC allow its output voltage always to
number of bits. These two parameters can be changed
tracks the variation in the input voltage. If the difference
according different applications. For the model consecutively
between vin and v_out is in range of ±0,5U LSB , no clock
are defined the work of the up/down counter, DAC,
impulses are passed to the counter and this is the end of subtracting the DAC output voltage from input circuit voltage,
conversion mode (DAC has reached the input value, and it’s analogue comparators with work ranges, logic gates and
digital code corresponds to it). The output comparator values assuming the results on output signals. Depending on the
are both '0' and the value of logic element G3 is '1' . The input result, output value of '1' or '0' is assumed.
voltage, the compensating DAC output voltage and end of Fig. 3 shows the behavioural VHDL-AMS model of
conversion signal eoc versus time are given on Fig. 2. tracking ADC. Library clause and the use clause make all
End of conversion
declarations in the packages std_logic_1164, logic_arith,
logic_unsigned, math_real and electrical_systems visible in the
model. This is necessary because the model uses logic types,
U LSB
Uin
operation with logic types and nature electrical from
packages. The proposed model is composed by an entity and
an architecture, where bold text indicates reserved words and
U _ out upper-case text indicates predefined concepts. The entity
declares the generic model parameters and specifies one
interface terminal of nature electrical, two input signal
t terminals – type std_logic, one output signal terminal – type
Fig. 2. Tracking process at AC input voltage for the tracking std_logic and one output signal terminal – type std_logic_vector,
converter on Fig. 1. with a length of 12 bits. The proposed model includes the
following terminals: ain – input voltage, clk, stop – clock and
stop signal, dout – output vector, eoc – end of conversion.

225
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

library IEEE; library ieee_proposed; Furthermore, the model has two inner terminals: sum_out and
use IEEE.std_logic_1164.all; use IEEE.std_logic_arith.all; use IEEE.math_real.all;
use ieee.std_logic_unsigned.all; use ieee_proposed.electrical_systems.all;
dac_out. They are used to specify the voltages vsum and
v_out respectively. Also six inner signals are defined: out
entity tracking_adc_12_bits is _plus, out_minus – outputs of K1 and K 2 ; clk _up, clk_down –
generic ( Nbits : integer := 12; ‐‐ number of bits of ADC's output summing and subtracting inputs of counter; s_out – DAC
v_ref : voltage := 5.12); ‐‐ ADC's reference voltage output and synch – internal synchronization of the clock signal.
The architecture is subdivided into four main parts
port ( terminal ain : electrical; ‐‐ADC's analog input terminal
signal clk, stop : in std_logic; ‐‐ Strobe clock and stop signal according to the functions of the different elements in basic
signal dout : out std_logic_vector (0 to 11);‐‐ Digital output structure: (1) up/down counter, (2) DAC, (3) subtractor and
signal eoc : out std_logic ); ‐‐ equilibrium signal (4) window comparator with additional logic element, gene-
end entity tracking_adc_12_bits; rating eoc signal.
The output value of the converter is formed by the formula:
architecture default of tracking_adc_12_bits is
VI V V
signal out_plus, out_minus: std_logic; ‐‐ output of comparators Z= = 2 n I = Z max I (1)
signal clk_up, clk_down: std_logic; ‐‐ up and down input of the counter VLSB Vref Vref
signal s_out: real := 0.0; ‐‐ output of dac
signal synch : std_logic; ‐‐ internal synchronization of clock signal where Z is the output digital value, V I is input analogue
terminal sum_out: electrical; ‐‐ sum of input signal and dac signal
terminal dac_out : electrical; ‐‐ dac output voltage, V LSB = V ref / 2 n is the least significant bit voltage
quantity v_out across i_out through dac_out to electrical_ref;
quantity vin across ain to electrical_ref; and is equal to division of reference voltage by 2 n bits, n is
quantity vsum across isum through sum_out to electrical_ref;
quantity vctrl_pos, vctrl_neg : voltage;
the number of bits, and Z max - is the maximum possible
digital value.
begin
The schematic representation of the symbol confirm to the
vctrl_pos == 0.5*(v_ref/2.0**Nbits); ‐‐ upper reference voltage description of Fig. 3 is given on Fig. 4. The generated model
vctrl_neg == ‐0.5*(v_ref/2.0**Nbits); ‐‐ lower reference voltage has a total number of five ports: ain – input terminal, clk–
process(Vin'above(vctrl_pos), Vin'above(vctrl_neg), clk_up, clk_down, synch)
clock signal, stop - stop signal, eoc – end of conversion signal,
dout1[0:11] – output logic vector with length of 12 signals.
variable counter : std_logic_vector (0 to 11):="000000000000"; ‐‐ init of the counter Number of bits in the model can be changed with define a
variable sum : natural;
new value for the parameter Nbits, the signal dout and the
begin variable counter. After the change a generation of new symbol
is needed.
if rising_edge(clk_up) and clk_down = '0' then ‐‐ define work mode of counter
counter := counter + '1';
elsif rising_edge(clk_down) and clk_up = '0' then
counter := counter ‐ '1';
end if;
dout <= counter;
sum := 0;

for i in counter'range loop ‐‐ modelling DAC

sum := sum * 2 + boolean'pos(counter(i) = '1' or counter(i) = 'H' ); Fig. 4. A schematic symbol of 12 − bit tracking ADC.
end loop;
s_out <= v_ref * real(sum) / real(2**Nbits);
‐‐ comparing the sum_out with ranges of comparators IV. MODEL PERFORMANCE
if vsum'above(vctrl_pos) then
out_plus <= '1' ;
out_minus <= '0'; Firstly is simulated only the modelled ADC, for the aim of
eoc <= '0'; that a piecewise linear voltage source is connected on the
elsif not vsum'above(vctrl_neg) then
out_plus <= '0'; input voltage pin. This source can provide a periodic voltage
out_minus <= '1'; profile, with time period 50ms . The input voltage set values
eoc <= '0';
elsif not vsum'above(vctrl_pos) and vsum'above(vctrl_neg) then
are 0,5 , 1 , 1,5 and 1V . The clock signal is with frequency of
out_plus <= '0'; 100kHz . The stop signal is with period of 400ms . The
out_minus <= '0';
eoc <= '1';
simulated schematic is given on Fig. 5. The output results are
end if; presented on Fig. 6. We can see clearly how the digital output
if (stop = '0') ‐‐ stop signal, bans the work of counter is tracking the variable value of the input voltage. According
then counter := (others=>'0');
end if;
to the formula the corresponding output values for input
voltages – 0,5 , 1 , 1,5V are 400 , 799,9 and 1200. Their
end process;
binary codes are 110010000 , 1100100000 and
v_out == s_out'ramp(1.0E‐6); ‐‐ assuming variables to output signals 1001011000 0. When obtaining the simulation results, we can
vsum == vin ‐ v_out; ‐‐ voltage difference see the correct work of the converter.
clk_up <= out_plus and clk;
clk_down <= out_minus and clk; Secondly, to proof the validity of the proposed model, a
synch <= clk; schematic of resolver to digital converter is created. It’s
end architecture default; structure is presented on Fig. 7.
Fig. 3. A 12-bit tracking ADC behavioural VHDL-AMS model.

226
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

TABLE 2. SIMULATION RESULTS FOR RESOLVER CONVERTER

shaft angle - θ sin(θ) cos(θ) out1 out2
o 0,2588 0,9659 31410 110810
15
30 o 0,5 0,866 60510 103010

45 o 0,707 0,707 84910 84910

Fig. 5. Tracking ADC simulated schematic. 60 o 0,866 0,5 103010 60510

75 o 0,9659 0,2588 110810 31410

90 o 1 0 113510 010

V. CONCLUSION
In this paper a generalized behavioural VHDL-AMS model
of tracking 12-bit ADC has been presented. The model is
implemented and the structure of its description confirm to the
format of the simulation program SystemVision 5.5. The
proposed model accurately simulates the actual performance
of typical tracking ADC. The efficiency of the model was
proved by comparison of the simulation results with
theoretically calculations for piecewise linear input voltages.
Furthermore, the workability of the model was shown by
simulation testing of a schematic of a resolver to digital
Fig. 6. Simulation results for four values of the input voltage. converter. The simulations were performed for several values
The schematic on Fig. 7. includes rotor, primary and of the shaft angles of the rotor. The obtained results confirm
secondary windings on the stator and two ADCs. The stator to the theoretically calculated parameters.
windings are displaced mechanically by 90 o . The primary
winding is excited with an ac reference. The amplitude of ACKNOWLEDGEMENT
subsequent coupling onto the stator secondary windings is a
function of the position of the rotor (shaft) relative to the This paper is part of a project, which is sponsored by the
stator. The resolver, therefore, produces two output voltages research program of the TU-Sofia, Bulgaria.
(S3 − S1, S2 − S4) modulated by the sine and cosine of shaft
angle. These output voltages are passed to two ADCs. The
ADCs track the change in the rotor position and their digital
REFERENCES
outputs respond to the respective analogue value. The
[1] G. Mihov. Digital Electronics, Sofia, Technical Univerity –
SystemVision libraries do not include a model of step motor. Sofia, 2005 (in Bulgarian).
To realize this structure is used an ac source modelled [2] M. Seifart. Analoge Schaltungen. 6 Auflage. Verlag Technik
vpsin(ωt) and two dc sources modelled sin(θ) and cos(θ). The Berlin, 2003 (in German).
dc sources present the sine and cosine of the corresponding [3] T. Floyd. Digital fundamentals, 10th edition, Prentice-Hall,
angle. Each of the dc sources is multiplied with ac source in 2009.
order to form output voltages relevant to the stator ones. The [4] V. Tietze, Ch. Schenk. Electronic circuits. 2nd Edition. Berlin,
Heidelberg, New York. Springer-Verlag, 2008.
simulation is made for six values of the angle − 15 o , 30 o , 45 o ,
[5] PSpice A/D library list, Release 9.2, Cadence Design Systems,
60 o , 75 o and 90 o . The ac source is with amplitude V p equal 2000.
[6] System Vision mixed-signal model library (ver. 5.5), Mentor
to 1V , dc offset voltage 1,5V and frequency 2kHz . The Graphics, 2007.
simulation results are given in Table 2. As you can see the [7] Matlab Simulink release 2011b libraries, The Mathworks,
results confirm to the calculated parameters. 2011.
[8] E. Peralias, A.J. Acosta, A. Rueda, J. Huerdas. VHDL-based
behavioural description of pipeline ADC, Measurement, vol.
31,pp. 47-60, 2002.
[9] M. Rizzi, B. Castagnolo. Efficiency in simulations tools in
designing sigma-delta ADC. AEU - International Journal of
Electronics and Communications, vol. 60, pp. 290-298, 2006.
[10] M. Kovacheva, I. Pandiev. Behavioural VHDL-AMS model for
half-flash analog-to digital converter. Annual journal of
electronics, vol. 5, No 1, pp. 39-42, 2011.
Fig. 7. Schematic of resolver to digital converter.

227
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Development of Parameterized Verilog‑AMS Model

of Photovoltaic Cells
Elissaveta Gadjeva1 and Georgi Valkov2
Abstract – A parameterized Verilog-AMS model of irradiation and the shading factor KSH models the behavior of
photovoltaic cell, module and panel is developed. The model is the cell under reduced light conditions. Its range is [0, 1],
available in two variants: one that includes a detailed Spice diode where 1 is for unshaded cell. R1 and R2 model the shunt and
description and the other which is simplified and includes only series resistance losses. Their values are R1=RSH and R2=RS.
the diode equations that affect the behavior of the PV elements.
Definitions for obtaining the model characteristics from the
The diode models the non-linear behavior, its parameters are
simulation results are provided. The model is realized in the IS and N, where IS is the saturation current and N is the
Dolphin Integration SMASH environment. emission coefficient.

Keywords – Photovoltaic cells, Parameterized models, Verilog-

AMS language.

I. INTRODUCTION
Fig. 1. Parameterized model of a single solar cell.
The design and optimization of photovoltaic systems
require development of computer models of PV cells and A generic symbol for the PV cell model is shown in Fig. 2.
panels. These models are of significant importance for the
simulation of maximum power point tracking (MPPT) control
systems and allow the investigation and optimization of the
dynamics of PV power systems.
Photovoltaic models of PV cells and modules are developed
Fig. 2. Generic symbol of PV cell.
in [1-3]. Methods for parameter extraction and simulation of
photovoltaic modules by taking the manufacturer specified B. Model of PV panel
data are developed in [4-6]. A sun-tracking system based on
the Verilog HDL and using an FPGA chip as controlling The model of PV panel uses the same schematic
platform is proposed in [7]. Combining all analog and digital representation as the single PV cell shown in Fig. 1, but with
blocks of a project, using a behavior HDL like Verilog-AMS scaled parameters to represent a whole panel. A panel consists
and VHDL-AMS [9] allow the entire project to be simulated of NP modules connected in parallel, where each module has
and verified as a whole. A SoC design of solar irrigation NS cells connected in series, as shown in Fig. 3.
control system based on FPGA is offered in [8].
Parameterized Verilog-AMS macromodel is created in the
present paper that can represent a single PV cell, as well as an
entire panel consisting of PV cells connected in series and
parallel. An additional model using a simplified version of the
diode is also described. Definitions for obtaining the model
characteristics from the simulation results are provided,
according to the pattern file syntax used by SMASH [10].

II. MODELS OF PV CELL AND PANEL

Fig. 3. Structure of PV panel.
A. Model of a single PV cell
In the case when NS=1 and NP=1, the model represents a
The schematic representation of the macromodel for a single PV cell. The values for model parameters are
single PV cell is provided in Fig. 1. I1 is a DC current source I1=ICC.KSH.NP, R1=RSH.NP, R2=RS.NS, IS=IS.NP, N=N.NS.
with value I1=ICC.KSH, where ICC is a function of the solar A generic symbol for the PV panel model is shown in
Fig. 4.
1
Elissaveta Gadjeva is with the Department of Electronics at
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000,
Bulgaria, E-mail: egadjeva@tu-sofia.bg.
2
Georgi Valkov is with the Department of Electronics at Technical
University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, Bulgaria,
E-mail: gvalkov@abv.bg. Fig. 4. Generic symbol of PV panel.

228
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

III. VERILOG-AMS IMPLEMENTATION I(n3, n2) <+ V(n3, n2)*Np/(Ns*Rs);

end
OF PV PANEL endmodule

A. Verilog-AMS code of PV panel with full diode description IV. SIMULATED CHARACTERISTICS IN DOLPHIN
INTEGRATION SMASH
The Verilog-AMS implementation of the model has the
following form: The characteristics for the Verilog-AMS model of PV
module pv_cell_01_vams(n1, n2); panel with full diode description are obtained for a single cell
inout n1, n2;
electrical n1, n2, n3; using DC sweep as the primary simulation analysis. To
parameter real Icc= 4.0; determine the dependence on RS, RSH, KSH and temperature, the
parameter real Ksh= 1.0; following parametric sweeps are run together with the primary
parameter real Rs = 1.0m; analysis: RS: value list: 0.1m, 1m, 10m, 100m; RSH: value list: 1,
parameter real Rsh=100.0; 10, 100; KSH: linear from 0.25 to 1, with a step size of 0.25 and
parameter real Ns = 1.0;
parameter real Np = 1.0; TEMPER: linear from 0 to 75, with a step size of 25. The
parameter real Is = 1.0u; remaining model parameters are given in Fig. 4.
parameter real N = 1.5; The current and power characteristics are obtained using
analog begin the waveforms I(V1) and IN(V1.POWER). The following
I(n3, n1)<+ V(n3, n1)*Np/Rsh - Icc*Ksh*Np;
I(n3, n2)<+ V(n3, n2)*Np/(Ns*Rs); characteristics are also obtained from the simulation results:
end current and voltage at the maximum power point IPP and VPP,
diode_pv_sp #(.Is(Is*Np), .N(N*Ns)) fill factor FF, maximum power PMAX, short circuit current ISC,
D_pv (.n2(n3), .n1(n1)); and open circuit voltage VOC. The parameter FF is defined as
endmodule
follows:
The block has two electrical terminals, n2 is the positive one
and n1 is the negative. The model parameters are defined. The PMAX
FF  , (1)
elements I1, R1 and R2 are modeled by their component PT
equations. A diode block is added between nodes n3 and n1, to
model the non-linear behavior. where: PMAX  max  P  , (2)
SMASH is a mixed language simulator and allows
subcircuits defined using the Spice language syntax to be PT  I SCVOC . (3)
added to Verilog-AMS nets and vice versa. The diode_pv_sp The definitions for obtaining PMAX, ISC, VOC, PT, FF, VPP,
block wraps a Spice diode model inside a subcircuit, using the and IPP from the simulation results are created using the
following Spice description: .MEASURE command according to the pattern file syntax
.SUBCKT diode_pv_sp n1 n2 PARAMS: Is=1 N=1 used by SMASH:
.MODEL diode_pv D (Is=Is N=N)
D1 n2 n1 diode_pv .MEASURE ANALYSIS=DC NAME=Pmax
.ENDS + WAVEFORM=IN(V1.POWER) EXTRACT=MAX
.MEASURE ANALYSIS=DC NAME=Isc
B. Model of PV panel with simplified diode description + WAVEFORM=I(V1) EXTRACT=MAX
.MEASURE ANALYSIS=DC NAME=Voc
+ WAVEFORM=I(V1) EXTRACT=X ATY=0
For the purpose of simulating PV cells behavior, the diode + CROSS=DOWN OCCUR=1
is only used in forward direction. Additional effects like .MEASURE ANALYSIS=DC NAME=PT VALUE={Isc*Voc}
junction capacitance can also be omitted. Hence it is possible .MEASURE ANALYSIS=DC NAME=FF VALUE={Pmax/PT}
to simplify the model of PV panel with Spice diode, by .MEASURE ANALYSIS=DC NAME=Vpp
+ WAVEFORM=IN(V1.POWER) EXTRACT=ATMAX ATY=0
replacing the diode with its equivalent equation. The model is .MEASURE ANALYSIS=DC NAME=Ipp
described by the following Verilog-AMS code: + WAVEFORM=I(V1) EXTRACT=Y ATX=Vpp
module pv_cell_02_vams(n1, n2); The current and power characteristics as a function of RS
inout n1, n2; are shown in Fig. 5.
electrical n1, n2, n3;
parameter real Icc = 4.0 from (0:inf); 200m 400m 600m
parameter real Ksh = 1.0 from [0:1]; I(V1) 4A
parameter real Rs = 1.0m from (0:inf); 0.1m
3A
parameter real Rsh = 100.0 from (0:inf); Rs=100m 10m 1m
parameter real Ns = 1.0 from [1:inf); 2A
parameter real Np = 1.0 from [1:inf); 1A
parameter real Is = 1.0u from (0:inf);
parameter real N = 1.5 from (0:inf); W(V1) 0.1m
parameter real Tn = 300.15 from (0:inf); 2
1m
parameter real XTI = 3.0 from (0:inf); 10m
1.5
parameter real Eg = 1.11 from (0:inf); 1
analog begin 500m
I(n3, n1) <+ V(n3, n1)*Np/Rsh - Icc*Ksh*Np Rs=100m
+ Np*Is*(exp(V(n3, n1)/(Ns*N*$vt))-1)
* exp(($temperature/Tn-1)*Eg/(Ns*N*$vt)) Fig. 5. Current and power characteristics as a function of RS: I(V1,RS)
* pow($temperature/Tn, XTI/(Ns*N)); and P(V1,RS).

229
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

High values for the series resistance tend to cause a 1 2 3 456 10 20 30

significant drop in the output power of the PV elements as IPP 3.7

well as reduction of IPP, VPP and FF (Fig. 6). 3.6

The current and power characteristics as a function of RSH 3.5

are shown in Fig. 7. High values for the shunt resistance have 3.4

no impact on the performance, while low values act like an 3.3

additional consumer, reducing the power at the output of the VPP 490m
PV elements and also reducing IPP, VPP and FF (Fig. 8). The 488m
shading factor KSH indicates what part of the PV cell or panel 486m
is not shaded. The output current is proportional to the solar 484m
482m
irradiation and hence proportional to KSH. Lower values for
480m
KSH result in a lower output power (Fig. 9), and also cause a FF 760m
reduction in IPP, VPP and FF, as shown in Fig. 10. 740m
The temperature dependences of the output current and 720m
power characteristics as a function of RS are shown in Fig. 11. 700m
At lower temperatures VOC and VPP are higher. As a result the 680m
I-V characteristic is shifted to the right; the output power at
the maximum power point and FF are also higher. There is
also a small increase in IPP (Fig. 12). Fig. 8. IPP, VPP and FF characteristics as a function of RSH: IPP(RSH);
VPP(RSH) and FF(RSH).
The model of simplified PV panel is compared to the PV
panel model with full diode description. The temperature
200m 400m 600m
dependence of the I‑V characteristic is shown in Fig. 13. I(V1) 4A
Ksh=1.00
100u 1m 2m 10m 3A
IPP 0.75
3.6 2A
0.50
3.2 1A
0.25
2.8 W(V1)
2
1.00 1.5
VPP 480m 0.75 1
440m 0.50 500m
400m 0.25
360m
320m Fig. 9. Current and power characteristics as a function of KSH:
FF I(V1,KSH) and P(V1,KSH).
700m
600m 400m 600m 800m
500m IPP 3.5
400m 3
2.5
Fig. 6. IPP, VPP and FF characteristics as a function of RS: IPP(RS); 2
VPP(RS) and FF(RP). 1.5
1
200m 400m 600m VPP 490m
I(V1) 100 4A 480m
10 470m
3A
460m
2A 450m
Rsh=1
1A 440m
FF 760m
W(V1) 100
2 756m
10
1.5
752m
1
Rsh=1 748m
500m

Fig. 7. Current and power characteristics as a function of RSH: Fig. 10. IPP, VPP and FF characteristics as a function of KSH:
I(V1,RSH) and P(V1,RSH). IPP(KSH); VPP(KSH) and FF(KSH).

230
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

200m 400m 600m

I(V1) 4A
4
2 3A
27 2A
3
52
1A 2
T=77 I(V1).spice
W(V1) 1
2 I(V1).simplified
2
1.5 0
27
1 0.2 0.3 0.4 0.5 0.6
52
500m
T=77 Fig. 13. Comparison between the model with Spice diode and the
Fig. 11. Current and power characteristics as a function of the simplified model. Current and power characteristics as a function of
temperature: I(V1,T) and P(V1,T). the temperature: I(V1,T) and P(V1,T).

20 40 60
IPP 3.76 REFERENCES
3.72
3.68 [1] H-L., Tsai, Ci-Siang Tu, and Yi-Jie Su. “Development of
3.64
generalized photovoltaic model using MATLAB/SIMULINK”,
3.6
Proceedings of the World Congress on Engineering and
3.56
VPP
Computer Science 2008, WCECS 2008, October 22 - 24, San
520m Francisco, USA, 2008.
480m [2] I.H., Altas, A.M., Sharaf. “A photovoltaic array simulation
440m
model for Matlab-Simulink GUI environment”, International
Conference on Clean Electrical Power,. ICCEP '07. 21-23 May
400m
2007,pp. 341 – 345, Capri, 2007.
FF
780m [3] Hernanz, R., C. Martín, J.J. Belver, L. Lesaka, Z. Guerrero, E.
760m Puelles Pérez.”Modelling of photovoltaic module”, International
740m
Conference on Renewable Energies and Power Quality,
720m
(ICREPQ’10) Granada (Spain), 23th to 25th March, 2010.
[4] D. Sera, R. Teodorescu, P. Rodriguez. “PV panel model based
on datasheet values”, IEEE International Symposium on
Fig. 12. IPP, VPP and FF characteristics as a function of the Industrial Electronics, ISIE 2007,June 4-7 2007, Vigo, Spain,
temperature: IPP(T); VPP(T) and FF(T). ISBN: 978-1-4244-0754-5, pp. 2392-2396, 2007.
[5] G. Farivar and B. Asaei. “Photovoltaic module single diode
The current characteristics match for all of the simulated model parameters extraction based on manufacturer datasheet
temperatures. The relative difference is smaller than 2×10-6 %, parameters”, IEEE International Conference on Power and
hence it is advisable to use the simplified model in order to Energy, PECon 2010Nov. 29-Dec. 1 2010, Selangor, Malaysia,
speed the simulation of large circuits, as the accuracy is not ISBN: 978-1-4244-8947-3, pp. 929-934, 2010.
sacrificed. [6] N.N.B. Ulapane, C.H. Dhanapala, S.M. Wickramasinghe, S.G.
Abeyratne, N. Rathnayake and P.J. Binduhewa. “Extraction of
parameters for simulating photovoltaic panels”, 2011 IEEE 6th
V. CONCLUSIONS International Conference on Industrial and Information Systems
(ICIIS), 16-19 Aug. 2011, Peradeniya, Sri Lanka, ISBN: 978-1-
A behavioral parameterized Verilog-AMS macromodel has 4577-0032-3, pp. 539-544, 2011.
been developed for PV cell and PV panel consisting of cells [7] Cheng Sen, Zhao Ping, He Hongkun, Ji Qianqian and Wei Xu.
connected in series and in parallel. The model of PV panel can “An improved design of photo-voltaic solar tracking system
also represent a single cell. The detailed model relays on a based on FPGA”, AICI'10, International Conference on
Spice diode model to describe the non-linear behavior of the Artificial Intelligence and Computational Intelligence, 23-24
October 2010, Sanya, China, ISBN: 978-1-4244-8432-4, pp.
PV cell. An alternative simplified model is also provided. It 267-271. 2010.
includes only the diode equations that are important for the [8] Ze Cheng, Dan Zhong, Baolin Li and Yanli Liu. “The SoC
modeling of the PV behavior. It is a pure Verilog-AMS design and realization of small scale solar irrigation control
implementation, and does not depend on mixed language system based on FPGA”, 2011 International Conference on
simulators. Both models are simulated in the Dolphin Electric Information and Control Engineering (ICEICE), Apr.
Integration SMASH environment. The basic PV 15-17, 2011, Wuhan, China, ISBN: 978-1-4244-8036-4, pp.
characteristics are obtained from the simulation results. 887-890, 2011.
[9] F. Pêcheux, C. Lallement, A. Vachoux, "VHDL-AMS and
Verilog-AMS as Alternative Hardware Description Languages
ACKNOWLEDGEMENT for Efficient Modeling of Multi-Discipline Systems", Paper
#1446, IEEE, ISSN 0278-0070, 2005.
The investigations are supported by the project [10] Dolphin Integration SMASH overview,
http://www.dolphin.fr/medal/smash/smash_overview.php
№122PD0026-03.

231
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Optical Control through Stencils Cutting in Surface

Mount Technology
Valentin Videkov1, Aleksei Stratev2, Georgi Furkov3
Abstract –This paper analyzes cutting techniques to conduct in terms of area, masks where the same precision in making
destructive testing in surface mounting. It showcases two
methods to control processes during manufacture of technology the holes on the entire area is applied. At the same time, there
equipment and process control. Experimental results from a
are no limitations related to chemical methods’ pickling to
process for manufacture of laser cut stencils are demonstrated.
produce small holes.
Keywords – SMT, optical control, laser cut stencils, An advantage of the method is also that a lower quantity of
consumables is used during the manufacture of the stencils.
Laser cutting treats the metal plate directly, as opposed to
chemical etching, where the metal plate undergoes a
I. INTRODUCTION procedure of preliminary treatment through photolithography,
in addition to expenditures incurred to pickle and rinse the
Surface-mount technology is a major technology used in metal plate.
the manufacture of electronic equipment, from basic LED An advantage of the laser cut stencils is the ability to get
light sources [1] to complex computer systems and mobile relatively vertical walls of the holes – Fig.1.
devices [2]. Control over the technological process is crucial
due to the widespread use of this technology. Destructive and
non-destructive control methods exist. Optical methods for
control are applied more frequently, including automated
process control methods [3]. In spite of this automation, there
are control points where the automated optical control systems
are inappropriate to use. Those are processes that are applied
relatively rarely and are designed to control individual
elements. For example, control over application of solder
paste in stencils. Solder paste-laying masks are manufactured
based on different technologies [4], and the optical control is
applied to assess the manufacturing technology. Both the
dimensions and the shape of the resulting holes can be
controlled though this method [5 by us]. The standard optical Fig. 1 Cutting of a laser cut stencil. 1- plane of the mask, 2 – edge
control in stencils sometimes is not able to evaluate the impact of the hole, 3 – wall of the hole
of all technological parameters and cannot address all
questions. This paper will demonstrate a method for optical This verticality of the cutting, combined with the extreme
control on stencils that allows for a broader assessment of the movement precision of the laser beam and the small
process for laser cutting of stencils. dimensions of the spot (cut), allow for complex holes
featuring different orientations – fig. 2.
II. LASER CUT STENCILS

Laser cut stencils feature many advantages compared to

chemically pickled ones. Foremost, one can produce large,

1
Valentin Videkov, Faculty of Electronic Equipment and
Technology at Technical University of Sofia, 8 Kl. Ohridski Blvd,
Sofia 1000, Bulgaria, E-mail: videkov@tu-sofia.bg.
2
Aleksei Stratev, Extramural PhD student at Technical University
of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, Bulgaria E-mail:
astratev@ivastech.com.
3 Fig. 2 Shapes of holes in laser cut stencils
Georgi Furkov, Extramural PhD student at Technical University
of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, Bulgaria, E-mail:
georgi_farkov@festo.com

232
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

III. OPTICAL CONTROL

Both destructible and non-destructible optical control can
be applied in the case of laser cut stencils. The shape and the
dimensions of the holes can be viewed in horizontal plane in
the case of non-destructible control. Variants of such an
observation in two modes are showed on Fig. 2. In the case of
observation in reflected light mode, one can see the surface of
a stencil -1, the shape and location of different holes – 2, 3.
For the purpose of precise control over the shape and
dimensions of the holes, an observation method based on
transmitted light is used and then figurations become
contrasting – 4. Fig. 5 Different surface in laser cutting
In the case of these stencils, the quality of the wall itself is
crucial, not only the preciseness of the hole and the verticality Fig. 5 shows the shapes of the wall at edge (angle of the
of the walls. It might feature a different surface as well as hole) – 1, top of a groove – 2, pore – 3, lusterless surface – 4.
defects – Fig. 3. The direct optical observation does not allow to assess the
surface of the cut in detail despite that special samples have
been made.
The height of the grooves, the presence of sub-surface
cavities, etc., should be assessed. The assessment of the
surface as a profile of the plane in different levels is of special
interest.

IV. EXPERIMENTAL RESULTS

A standard approach to profile a surface involves the use of

mechanical profile-measuring devices. Here, there are two
restrictions. The wall of the hole is relatively small and
narrow, requiring high level of precision during manipulation.
Fig.3 A wall of cut stencil. 1 – wall, 2 – defect, metal droplet, 3 – On the other hand, the mechanical profile-measuring device is
edge not applicable in the case of complex shapes – for example,
curved groove.
Samples that have been cut along the holes have been used, Bearing the abovementioned in mind, a decision has been
allowing for better observation, while enabling observation of made to make horizontal cross-sections, taking profiles of the
edges and walls – Fig. 4. hole. For this purpose, epoxy resin has been poured on
samples of stencils, which have been placed vertically to the
cutting plane – Fig. 6.

Fig. 4 Samples for observation of cuts

Fig. 6 Samples of cutting in horizontal plane.
At the above figure, one can see the surface of the stencil -
1, its wall – 2, and the edge of the hole – 3, where angled The width of the sample has been under continuous control
lighting is used. during the cutting process to determine the cutting plane. The
In optimizing the modes of operation of the laser (power, cutting has been performed at 250 rpm, pressure of 30
impulse frequency, duration, movement speed), the possibility Newtons per sample, and processing period of 30 seconds.
to compare the outcome of cutting against the consequential Cut levels have been within 30 µm. In the beginning, cuts
surface of the wall is essential. It can take the shape of appeared at the levels of the metal droplets that occurred due
channels, grooves, lusterless, and others – Fig. 5. to the cut – Fig.7, and then at the surface of the stencil.

233
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 10 Contours of the cut surface after extracting

Fig. 7 Horizontal cutting of a stencil. 1 – metal droplets.

When performing cutting at particular distances, the shapes

of the wall as resulting from different handling techniques
emerged – Fig. 8.

Fig. 11 3D diagram

V. CONCLUSION
This paper provides a method for destructive control over
Fig. 8 Different modes of cutting. Cuts after grinding. laser cut stencils designed for surface-mount assembly. This
method allows for a wall profile of the hole, measured in
Software has been used to process the optical image to absolute measurements, to be achieved and makes it possible
increase contrast and get sharp contour – Fig. 9. to complement the assessment of cutting parameters.
Combining profiles that have been accomplished at different
cut levels with digital presentation of those profiles allows for
creation of a 3-D digital image of the surface.
The efforts to obtain such a digital image using other
methods, like mechanical scanning through profile-measuring
device, face difficulties.

REFERENCES
[1] http://www.lc-led.com/ecommerce4.html open 2012-02.
[2] http://www.national.com/mpf/LM/LMV243.html#Overview
open .2011-12.
Fig. 9 Contrast processing of a cutting. [3] Tae-Hyoung Park at all, Path Planning of Automated Optical
Inspection Machines for PCB Assembly Systems, International
A diagram of the contour itself is easy to extract, following Journal of Control, Automation, and Systems, vol. 4, no. 1, pp.
the application of a respective processing method, where such 96-104, February 2006
a contour exists – Fig. 10. [4] Steplewski, W., at all, Stencil Design for Lead-Free Reflow
The transformation of the resultant contour into digital data Process, Electronics Technology, 30th International Spring
is not hard to realize as well. All and every cut is processed Seminar on 9-13 May 2007, ISBN: 987-1-4244-1218-1, pp. 330
using the demonstrated methodology and is transformed into - 334
[5] Stratev A., Farkov G., Videkov V., Optical Control of Laser Cut
digital data. Those may be presented graphically and may be Stencils, XLVI International Scientific Conference on
used in different types of processing. The below figure 11 Information, communication and energy systems and
shows a sample how the results derived from the wall’s technologies, Proceedings of Papers, V2, Serbia, Nish, 2011,
contours can be displayed as a 3-D diagram. ISBN 978-86-6125-3, pp. 529 – 532

234
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Power Quality According to EN 50160

Nikolce Acevski1, Kire Mijoski2 and Tomce Mijoski3
Abstract – Although electricity, because of its possibility for suppliers equalize the concept of power quality with supply
easy conversion into other types of energy is the most used and reliability (continuity in delivery), defining the system as
most abundant form of energy, its features, except the frequency 99.98 percent reliable. Manufacturers of electrical equipment
and amplitude of voltage is standardized too late. can define electrical energy as being quality if their equipment
Determination of additional criteria for evaluating the quality
functions properly. Of course, such approach is subjective and
of electricity has become necessary with the increasing of the use
of electric consumers with nonlinear character. Such electrical open to criticism. But, however, the most reasonable, and in
devices and equipment, on one hand they need quality voltage for any case there must be a priority, the definition behind which
their operation, on the other hand inject many disturbances that stands the user of electricity, its consumer. The PQ analysis
negatively affect the characteristics of voltage which is generally includes the following voltage parameters: power
connected. These are a kind of customer whose work is mainly frequency, supply voltage variations, rapid voltage changes
based on microprocessor and logic circuits and particularly is (and flicker), supply voltage dips, short interruptions, long
important to be exposed to less interruption at work because interruptions, temporary over voltages, supply voltage
they can cause major economic damage. With the increasing of unbalance, harmonic voltage, mains signalling voltage.
the number of nonlinear consumers it is also increased the
interest for power quality in recent decades.
B. Standard EN 50160
Keywords – Distribution network, electric power quality,
European norm EN 50160, network performance analyser. EN 50160 gives the main voltage parameters and their
permissible deviation ranges at the customer’s point of
I.INTRODUCTION common coupling in public low voltage (LV) and medium
voltage (MV) electricity distribution systems, under normal
Electrical energy is a product and, like any other product, operating conditions.
should satisfy the proper quality requirements. If electrical Supporting the requirement to define voltage characteristics
equipment is to operate correctly, it requires electrical energy in terms of frequency, magnitude, waveform and symmetry,
to be supplied at a voltage that is within a specified range EN 50160 provided definitions and in some cases
around the rated value. A significant part of the equipment in measurement methods and compliance levels for 12
use today, especially electronic and computer devices requires characteristics of the supply voltage:
good power quality (PQ). However, the same equipment often
causes distortion of the voltage supply in the installation, TABLE I
EN 50160 COMPLIANCE LIMITS
because of its non-linear characteristics, i.e. it draws a non-
sinusoidal current with a sinusoidal supply voltage. Thus, Supply voltage characteristics according to
No Parameter
maintaining satisfactory PQ is a joint responsibility for the EN 50160
supplier and the electricity user. The responsibilities of the Power frequency LV, MV: mean value of fundamental
suppliers and the consumers of electricity are defined by measured over 10 s ±1% (49.5 - 50.5 Hz) for
1
99.5% of week -6%/+4% (47- 52 Hz) for
internationally accepted electrical standards. In our area the 100% of week
most suitable standard for power supplying in distribution Voltage magnitude LV, MV: ±10% for 95% of week, mean 10
2
networks is the European standard EN 50160. variations minutes rms values
This paper presents the results of measuring the PQ at Rapid voltage changes LV: 5% normal 10% infrequently Plt ≤ 1 for
95% of week MV: 4% normal 6%
several points in DN using auxiliary sophisticated network 3
infrequently Plt ≤ 1 for 95% of week
analyzer. (formula 1)
Supply voltage dips Majority: duration <1s, depth <60%. Locally
4 limited dips caused by load switching on:
II. POWER QUALITY LV: 10 - 50%, MV: 10 - 15%
Short interruptions of LV, MV: (up to 3 minutes) few tens - few
5
supply voltage hundreds/year.Duration 70% of them < 1 s
A. Term of PQ Long interruption of LV, MV: (longer than 3 minutes) <10 -
6
supply voltage 50/year
Different ways of understanding what PQ is are present Temporary, power LV: <1.5 kV rms MV: 1.7 Uc (solid or
7 frequency impedance earth) 2.0 Uc (unearthed or resonant
worldwide, including Macedonia. For example, power
overvoltages earth)
Transient overvoltages LV: generally < 6kV, occasionally higher;
8
1 rise time: ms - μs. MV: not defined
Nikolce Acevski is with the Faculty of Technical Sciences,
Supply voltage LV, MV: up to 2% for 95% of week, mean
7000 Bitola, R. Macedonia, e-mail: nikola.acevski@uklo.edu.mk.
2 9 unbalance 10 minutes rms values, up to 3% in some
Kire Mijoski master student, Faculty of Technical Sciences, locations
7000 Bitola, R. Macedonia, e-mail:.kire.mijoski@yahoo.com Harmonic voltage 95% of the time in 1 week, THD <8% (Table
3
Tomce Mijoski master student, Faculty of Technical Sciences, 10
2, formula 2)
7000 Bitola, R. Macedonia, e-mail:.tomcemijoski@yahoo.com.

235
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

11 Interharmonic voltage LV, MV: under consideration Next measurement is performed in the period from
Mains signaling 9% - 100 Hz; 1% - 100 kHz, 99% of the time 20.01.2011 to 28.01.2011 at 0,4 kV voltage level.
12
voltage in 1 day
Measurements were performed simultaneously at supply
terminal of one consumer and at the LV level of TS Crniliste
12
Psti3

20/0,4 which supplies this consumer.
Pit  3 (1)
i 1 12
B. Showing the results
40
THD   U 
2 After finishing the process of measurement, data is
h (2)
h2 transferred to the PS. Then, by using special software package
the summary results of measurements are obtained according
TABLE II to standard EN 50160. Because of the size of the number of
VALUES OF INDIVIDUAL HARMONIC VOLTAGES AT THE SUPPLY the data it is not possible to display all the data and charts
TERMINALS FOR ORDERS UP TO 25, GIVEN IN PERCENT OF UN provided from measurements. It will be shown only those who
are most interesting and most important for this paper.
Odd harmonics Even harmonics In the graphical results presented in additional in this paper,
Not multiples of 3 Multiples of 3 Not multiples of 3
the limit values according to standard EN50160 are marked by
Relative Relative Relative
Order h voltage Order h voltage Order h voltage violet color.
(%) (%) (%) Line voltage variations at 20 kV voltage level of TS Polog
5 6,0% 5 6,0% 5 6,0% are given in the first picture. It can be noticed that the line
7 5,0% 7 5,0% 7 5,0% voltages variations throughout the period were within the
11 3,5% 11 3,5% 11 3,5% allowable limits according EN 50160.
13 3,0% 13 3,0% 13 3,0%
On the second and the third picture are given the results for
17 2,0% 17 2,0% 17 2,0%
19 1,5% 19 1,5% 19 1,5% short interruptions of supply voltage and values of flicker
23 1,5% 23 1,5% 23 1,5% severity and total harmonic distortion factor at 20 kV level of
25 1,5% 25 1,5% 25 1,5% TS Polog 110/20 kV/kV.
It can be seen that on 10.11.2011 the THD is more than 7.5
% (very close to 8%) and there is deviation of flicker severity
III. SYSTEM ANALYZER OMNI-QUANT (Plt>1) about several hours in 26.11.2011, but still the values
of these parameters also vary within the allowable limits
The OMNI-QUANT mobile is preferably used as portable according to EN 50160.
device in changing locations. Four voltage and current As it was mentioned above that this is completely automatic
measuring inputs each one allows power measurements, fault TS with automatic voltage regulation and integrated device
analyses and recording functions along with the evaluation of for continuous control of power quality and this significantly
the voltage quality. affects to the values of all measured parameters to moving in
Complete scanning and calculation of the following values: approved limits.
• Phase voltage (L-N) and phase-to-phase voltage (L-L) In Fig. 4 and Fig. 5 are presented the results from
• Star point voltage and symmetry L1...L3 measurement of PQ at 0.4 kV voltage level of TS Crniliste.
• Frequency (identical for all channels) Similar as in the previous example (TC Polog) most of the
• Current, total current L1...L3, total current L1...L3 + N parameters are moving within the limits according to the
• Power (P, Q, S, power factor, distorted reactive power) standard EN 50160. From the shown parameters, the flicker
• Power of the fundamental (active power, reactive power, severity, the total harmonic distortion factor and the supply
apparent power, cos) voltage unbalance did not deviate from the allowable limits
• L1...L3 total of the above power variables prescribed by EN 50160 over the entire period. With the
• Harmonics 1...50th order exception on 23.01.2011 when it has a deviation of phase
• Intern harmonic of U and I up to 2.5 kHz voltages above the limits for several hours.
• Ripple control level In contrast to the values measured at 0.4 kV level of TS
• Distortion factor (THD) of V and A. Crniliste, the values measured in the same period at supply
terminal of one consumer supplied by TS Crniliste,
IV. EXAMPLES OF MEASURING THE PQ ACCORDING significantly differ. In fact, with the exception of the total
harmonic distortion factor, which moves within the allowable
TO EN 50150
limits, the other parameters significantly deviate outside the
limits. The results from this measurement are presented in Fig.
A. Measuring 6 and Fig. 7.

Practical measurement of power quality using measuring

device OMNI-QUANT was made in several times in different
V. CONCLUSION
points in distribution system.
The first measuring was performed in the period from Today’s electronic loads are susceptible to transients, sags,
01.11.2011 to 01.12.2011 at 20 kV voltage level of the TS swells, harmonics, momentary interruptions, and other
Polog 110/20 kV / kV. It is important to note that this is disturbances that historically were not cause for concern. For
completely automatic TS with automatic voltage regulation sensitive loads, the quality of electric service has become as
and integrated device for continuous control of power quality. important as its reliability. Power quality is a new
phenomenon. Events such as voltage sags, impulses,
harmonics, and phase imbalance are now power quality
236
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

concerns. Power quality problems have a huge economic

impact. As a result, any discussion of power system reliability
must also include power quality.
The main document dealing with requirements concerning
the supplier’s side is standard EN 50160, which characterizes
voltage parameters of electrical energy in public distribution
systems.
In this paper were presented several measurements of PQ
using modern network analyzer OMNI-QUANT. By using a
special software package the summary results of
measurements are obtained according to standard EN 50160.
From the first measurement can be seen that measured
values fully satisfy the limits prescribed in EN 50160. This
PQ is due to the fact that this is completely automatic TS, with
automatic voltage regulation and integrated device for
Fig. 2. Statistical report for short interruptions of supply voltage in
continuous control of power quality. the observed period
Serious deviations from the limits according to EN 50160
are detected during measuring the PQ at supply terminal of
one consumer supplied by TS Crniliste. LV lines with great
length and TS with outdated equipment are the reason of such
results for the PQ. Hence, the investments in distribution
systems are necessary to achieve satisfactory results for the
PQ.

REFERENCES
[1] Angelo Baggini. Handbook of Power Quality. University of
Bergamo, Italy: 2008.
[2] Dr. Željko Novine. „Kvaliteta Elektrićne Enrgije“. Prirućnik.
Sveučilište J. J. Strossmayera, Osijek:2006.
[3] Љ. Николоски “Хармоници и Други Повратни Влијанија во
Електроенергетскиот Систем” Предавања на ЕТФ
Скопје:2008.
[4] EN 50160, Voltage characteristics of electricity supplied by
public distribution systems, 1999.
[5] H. Markiewicz, A. Klajn “Voltage Disturbances Standard EN
50160 - Voltage Characteristics in Public Distribution Systems”
Wroclaw University of Technology July 2004.
[6] Љ. Николовски, Г. Рафајловски, К. Најденовски, Д.
Арсеновска, Х. Хаидвогл, Ј. Зирлинген. “Технички Правила
за Проценка на Повратните Влијанија во Мрежата”, Скопје:
2010.

VI. ADDITION

Fig. 3. Flicker severity and total harmonic distortion factor at 20 kV

level of TS Polog 110/20 kV/kV

Fig. 1. Voltage magnitude variations at 20 kV level of TS Polog

110/20 kV/kV

237
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 4. Flicker severity and total harmonic distortion factor at 0,4 kV Fig. 6. Flicker severity and total harmonic distortion factor at supply
level of TS Crniliste 20/0,4 kV/kV terminal of one consumer supplied by TS Crniliste

Fig. 5. Supply voltage unbalance and voltage magnitude variations at Fig. 7. Supply voltage unbalance and voltage magnitude variations at
o,4 kV level of TS Crniliste 20/0,4 kV/kV supply terminal of one consumer supplied by TS Crniliste

238
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Using H∞ synthesis for finding settings of single channel

power system stabilizers of synchronous generators
Konstantin Gerasimov1, Petko Petkov2 and Krum Gerasimov3
Abstract –This paper presents a methodology for finding set- 4 phase-shifting blocks and there are as well PSS2A of ABB
tings of single channel power system stabilizers by approxima- with 2 blocks.
tion of the frequency response of the synthesized H∞ controller
satisfying the requirements for maximal damping of the syn-
chronous generator electromechanical oscillations and minimiza-
tion of the measurement noise. Test results are presented for a
real synchronous generator from the Bulgarian electric power
system. The advantages of the proposed methodology are dis-
cussed.

Keywords – H∞ synthesis, power system stabilizer, tuning

I. INTRODUCTION
Modern power plants are equipped with power system sta-
bilizers (PSS) for damping the electromechanical oscillations
of synchronous units. In accordance to their structure we dif-
ferentiate them as single- and dual-channel. A typical single
channel PSS with rotor speed as input signal (PSS1A) [1] is
shown in Figure 1.
Fig. 2. Block-diagram of Alstom realization of PSS2A
The input filters are tuned in such a way that they do not
pass the settled deviations of the regime parameters, and the
PSS phase-shifting blocks – to maximally damp the electro-
mechanical oscillations. The settings can be determined by a
Fig. 1. Block-diagram of PSS1A variety of different methodologies [2-4].
The purpose of this paper is to present a methodology, de-
In the electric power system (EPS) of Bulgaria mainly sin-
veloped by the authors, for single channel PSS tuning based
gle channel PSS are used, with input signal form the equiva-
on H∞ synthesis, and to discuss its advantages.
lent sum of the generator active power (Pe) and rotor speed
(ω). This equivalent input signal is obtained after the signals
of Pe and ω pass through input filters and then once again II. METHODOLOGY FOR SINGLE CHANNEL PSS
through a torsion filter which rejects the torsion oscillations
TUNING BASED ON H∞ SYNTHESIS
originating from the generator rotor. These PSS are classified
as type PSS2A and PSS2B.
The general structure of PSS2A of Alstom is shown in Fig- A. Mathematical model
ure 2. The difference between the different manufacturers’
modifications is in the number of the phase-shifting blocks For analysis of the electromechanical oscillations of the
included (and in some elements in the input filters). For ex- motors in EPS a mathematical description linearized around a
ample, in Bulgaria there are PSS2A from Alstom with certain operating point is used. The size of this mathematical
description is too big due to the great number of elements in
1
the modern united systems. Because the purpose is to tune a
Konstantin Gerasimov is a PhD student at the Department “Pow- particular PSS of a particular synchronous generator, the au-
er Engineering”, Technical University of Varna, Studentska Str. 1,
thors have developed a methodology for frequency aggrega-
Varna, 9010, Bulgaria , E-mail: kkgerasimov@tu-varna.bg
2
Petko Petkov is a Professor at the Faculty of Automation, Tech-
tion of the multidimensional EPS mathematical description in
nical University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, Bulgaria, respect to the studied generator buses [5,6]. The descriptions
E-mail: php@tu-sofia.bg results in the following structure:
3
Krum Gerasimov is a Professor at the Department “Power Engi-
neering”, Technical University of Varna, Studentska Str. 1, Varna,
9010, Bulgaria , E-mail: k.gerasimov@tu-varna.bg

239
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Having in mind the above, the synthesis model in Figure 4

objectifies to the structure in Figure 5:

Fig. 3. Block-diagram of the linearized mathematical

description for determination of PSS settings of a generator
The mathematical description of the building elements in
Figure 3 are obtained according to [2,5,7].
Fig. 5. Block-diagram of the H∞ synthesis, based on signals,
of stabilizer K
B. H∞ synthesis The weighting functions of reference (Wp, Wv) and disturb-
ance (Wnoise) can be constant or dynamic and describe the rela-
The general formulation of the H∞ control problem can be tive importance and/or the frequency contents of the inputs.
presented by the block-diagram in Figure 4 [8,9]. The weighting function Wω sets the requirements in respect to
the degree of damping of the rotor speed, and Wu – the re-
quirements concerning the limitations of the PSS output sig-
nal. As seen in Figures 1 and 2 the PSS output is equipped
with non-linear limiters.

C. Algorithm

Fig. 4. Block-diagram of H∞ control design The algorithm for single channel PSS tuning consists of the
following steps:
In this form the “external” input w is the vector of all sig- 1) Formulation of the mathematical description of the
nals which come into the system and the “error” z is the vec- studied generator, as shown in Figure 3;
tor of all signals which are necessary to describe the behavior 2) Formulation of weighting functions and of transfer ma-
of the closed-loop system. P contains the plant transfer matrix trix P, as shown on the block- diagram in Figure 5;
G and weighting functions which are specific for every syn-
3) Using the constructed transfer matrix P an H∞ stabi-
thesis problem. K is being synthesized (searched) control
lizer is synthesized by means of the MATLAB® func-
function. The standard task for H∞ optimal control is to find a
tion hinfsyn;
stabilizing function K which minimizes: 4) Tuning of the fixed-structure single channel PSS in
 max   Fl  P, K  j  
Figure 1 or in Figure 2 so that its frequency response is
Fl  P, K  
(1) as close as possible to the frequency response of the

mathematically synthesized stabilizer. This is achieved
In MATLAB® this task is solved by the hinfsyn function by searching for coincidence, of the phases or of the
from Robust Control Design® 3 toolbox. amplitudes of the frequency response in the frequency
The synthesis of control function based on signals is a range of the electromechanical oscillations, using ap-
common approach to MIMO problems for which simultane- proximating functions which solve non-linear least
ously a few different (and usually controversial) goals are square problems.
required. In this particular problem the following goals are 5) Analysis of the behavior of the tuned PSS. The step
set: and frequency response are recalculated and the ful-
- maximal damping of the electromechanical oscillations fillment of the goals, set during the synthesis, is as-
manifested in rapid damping of the rotor speed devia- sessed. If the results are unsatisfactory, first a return to
tion and the generators active power deviations. Thus step 4 is made and the structure and parameters of ap-
the generators influence over the rest of the EPS during proximation are varied. If even this cannot lead to sat-
transient processes will be minimized; isfactory results a return to step 2 is made where the
- maximal filtration of the measurement noise. Passing weight functions have to be reconsidered and from
this noise through PSS leads to high frequency oscilla- there on the process repeats.
tions in the excitation circuit and thus the other genera- 6) Construction of model with uncertainties for the pur-
tor regime parameters. It is even possible that there pose of the robust analysis. This can be done with the
may be a 50 Hz component in the noise which can re- help of the developed by the authors software tool
sult in very troublesome resonance phenomena. RobustPSS [10], allowing modeling of structured un-
certainty, presented in state space, and of unstructured

240
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA
Bode Diagram

uncertainty, presented in the frequency domain, which 160

140
1

are caused by the uncertainties in the generator load 120

and the system operating point. 100

Magnitude (abs)
2
80
7) Analysis of the robust stability. It is done by means of 60

the MATLAB® function robuststab. 40

20

0
90

III. TEST RESULTS 0

2

Phase (deg)
The proposed algorithm is tested for tuning Alstom PSS2A -90
1
of a real 370 MW synchronous generator from a Bulgarian -180

thermal power plant. The frequency equivalentation of the -270

5 10 15 (Hz)
Frequency 20 25 30
EPS in respect to the generator bus is reduced to order of 20.
Fig. 6. Frequency response of: 1 – the H∞ synthesized controller
The weighting functions used in the synthesis are as follows:
2 –PSS2A, tuned by approximation of the amplitude
W p  1; Wv  1; Bode Diagram

400

0.0521  s  9.48  s  1293  s  7.429 10

3 2 4
Wnoise 
350
; 2

s 3  1696  s 2  2.147 105  s  9.018 107

300

250

Magnitude (abs)
46.17  s 2  41.5  s  1.442 200

Wu  0.8; Ww  2
1
150

s  0.1198  s  0.0003 100

The H∞ synthesis is carried out under the assumption that

50

0
90
the input filters are tuned well because it is a common prac- 45 1
tice, due to subjective reasons, that it is not allowable to 0
2
Phase (deg)

-45
change the settings of the input filters. In this particular case -90

the settings are as follows: ks3 = 1 p.u.; TW1 = 7 s; -135

TW2 = 7 s; TW3 = 7 s; ks2 = 0.86 p.u.; T7 = 7 s; T8 = 0.6 s; -180

-225
T9 = 0.15 s; M = 4; N = 1. 5 10 15 (Hz)
Frequency 20 25 30

Under these conditions the H∞ synthesized controller has Fig. 7. Frequency response of: 1 – the H∞ synthesized controller
2 –PSS2A, tuned by approximation of the phase
frequency response shown in Figures 6 and 7. In Figure 6 it is
compared with PSS2A tuned by approximation of the ampli-
tude and in Figure 7 – tuned by approximation of the phase.
One should not forget that the fixed-structure PSS settings can
vary only in certain ranges. In this case T1, T3, T11, and T31 =
0 ÷ 10 s, while T2, T4, T21, and T41 = 0.015 ÷ 3 s. Due to sub-
jective reasons, the authors have chosen ks1 to vary between 5
and 20 p.u.
It is clear that in this particular case the approximation by
amplitude gives better results and this is why it will be used.
The obtained in this way settings are: ks1 = 5 p.u.;
T1 = 0.0951 s; T2 = 0.0367 s; T3 = 0.6686 s; T4 = 0.0367 s;
T11 = 0.0967 s; T21 = 0.0367 s; T31 = 0 s; T41 = 0.0664 s.
Fig. 8. Frequency response of PSS2A (1), tuned by approximation of
The gain ks1 is relatively small and we can afford to in-
the amplitude and ks1 = 9 p.u.,
crease it 1.8 times without the PSS output signal to reach the compared to the PSS output signal limitation (2)
PSS output signal limitation (see Figure 8).
0.14
In general we could fine adjust the gain ks1 because it
doesn’t affect the PSS phase compensation and exactly it is 0.12

crucial for the right operation of PSS. But by changing ks1 we

0.1
change the degree of damping. For this reason from here on 1
Singular value (abs)

the presented results are for ks1 = 9 p.u. and the effect can be 0.08

clearly seen in Figure 9. It shows a significant damping of the

0.06
rotor speed oscillations when the tuned PSS is switched on. A
significant damping of the active power oscillations is 0.04 2

achieved as well and this can be seen in Figures 10 and 11.

0.02

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency (Hz)

Fig. 9. Maximal singular values of transfer matrix from all inputs to

Δω of the generator: 1 – without PSS; 2 – with PSS2A (ks1 = 5 p.u.);
3 – with PSS2A (ks1 = 9 p.u.)

241
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

IV. CONCLUSION
From: Vref
1.5

1
1
To: dU

0.5 As a conclusion it can be summarized that the proposed al-

2
0
1.5
gorithm has as main advantage that given a particular fixed
1 structure PSS, one can tune it considering simultaneously the
0.5
Amplitude

following limiting requirements:

To: dP

0
-0.5
-1
2
1
1) ensure quality of the transient processes (better
-1.5
damping of the generator electromechanical oscilla-
-0.01
0
2 tions);
2) the PSS output signal doesn’t reach its limitation;
To: dw

-0.02
-0.03
-0.04 1 3) suppresses the measurement noise.
-0.05
0 1 2 3 4 5
Time (sec) 6 7 8 9 10

Fig. 10. Step response of the nominal generator model for step
change of Vref: 1 – without PSS; 2 – with PSS2A (ks1 = 9 p.u.) ACKNOWLEDGEMENT
From: Vref

0.8
1 The authors would like to express their appreciation of eng.
Venci Zahov (head of division “System analyses & Transmis-
To: dU

0.6 1

sion Planning” in National Dispatching Center by the Bulgari-

0.4
0.2
0
1.5 2 an Electricity System Operator) for his generous and active
1
collaboration.
Amplitude

0.5
To: dP

0
-0.5 2
-1
1
0.01
REFERENCES
0
2
-0.01
To: dw

-0.02 [1] Rogers Gr., Power System Oscillations, Springer, 1999,

-0.03
ISBN: 978-0792377122
-0.04 1
0 0.5 1 1.5 2 Time2.5
(sec) 3 3.5 4 4.5 5 [2] Gerasimov K., Mathematical modeling of electromechanical
Fig. 11. Step response of the generator, modeled with uncertainties, transients in electric power systems and algorithms for analysis
for step change of Vref: of stability at small disturbances, D.Sc. Dissertation, 2006,
1 – without PSS; 2 – with PSS2A (ks1 = 9 p.u.) TU-Varna (in Bulgarian)
[3] Gerasimov Kr., Y. Rangelov, K. Gerasimov, Y. Kamenov.
The uncertain model used for robust stability check consists Methodology for tuning PSS2A system stabilizers, International
of output unstructured multiplicative uncertainty of 10% in Journal of Reasoning-based Intelligent Systems (IJRIS),
the EPS model and structured uncertainty in the generating ISSN (Online): 1755-0564, InderScience Publishers
unit state space realization describing change of the active [4] Murdoch, A., Boukarim, G. Perfomance criteria and tuning
power in its whole allowable range. In Figure 12 is shown that techniques, IEEE Tutorial Course – Power System Stabilization
a sufficient robust stability margin of 137 % is achieved. via Excitation Control, June 2007, pp. 26-34,
ISBN: 978-1424462308
0.73 [5] Kamenov, Y., K. Gerasimov, Y. Rangelov. Structuring the
Nominal Mathematical Model of the Electric Power System for
the Aims of Robust Analysis, Proc. of ICEST 2011, Niš, Serbia,
0.72

0.71 June 29 - July 1, 2011, Volume 2, pp.481-484,

ISBN: 978-8661250323
0.7
[6] Gerasimov K., Frequency method for analysis of small signal
0.69
stability of large electric power system, Ph.D. Dissertation,


1983, TU-Sofia (in Bulgarian)

0.68 [7] Notov P., K. Gerasimov, Transient processes in electric power
systems, TU-Sofia, 1997, ISBN: 9544382054
0.67
[8] Petkov P., M. Konstantinov, Robust Control Systems, ABC
0.66 Tehnika, 2002. (in Bulgarian) ISBN: 9548873516
[9] Petkov P., G. Lehov, A. Markovski, Handbook on Robust Con-
0.65
0 5 10
Frequency, [Hz]
15 20 25 trol Systems, ABC Tehnika, 2006. (in Bulgarian)
ISBN: 978-9548873772
Fig. 12. Upper bound on the mixed structural singular value μ of the [10] Gerasimov K., Y. Kamenov, Software tool for robustness ana-
generator with PSS2A with ks1 = 9 p.u. lisys of PSS in electric power plants, TU-Varna Yearbook,
2011. (in Bulgarian) ISSN: 1311-896X

242
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Technical Conditions for PV Plants Connection on the

MV Distribution Grids in the Republic of Macedonia
Ljupco Trpezanovski1, Metodija Atanasovski2 and Dimitar Dimitrov3
Abstract – In this paper the technical conditions, which have to II. CONDITIONS FOR THE ALLOWED PV PLANT
be fulfilled for connection of a Photovoltaic Power Plant (PV
plant) on distribution MV grid in the Power System (PS) of the
INSTALED POWER
Republic of Macedonia, are given. As the base, the technical
conditions from the aspect for obtaining a permission from the The installed power of the PV plant has impact to the
distribution system operator for connection of a PV plant, the voltages in the distribution grid during the plant switching-on
following conditions are described: nodes voltages changing, or switching-off transient period. This voltage impact should
flickers appearances, increasing of higher-harmonics currents, not exceed value of ∆um.=.2% at the plant connection point to
increasing of short currents intensity in the grid, changing of
feeder load segment distribution and location of energy-meter.
the distribution grid. The small PV plant may connect to the
The approach and results for checking the technical conditions distribution grid according to the criterion of allowed installed
which have to be fulfilled for a real case 1 MW PV plant power, only if the condition given by Eq. 1 is satisfied:
connection to a 10 kV distribution grid are presented.
S 3 pc
S nPV ≤ (1)
Keywords – Renewable energy sources, Photovoltaic system, 50 ⋅ k
Photovoltaic plant connection, Dispersed generation.
where:
− S nPV is rated installed apparent power of the plant in MVA,
I. INTRODUCTION − S3 pc is three phase short-circuit power in MVA and
− k is coefficient, which has value 1 for inverters DC/AC.
According to the Base Study on Renewable Energy In the MV distribution grids in the power system of the
Resources in the Republic of Macedonia [1] and Strategy for Republic of Macedonia, there are three standard values for
power plants development in the period 2008-2020 in the maximal permitted three phase short-circuit currents (powers):
Republic of Macedonia [2], it is planning to construct 10 – 30 − for 10 kV grid, current of 14.5 kA and power of 250 MVA,
MW PV Power plants till 2020, with total electricity − for 20 kV grid, current of 14.5 kA and power of 500 MVA,
production of 16 – 60 GWh/year. There are two categories of − for 35 kV grid, current of 12 kA and power of 750 MVA.
PV plants: 1. with installed power ≤50 kW and feed-in tariff It is worth to mention, that no matter how big is installed
30 €c/kWh and 2. with installed power between 50 kW and 1 power of the PV plant, the maximum voltage deviation in the
MW and feed-in tariff 26 €c/kWh. At the moment, the total connection point in steady-state conditions, should not exceed
allowed capacities which the Government would support by the interval ∆um.=.±5% from the MV grid rated voltage [3].
feed-in tariffs are: 2 MW for the first and 8 MW for second
category. Recently, the Energy regulatory commission of the
Republic of Macedonia issued a register of connected PV III. CONDITION FOR FLICKERS GENERATING
plants [6]. Till now, six PV plants from the first (total capacity
of 220.765 kW) and two from the second (total capacity of The flicker criteria can be assessed by the disturbance
1246.7 kW) category have been put into operation. factor Afd for the PV plants with flicker duration more than
According to the technical recommendation [3], [4] and two hours. This factor can be calculated with Eq. 2:
Distribution grid code [5], there are several technical 3
conditions which have to be fulfilled in order to connect a PV  S 
A fd =  c f ⋅ nPV  (2)
plant on distribution grid. In the following sections the  S 3 pc 
 
technical conditions which have to be fulfilled by the PV
plant, necessary for obtaining permission for connection to the where:
MV grid of the distribution company EVN, are explained. − cf is flicker coefficient ( S nPV , S3 pc .are defined in section II).
1
With this coefficient is assigned the ability of the PV plant to
Ljupco Trpezanovski is with the Faculty of Technical Sciences at produce flickers. After PV plant connection to the distribution
University St. Clement Ohridski – Bitola, I. L. Ribar bb. 7000 Bitola,
Republic of Macedonia, E-mail: ljupco.trpezanovski@uklo.edu.mk.
grid, the flicker coefficient should be measured in real
2 operating steady-state conditions. The value of this coefficient
Metodija Atanasovski is with the Faculty of Technical Sciences
at University St. Clement Ohridski – Bitola, I. L. Ribar bb. 7000
usually is cf >20, but it can reach a value up to 40. For every
Bitola, Republic of Macedonia. PV plant there should be issued certificate that condition for
3
Dimitar Dimitrov is with the Faculty of Electrical engineering long time duration flickers is fulfilled. Calculated value of the
and Information Technologies, University St. Cyril and Methodius – disturbance factor with Eq. 2, should be Afd ≤0.1 as a proof
Skopje Karpos II, 1000 Skopje, Republic of Macedonia. that the PV plant would not generate flickers [3].

243
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

IV. CONDITION FOR HIGHER-HARMONICS The automation installed in the PV plant will allow
CURRENTS switching on the plant to the grid when all three phases in the
grid have normal operating voltages. The conditions for
The criterion of permitted content of higher-harmonics synchronized connection are voltage difference ∆UPV.<±0.1Un,
currents can be checked by applying Eq. 3 [3]: ∆.fPV.<.±0.5Hz and phase angle difference ∆ϕ.<±10% [3], [4].

I hhp = I hhr ⋅ S 3 pc (3)

VI. CONDITION FOR ELECTRICITY MEASURING
where:
Despite that in the Grid Code [5] is not clarified, the
− Ihhp is the permitted value of the higher-harmonic current distribution company obliges the dispersed producers, the
on the generator voltage level, in A, measuring of produced and consumed electrical energy should
− Ihhr is the value of the higher-harmonic current in A/MVA be performed on one indirectly measuring place on 10 (20) kV
(reduced on the S3 pc in the connection point on the grid). side, with measuring transformers placed in separate
The values of maximum permitted content of higher- measuring cabinet [8].
harmonics currents reduced on the three phase short-circuit The current transformers (CT) in each phase, should have
power in the PV plant connection point on the grid, are given two cores X/5/5 A, 50Hz, 10/25 VA, class 0.5/5P10 (FS<5 for
in Table 1. the first core and FS>10 for the second core). The voltage
transformers (VT) in each phase should have the following
TABLE I characteristics: 10000/ 3 (20000/ 3 )/100/ 3 /100/ 3 V/V,
MAXIMUM PERMITED HIGHER-HARMONICS CURRENTS
50 Hz, 25/25 VA, class 0.5/3P.
harmonic ν 5 7 11 13 17 19 23 25 Electricity meter have to be with two-way measuring
0.7 0.6 0.5 0.3 0.3 0.2 0.2 0.2 system of produced and consumed electrical energy. The type
Ihhr A/MVA
of this meter must be 5(6) A, 3x(100/ 3 )/100 V, class 1/2
If the current of any higher-harmonic ν exceeds the (A/R) with optical port, internal connecting clock, CS/RS
permitted value, the voltage for the ν-th harmonic Uhν should communication and connected with modem to the system of
be calculated. If the voltage of the 5-th higher-harmonic is distance reading in the company EVN Macedonia. No other
Uh5 ≤0.2%⋅Un and for the rest higher-harmonics from Table 1 additional equipment is allowed to be connected on the
Uhν ≤0.1%⋅Un (where Un is rated voltage of the grid), than the secondary windings of the measuring transformers.
criterion of permitted higher-harmonics currents is fulfilled. The measuring transformers and the electric energy meter
On the contrary the owner of the PV plant should undertake are provided by the distribution operator - company EVN
measures to eliminate higher-harmonics currents [3]. Macedonia and stay in their property.
Another way to check the condition for higher-harmonic
currents is to compare total harmonic distortion factor THD VII. CASE STUDY: 1 MW PV PLANT CONNECTION
allowed by the distribution company, with the certified THD
factor of the DC/AC inverter [8]. In this paper a case study for connection of 1 MW PV plant
on 10 kV distribution feeder named “Egri”, from substation
V. CONDITION FOR SHORT-CIRCUIT POWER TS 35/10 kV/kV Bukri, is analyzed. The schematic diagram of
the PV plant main components is shown on the Fig. 1.
If the connection of the PV plant causes increasing of the
PV modules N PV,p PV generator
three phase short-circuit power (current) over the allowed
values for the equipment, the following measures should be
undertaken:
- to limit the short-circuit currents in the PV plant, Inverter
+ 10 kV
- to replace the switching and other equipment which not meet N PV,s –
requirements for short-circuit currents, – ~ 10 kV
grid
- to connect the PV plant in other place of the grid.
Because the PV plants with installed power less or equal to
1 MVA don’t influence significantly on short-circuit power
(current) increasing, it is not necessary to check this criterion
for PV plants constructed in the Republic of Macedonia. Fig. 1. Schematic diagram of the PV plant main components.
The relay protection of the MV feeder will trip the feeder
circuit-breaker in cases of short-circuit faults in the feeder. In Total number of the modules will be 4320 each per 230 Wp.
these cases is not allowed island operation of the PV plant (or The nominal power of the DC generator of the PV plant will
PV plants if more then one are connected on the same feeder). be 993.6 kWp. Because the power factor for this type of plants
The relay protection of the PV plant should switch of the is cosϕ.≈1.0 and modules power is with positive tolerance in
plant from the grid immediately. the following calculations can be taken that S nPV ≈ 1 MVA.

244
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The DC/AC inverter which is planed to be used is Sunny B. Check for the flickers generating
Central HE 1000MV for direct medium-voltage feed-in,
because the MV/LV transformer is build in the same house. As it was mentioned in section III., the flicker coefficient cf
Distribution feeder named “Egri” with all its elements as a should be measured in real operating steady-state conditions.
part from 10 kV grid is shown on Fig. 2. However, if the maximum possible value of this coefficient
All data for the elements connected in and between the cf =40 is taken into Eq. 2 and S3 pc =250 MVA, S nPV =1MVA,
nodes are given on the same Fig. 2. A new PV plant will be the calculated value of disturbance factor Afd.=0.0041.
connected in the node J4. In this case Afd.<<0.1, so it can be concluded that the new

Fig. 2. 10 kV feeder named “Egri” with data for the connected elements and connection point of 1 MW PV plant.

A Load-flow and short-circuit analysis for the 10 kV feeder PV plant will not generate flickers with duration more than
shown on Fig. 2 are performed with Neplan 5.0 software [9]. two hours. After construction of the PV plant the certificate
Two cases are analyzed. In the first case the new PV plant is for measured cf should be issued.
not connected and in the second case this plant is connected in
node J4. Also the calculations for checking the technical D. Check for the higher-harmonics currents
conditions which have to be fulfilled to connect this PV Plant
of 1 MW, are done. Under the normal operating conditions the total harmonic
distortion factor THD allowed by distribution company EVN
A. Check for the allowed PV plant installed power Macedonia is 8% [8]. According to the inverter manufacturer
data, this factor has value THD<3%. Taking into account
According to Eq. 1 and taking into account the values for these data, it is obvious that this condition is fulfilled and the
three phase short-circuit power (250 MVA for 10 kV grid) generated higher-harmonic currents have acceptable values.
and k.=1, the maximal allowed installed power for the PV
plant connected in node J4 can be S nPV ≤ 5 MVA. Because the E. Check for the short circuit power (current)
installed power of new PV plant is 1 MVA (actually 1 MW
because PV plants work with cosϕ.=1.0) it is obvious that The DC generator of the PV plant works as constant current
condition for maximal allowed installed power is fulfilled. source. For the planned PV modules with power of 230 Wp
The load-flow analysis confirmed that PV plant connection maximum power point current is IMPP.=7.8 A and maximum
wouldn’t change the node voltages and line currents over short circuit current is Isc.=8.3 A. Increasing of the short circuit
permitted values. In this case, the voltage drops and energy current is only 6.4% over the IMPP.. Because of the presence of
losses in the lines are smaller than in the first case. Also the the PV plant in the MV grid, the total increasing of the three
power losses in the entire grid are reduced [7]. phase short circuit current and power at the point of

245
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

connection will be also 6.4%. This is not significant increase REFERENCES

which can take negative influence on the electrical equipment.
[1] Team of ICEIM-MANU. Base strategy for use of renewable
energy sources in R. Macedonia until 2020, Government of R.
VIII. CONCLUSION Macedonia, 2010.
[2] Team of ICEIM-MANU. Strategy for energy sector
In the last few years, the construction and connection of the development in R. Macedonia until 2030, Government of R.
PV plants in MV grids in the PS of the Republic of Macedonia, 2010.
Macedonia are occurred. The technical regulations for PV [3] Group of Authors, Technical Recommendation 16. Base
plants were very poor and lot of these documents were conditions for connection of small power plants on the
directly accepted from the foreign regulations which distribution grid of Serbia (adopted by EVN Macedonia), JP
sometimes do not coincide with the PS situation and the Elektroprivreda Srbije, 2003.
[4] Group of Authors, Technical Recommendation for application
present legislation. Distribution operator EVN Macedonia
of valid lows and rules for connection the power plants from
introduced additional regulations in this domain. renewable energy on distribution grid, HEPODS doo,
This paper deals with necessary technical conditions which distribution system operator of Croatia, 2010.
have to be fulfilled for PV plant connection on the MV grid. [5] Management committee of EVN Macedonia, Grid Code for
The conditions as: the allowed PV plant installed power, the distribution of electrical energy, Official gazette of the Republic
flickers generating, the increasing of the higher-harmonics of Macedonia, no. 83, 2008.
currents, the increasing of the short circuit currents (power) [6] Energy regulatory commission of the Republic of Macedonia,
intensity in the grid, the location and the type of energy-meter www.erc.org.mk
are described. These technical conditions are checked for a [7] Lj. Trpezanovski, M. Atanasovski, Analyze of 10 kV distribution
feeder “Egri” from TS Bukri, before and after connection of PV
real PV plant of 1 MW which is planned to be connected in
plant of 1 MW from investor Torpedo-Mobil Bitola, Faculty of
the 10 kV distribution grid in the middle of 2012. All Technical Sciences Bitola, 2010.
necessary and proper parameters are calculated and compared [8] EVN Macedonia, Decision for 1 MW PV plant “Torpedo-Solar
with allowed values given by the grid code. GT” connection on the distribution grid, EVN Macedonia, AD
From the calculations and values of obtained results, it can Skopje, 2010.
be concluded that connection of PV plants with power equal [9] Neplan 5.0 software, BCP Switzerland (Educational version).
or less than 1 MW will not have negative influence to the
normal operation conditions of the MV distribution grid and
the appertaining electrical equipment.

246
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Control cards. Control cards and control points as part

of the manufacturing process.
Violina Georgieva1 and Alexander Hadjidimitrov2
Abstract – In process manufacturing the control over the
process usually is more important than the control over the Control cards are used in production departments (products
product itself. In production departments the registration of the and PCBs.)
operations and the quality control is done using “Control cards” The manufacturing process consists of two stages –
The control card is used afterwards in the analyses of defects,
production of semi-finished products (PCBs) and production
claims, the load of machines and employees, etc.
of products.
Keywords – Manufacturing, Production process, Control card,
Control point, Integrated system. The production of PCBs consists of:
 Delivery of the components needed
 Print of the blank PCB
I.INTRODUCTION  SMD assembly
 Visual control
The current paper describes the usage of control cards as a  Conventional assembly
tool for quality control of the production process. The authors  Visual control
share their experience in the creation of a software module for  Control card filling
control cards management as a part of an enterprise resource  Transfer of the produced PCBs to the manufacturing
planning (ERP) system. This module is integrated to the department
existing manufacturing module of the ERP system. The
system is implemented at a manufacturing enterprise that The production of the products consists of:
produces electronic equipment (cash registers, fiscal printers,  Preliminary operations
scales, etc.). Before the creation of the module the control  Test of PCBs, recording of firmware
cards were created manually on paper. The implementation of  Assembly of the components of the product
the module automated the process of Control Cards  Assembly of the final product
management and increased the control over it. The paper  Packaging
describes the basic concepts of control cards, the main issues
that are met during their creation, the structure and the The registration of the planning and production processes in
concepts of the module created. The usage is illustrated with the system consists of the following stages:
examples from the enterprise, where the system is  Sales order – it is created every time when a new
implemented. order from a customer is received.
 Main production schedule – this is the planning
II. COMPANY STRUCTURE stage. It is based on the sales orders received.
Usually it is prepared once a week. As a result new
production orders for the products needed are
The main departments in the company are:
created.
 Trade (international and domestic) – manages the
 Start of the production orders – each order is started
sales of production and the purchase of materials and
when the production of the product in the
goods.
manufacturing department is started.
 Manufacturing of products
 Registration of the production – at this stage are
 Manufacturing of Printed circuit boards (PCBs) –
filled up the serial numbers and lot numbers. The
Semi-finished products
control cards are also filled.
 Software – develop software and firmware for the
 Consumption of materials – the system proposes the
products produced
materials to be consumed based on the “Bill of
 Development – develop new products and
materials” of the product. Manual adjustment of the
modifications of the existing products
quantities proposed is allowed.
 Accounting
 Finalization of the production order.
 Management

1
Violina Georgieva – PhD. Student, Technical University Sofia,
III. APPLICATION OF CONTROL CARDS
Faculty of Automatics. Sofia 1000, 8 “Sv. Kliment Ohridski” Blvd.,
violina_jg@abv.bg. The quality control of the products is based on the control
2
Alexander Hadjidimitrov – ERP consultant at Team VISION of the production process itself. The documentation of the
Bulgaria Ltd. Sofia 1712, 41 “Alexander Malinov” Blvd., process is done using control cards. Thus is tracked that all
Alexander@Hadjidimitrov.com.

247
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

operations needed are done and also is tracked who are the
employees that have done the operations. This way two This control card template becomes a part of the product
important targets are achieved: specification. It is an integral part of the production
 Following the sequence of operations described in documentation. It is used in the preparation of the current
the control card is guaranteed that no operations are control cards of the products during the production process.
skipped; Current control cards are filled up during the production
 In case of defective product is easy to find which process. Based on the production process organization
operation has caused the defect and to trace who is different approaches are possible:
the employee who has done the operation and when
is the operation done. Thus the reasons of the defect  A separate control card is filled in for each product
could be analyzed and actions to be undertaken in (for each instance).
order to change the process to avoid future problems  A separate control card is filled in for each lot of
of the same type. products (e.g. for the products produced by a single
work shift).
The most important thing during the creation of a control  A single control card is filled in for the total quantity
card is to define a proper list of control points. Each control produced.
point describes one operation or a group of operations that are
done in conjunction during the production process. Defining In regard of the selected approach some modifications in
the list of operations should keep in mind that operations the control cards are possible. In general it contains the
should be grouped in control points in a way that no operation following fields:
is skipped but also the list of control points should be short
enough in order to be easy to be documented and followed.  ID and name of the product
It’s advisable to group related operations that are done  ID of the production order
together by the same employee. As an example if during the  Production date
assembly of a product 5 Integrated Circuits (ICs) should be  Serial number / Lot number (or a list of numbers)
mounted in the sockets on a PCB by the same employee these  A list of control points
could be grouped to one control point (“IC mounting) as  For each control point – a flag that the operation is
during this operation the employee receives a PCB with the done and the ID of the employee (or a list of
sockets mounted, mounts the ICs in the sockets and employee IDs) who has done the operation. If the
afterwards transfers the ready PCB to the next stage. In other control card is for a production period longer than
case if the employee should mount only three of the ICs, a test one day – also a production date should be added to
should be carried out and afterwards the two other ICs should this list [3].
be mounted then maybe it would be better to separate the
process in 3 control points (“Mounting of the first group of
ICs”, “Test”, “Mounting of the second group of ICs”). This
IV. IMPLEMENTATION OF THE CONTROL CARDS IN
will help the process to be tracked properly in order to allow THE ERP SYSTEM
future analyses in case of defects or claims.
In order to be possible to use control cards in the production Depending on the enterprise and the software used for
process is required to prepare templates of the control cards production planning and management different approaches in
for each product (semi-finished or final) that is tracked the implementation of control cards are possible. In the case
separately (e.g. for each PCB, each cash register, etc.) Each described in the current paper an ERP system is used. The
template contains a list of control points listed chronologically functionality for control card management is implemented as
during the production process. If some operation is repeated at an additional development especially for the specific needs of
different stages during the process (e.g. testing of different the enterprise. It is integrated to the standard functionality of
modules) then it is documented as several control points at the the system for management of items (products) and
appropriate positions (one control point for each occurrence) production orders.
in the list. The control points are developed as a separate table. It
A sample template of a control card could look like: contains the full list of operations that are tracked separately.
These control points are used afterwards to assemble the
 Main board test control cards of all products. The table contains the following
 Recording of firmware fields:
 Operator display test
 Fiscal module test  ID – a unique ID of the point in the system
 Top cover assembly  Description a description of the tracked operation
 Bottom cover assembly
 Display assembly The product control cards are developed as a separate table
 Final product test related to the Item table (that contains the list of items in the
 Final product assembly system). This way for each item is specified the list of the
 Packaging control points that should be tracked. The table contains the
list of the control points that form the control card of the item.

248
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The control points are selected from the global list of control
points in the system (the table described above.) This table
contains the following fields:

 Item ID – contains the unique ID of the item. It is

related to the Item table of the system. Usually for a
single item a several lines in the table with the same
item ID are created – each line contains one control
point from the item control card.
 Control point – contains the ID of the control point.
Fig. 1 Product control card template
The field is related to the table with the global list of In case of definition of new products a new control card
control points. template is created (see Fig. 1).
This functionality is used for management of control cards During the creation of new production orders for each line
during the production process. As the manufacturing that contains a product the system automatically creates a
functionality itself is based on production order the blank control card based on the template set for that product
corresponding control cards are developed as documents (see Fig. 2).
related to the lines of the production orders. The production
order itself is a document containing one or several lines.
Each line describes the production of a single item. For each
item are stored its ID and the quantity for production.
The functionality consists of the following tables:

 Production employee – contains the list of employees

in the production departments. Contains the
following fields:
o ID – a unique ID of the employee in the
system
o Name – the name of the employee
 Production order line control card – contains a list of Fig. 2 Production order control card
the control points for the item in the production order
line. Contains the following fields:
o Production order ID – the ID of the When an operation is completed the employee in charge
production order for which the control card sets the “Approved” flag for the corresponding control point
is prepared; and fills the list of employees who have performed the
o Production order line ID – the ID of the line operations.
in the production order for which the control The control card could be printed on paper at any moment
card is prepared; if needed.
Control point ID – ID of the control point in
the control card;
V. BENEFITS FROM CONTROL CARD USAGE
o Approved – a flag that means that the
operations described in the control point are
The properly prepared control card template contains all
done successfully;
operations that should be performed during the production of
 Performed By – this table is related to the
an item. This guarantees that during the production process no
“Production order line control card” table. For each
operation will be skipped (because otherwise the
control point here are listed the employees who have
corresponding control point will not be filled in and the
performed the operations. The table contains the
control card will be incomplete. This should be noticed when
following fields:
the production order is finished.) The ERP system described
o Production order ID
in this paper checks the control cards during the production
o Production order line ID
order closure and does not allow the order to be finished if
o Control point ID
there are control points that are not marked as “Approved”.
o Employee ID – ID of the employee who has
The second benefit from control cards usage is the
performed the operation
possibility for future analyses and control. They are performed
in cases of defective production when is needed to find the
During the initial implementation of the system the global
possible reasons for the defects. Each produced item has a lot
list of control points is set. Periodically in cases of need of
number or a serial number. Based on this number the system
new operations the list is updated with the new control points.
can identify the production order the item was produced by. In
Also periodically when new employees are employed the
the production order control card could be identified the
list of production employees is updated.
employees who have performed the operations. This allows to
trace the production process for the specific item and to give

249
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

hints for the possible reasons for the defect [3]. That’s why many errors and misusage could occur. For this
reason it is very important during the implementation of this
tool to guarantee the involvement of the team who is supposed
VI. RISKS to fill and use them in order to guarantee that the information
filled is correct. Otherwise they could not pay enough
The successful usage of control cards could be attention to it and to feel it like an additional useless time-
compromised if some risks are neglected. consuming documentation that should be filled. This also will
The first risk is the improper definition of the control compromise the process.
points. As the control card is a list of the operations that
should be tracked and controlled, the proper definition of this
list is the foundation for the successful usage of control cards. VII. CONCLUSION
Possible problems include:
 Skipping important operations – in case a control Control cards are a tool that allows guaranteeing of the
point is not created for an important operation than proper flow of the production process and as a result to
that operation will not be tracked by the control card. guarantee the quality in the terms of not skipping operations
As a result it could be skipped during the production due to errors. For the proper usage of the tool however it is
and also it will not be documented and future control very important the proper implementation in the organization
and analyses will be impossible. to be done and to guarantee the involvement of the team who
 A very detail list of the operations – in case the is supposed to use them.
operations are not grouped in a proper way to control
points a very long list of control points might be
produced. This could make the process of filling
REFERENCES
control cards very tough and the employees to stop to
[1] G. Tsvetkov, “Production management”, Sofia, 2006,
pay attention to it. As a results they could mark
“Softrade”.
operations as “Approved” without checking their [2] I. Dakov, “Production engineering”, Sofia, 2003, “Luren”
actual status and thus to compromise the process. publishing house - “Luren commerce” Ltd.
[3] E. Andronov, M. Aleksandrova, “Operations management”,
The second big risk is the human factor. In general the „Stopanstvo”, Sofia, 2005.
control cards are created by people and are used to document
the work of people.

250
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Computer Simulation and Analysis of Two-Coordinate

Position Electric Drive Systems
Mikho Mikhov1 and Marin Zhilevski2
Abstract – Analysis of some algorithms for position control of V pr2 – position reference signals; Vsr1 and Vsr2 – speed
two-coordinate electric drive systems is presented in this paper.
Models for computer simulation with various types of motors reference signals; Vcr1 and Vcr2 – current reference signals;
have been developed. Detailed studies by means of computer
V pf1 and V pf 2 – position feedback signals; Vsf1 and Vsf 2 –
simulation and experimental research have been carried out. The
results obtained can be used in the design and tuning of such speed feedback signals; Vcf1 and Vcf 2 – current feedback
types of drive systems with position control.
signals; θ1 and θ 2 – angular positions; S1 and S 2 – linear
Keywords – Two-coordinate electric drive, Position control, displacements.
Computer simulation. The set of achievements required for the drive system can
be formulated as follows:
- forming the necessary motion trajectories at given position
I. INTRODUCTION cycles;
- maximum starting torque to ensure good dynamics;
Two-coordinate electric drive systems are widely used in - reversible speed and torque control;
many industrial applications. - compensation of the disturbances.
Generally, motions in these systems are formed by the
respective trajectories along both coordinate axes. Control III. MODELLING OF THE DRIVE SYSTEM
algorithms affect the performance, productivity and energy
consumption [2], [3], [4]. The vector-matrix model of the DC motor drive under
Time shortening of the transient regimes at positioning is consideration is as follows:
essential for mass production of parts, because it increases the
respective machine effectiveness. ⎡ ⎤ ⎡ 0 ⎤
Mathematical modeling and computer simulation provide ⎡ dθ i ⎤ ⎢0 1 0 ⎥ ⎡θi ⎤ ⎡ 0 ⎤ ⎢ ⎥
⎢ ⎥
very good opportunities to explore different control ⎢ dt ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ dω ⎥ ⎢ ⎢ ⎥ ⎢− 1 ⎥
algorithms aiming at optimizing of motion trajectories [2], [3], K ti ⎥ ⎢ωi ⎥ ⎢ 0 ⎥
⎢ i ⎥ = ⎢0 0 ⎥ +⎢ ⎥ vi + ⎢⎢ J Σ i ⎥il , (1)
[5]. ⎢ dt ⎥ ⎢ JΣi ⎥ ⎢ ⎥ ⎢ ⎥ ⎥ i
With respect to modernization of a class of machine tools ⎢ diai ⎥ ⎢ ⎢ ⎥ ⎢ ⎥
K ei Ra ⎥ ⎢iai ⎥ ⎢ K ci ⎥ ⎢
some two-coordinate electric drives have been analyzed, ⎢ ⎥ ⎢ − i ⎥⎢ ⎥ ⎢L 0 ⎥
⎣ dt ⎦ ⎢0 − L Lai ⎥⎦ ⎣ ⎦ ⎣ a i
⎥ ⎢ ⎥
allowing the choice of the appropriate once meeting the ⎣ ai ⎦ ⎣ ⎦
required performance. Models of such drives have been
developed used for studying of different position control where: θ i is angular position; ωi – motor speed; iai -
algorithms for the respective dynamic and static regimes at
different operation modes. armature current; K ei - back EMF voltage coefficient; K t i -
torque coefficient; Ra i - armature circuit resistance; Lai -
II. FEATURES OF THE DRIVE SYSTEM
armature inductance; K ci - amplifier gain of the chopper; vi
The simplified block diagram of the system under - input control signal of the power converter; J Σ i - total
consideration is represented in Fig. 1, where the notations are
as follows: CP – control panel; CD – control device; PC1 and inertia referred to the motor shaft; ili - armature current
РC2 – position controllers; SC1 and SC2 – speed controllers; which is determined by the respective load torque; i = 1,2 –
CC1 and CC2 – current controllers; C1 and С2 – power number of the coordinate axes.
convertors; М1 and М2 – DC motors; SS1 and SS2 – speed Both subsystems have identical cascade structures with
sensors; DM1 and DM2 – driven mechanisms; CF1 and CF2 – subordinate regulation of currents, speeds and positions.
current feedback blocks; SF1 and SF2 – speed feedback Control loops optimization and tuning of the respective
blocks; PF1 and PF2 – position feedback blocks; V pr1 and controllers have been done sequentially, starting from the
innermost ones [1].
1
Mikho Mikhov is with the Faculty of Automatics at Technical For the used mechanical gear maximum operating speed of
University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, Bulgaria, E- the motor for the respective coordinate axis is chosen to be
mail: mikhov@tu-sofia.bg. equal to the nominal value:
2
Marin Zhilevski is with the Faculty of Automatics at Technical
University of Sofia, 8 Kl. Ohridski Blvd, Sofia 1000, Bulgaria, E- ω maxi ≤ ω nom i . (2)
mail: electric_zhilevski@abv.bg.

251
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 1. Block diagram of the two-coordinate drive system.

The maximum rate of speed change in the respective axis algorithms are presented in Fig. 2, where the symbols used
can be determined from the following equation: are: О (0, 0) – initial position; А ( S1 f , S 2 f ) – final position

ε max i = М maxi J Σ i , (3) of the specified cycle.

Fig. 2a shows a trajectory obtained by successive
movement along the coordinate axes. The total time for
where М max i is the maximum torque, which the respective positioning is as follows:
motor can develop along this coordinate axis; J Σ i – total
t p = t p1 + t p2 , (9)
inertia referred to the motor shaft.
For the deceleration motion in this case the following
where: t p1 is the motion time along the coordinate axis x ;
relationship is valid:
t p2 – the motion time along axis y .
2
Δθ d maxi = ω nom 2ε d maxi . (4)
i

Because subordinate regulation of coordinates is applied,

the output voltage of the respective position controller is the
assigning speed signal.

( ) ( )
U sri = K рci V pri − V pfi = K рci K pfi θ ri − θ i = K sfi ω ri . (5)

Equation (4) for Δθ d maxi and ω nomi takes the following a) b)

form:
Fig. 2. Motion trajectories for
K рci K pfi Δθ d maxi = K sfi ω nomi . (6) two-coordinate position control:
a) Consecutive motion along the
coordinate axes;
After substituting (4) to (6) and solving with respect to the b) Simultaneous motion along
position controller coefficient, the equation becomes: the coordinate axes;
c) Combined motion along the
K sfi ω nomi 2 K sfi ε d maxi c) coordinate axes.
K рci = =
( )
. (7)
2
K pfi ω nom 2ε d maxi K pf ω nomi
i
i Fig 2b represents a trajectory obtained by simultaneous
movement along both coordinate axes. In such way of control
For the corresponding mechanical gear the linear speed and position time is:
linear position can be determined as follows:
t p = t p1 = t p2 . (10)
Vi = ωi / K gi ; S i = θ i / K gi , (8)
Fig. 2c shows a trajectory obtained at combined motion
where K gi is the respective gear coefficient. along the coordinate axes. If both drives work at the same
In general, when two-coordinate systems with position speeds, the total time of positioning is equal to the time
control are used, the motion trajectories are formed by the necessary for the drive with longer displacement time set.
respective displacements of both axes.
Motion trajectories for the studied position control t p = t p1 . (11)

252
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 3. Simulation model of the two-coordinate drive system.

In the MATLAB/SIMULINK environment some models of coordinates.

systems for two-coordinate electric drives have been
developed with different types of motors. They allow for
detailed studies of the respective static and dynamic regimes
and analyses of performance. A simplified block diagram of
one of the models is presented in Fig. 3.

IV. EXPERIMENTAL RESULTS AND ANALYSIS

The electric drives for both coordinate axes are identical.
The DC motors used for modeling and computer simulation
have the following parameters:
Vnomi = 30 V ; I anom = 15.7 A ; ω nomi = 115.19 rad/s .
i
Fig. 4 shows the time-diagrams obtained by computer
simulation during processing a position cycle along the x
coordinate. The set displacement S1r , the load torque Tl1 , the
armature current I a1 , the displacement speed V1 and the
linear position S1 along this coordinate axis are shown.
Armature current is limited to the maximum admissible value
I a1 max . The respective control loops setting provide compen-
sation of the disturbances and smooth positioning without
overshooting. The disturbances applied sequentially on the
electric drive are ΔTl1 = ± 0.25Tl1nom
Fig. 4. Time-diagrams for a set position cycle along one axis.
In Fig. 5 the trajectories of movement to the same final
position have been shown, obtained through different Fig 5b shows the trajectory when simultaneous motion is
algorithms of movement on both axes. performed along both coordinate axes.
Fig. 5a represents a consecutive motion along the x and y

253
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 5в illustrates combined motion along the coordinate coordinate axis у is represented in Fig. 7. The set distance in
axes. this case is 0.32 m.

Fig. 7. Time-diagram for motion along the coordinate axis y.

а) Fig. 8 illustrates reverse control. The set distance in both

directions is 0.32 m.

b) Fig. 8. Time-diagram illustrating reverse control along the axis y.

The behavior analysis shows that the presented position

control algorithms provide for good performance suitable for
practical applications.

V. CONCLUSION
Models for computer simulation of two-coordinate electric
drive systems with various algorithms for position control
have been developed.
On the basis of computer simulations and experimental
c)
studies the presented algorithms for position control have
Fig. 5. Trajectories for different algorithms of control. been analyzed.
This research and the results obtained can be used in the
Detailed experimental studies have been carried out for design and tuning up of such two-coordinate systems of
different versions of controllers’ tuning and operation regimes. electric drives.
Some time-diagrams are presented in Fig. 6, 7 and 8.
Fig. 6 shows a linear speed trajectory, obtained REFERENCES
experimentally for displacements of 0.62 m along the
[1] M. Mikhov, Electric Drives Control Systems, Technical
coordinate axis x. University of Sofia, Sofia, 2009, ISBN 978-954-438-628-3.
[2] M. Mikhov, I. Tatarov, Analysis of a Two-Coordinate Driving
System Aiming at Performance Improvement. Bulletin of the
PGU, Vol. 61, No. 3, pp. 105-110, Ploiesti, Romania, 2009,
ISSN 1224-8495.
[3] M. Mikhov, G. Mitrinski, Analysis of Some Algorithms for
Positioning Control of Two-Coordinate Driving Systems.
Proceedings of the ICRDMI, Vol. 1, pp. 638-643, Donji
Milanovac, Serbia, 2010, ISBN 978-86-6075-017-6.
[4] G. Blažiunas, Accuracy Investigations of Multifunctional Two-
coordinate Drive System, Electronics and Electrical
Engineering, Vol. 51, No. 2, pp. 8-14, 2004, ISSN 1392 – 1215.
Fig. 6. Time-diagram of motion along the x coordinate.
[5] C. Ong, Dynamic simulation of electric machinery using
MATLAB/SIMULINK, New Jersey, Prentice Hall, 1998, ISBN
A linear speed diagram, obtained for motion along the 0-13-723785-5.

254
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Automated multichannel broadband spectrum

analysis of fiber-optic grating sensors
Plamen Balzhiev1, Wojtek Bock2, Tinko Eftimov3 and Rumen Arnaudov4
Abstract – In this paper we report methods for spectrum improvement in position accuracy and micro-stepping control
analysis and data processing algorithms. An automated multi- is applied.
channel spectrum measurement system is introduced with In this paper we report on the further development of a
controlled fiber-optic signal switching and spectra analysis with previously proposed spectrally and spatially multiplexed
linear CCD photodiode array, diffraction grating and precise
stepper motor. The designed system demands advanced
sensor network using an InGaAs CCD photodiode array and
measurement and data processing techniques. The paper reports opto-mechanic switches. We also present results on the
the implemented methods for automated multi-channel implemented methods for automated multi-channel
measurements, accuracy improvement, noise cancellation measurements, accuracy improvement and noise cancellation
techniques and fiber-optic grating sensor measurements techniques and fiber-optic grating sensor measurements..

Keywords – long-period grating sensors, fiber-optic sensor

interrogation and multi-channel spectrum measurement. II. MULTI-CHANNEL SPECTRUM
MEASUREMENT SYSTEM
I. INTRODUCTION
A. System description
Over the past decade optical fiber-based sensors are gaining
significant progress and popularity. Optical fiber gratings are The basic scheme of the multi-channel spectral
often classified as Fiber Bragg gratings (FBGs) and Long measurement system is shown in Fig. 1. The radiation of a
Period gratings (LPGs) [5]. Because of the large periodicity, C+L band amplified spontaneous emission (ASE) broadband
LPGs are usually easier to fabricate in mass production in source (Joinwit) is coupled to port 1 of a 3-port optical
comparison with FBGs. LPGs have found applications in circulator. Light reflected from the end of each channel is
various devices like equalizers for erbium doped fiber redirected from port 2 to port 3 and then to the diffraction-
amplifiers, band-rejection filters and sensors for strain and grating-based spectrometer and the CCD photodiode array
temperature [3]. Because of their great width, spectral detector [4].
multiplexing is limited; the absence of a reflected signal Refl.1
demands detection of center wavelength shifts in a noisy Refl.4
LPG1

minimum; and since resonance coupling in LPGs is to a

cladding mode, the fiber typically has to be stripped, which
LPG3

LPG4
creates challenges for long-term reliability and packaging.
LPG2

On the other hand, LPGs are sensitive to a number of ASE

physical quantities such as surrounding refractive index,
hydrostatic pressure, bending and twisting. They therefore
offer significant application opportunities. However, in order SW2 SW3
to reduce the effective price per sensor, simple and efficient
Fiber-optic
multiplexing systems must be developed. Wavelength and
Circulator 1 SW1
time-division multiplexing are well advanced with FBG
sensor networks [1,2], but comparatively little has been 2
reported on the multiplexing of LPGs [10]. Also a precise
stepper motor is introduced to extend measured spectrum 3
range and resolution by rotating the diffraction grating. Diffraction Precise
Precise stepping drives are presented in [12, 13], where Attenuator Grating Stepper
Motor
1
Plamen Balzhiev is with the Faculty of Telecommunications at
Technical University of Sofia, 8 Kl. Ohridski Blvd, Sofia, Bulgaria,
E-mail: baljiev@gmail.com
2
Wojtek Bock is with the Canada Research Chair in Photonics,
Université du Québec en Outaouais, Québec, Canada, E-mail:
Wojtek.Bock@uqo.ca
3
Tinko Eftimov is with the Faculty of Physics, Plovdiv University, CCD Detection Unit Control Unit
Bulgaria, E-mail: tinkoeftimov@yahoo.ca
4
Rumen Arnaudov is with the Faculty of Telecommunications,
Technical University of Sofia, Bulgaria, E-mail: ra@tu-sofia.bg Fig. 1. Multi-channel spectrum measurement system

255
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

At port 2 there is an arrangement of three electrically The LabView application automatically configures
controlled 1x2 fiber optic switches that allow an arbitrary diffraction grating angle position via the precise stepper motor
access to four sensing channels, which can accommodate up and the measured channel. After the current channel is
to four LPGs depending on their bandwidth and sensitivity to measured and visualized, the program automatically
a particular physical quantity [9, 14]. At the end of each configures the next channel for measurement.
channel there is a tunable reflector which returns light back to
the sensing channel and through port 3 the light is collimated C. Correlation analysis for accuracy improvements and noise
onto a 600 lines/mm diffraction grating so the spectrum is cancellation
observed by a CCD photodiode array.
The four measured fiber-optic channels are set in the Correlation analysis in two separate spatially shifted signals
following configuration – in Ch.1 two LPG sensors are is introduced to increase accuracy of spectral measurement
placed, Ch2 and Ch.3 investigate single LPG and Ch.4 is
and to reduce signal noises. The measured broadband signals
utilized to perform reference signal measurement and system from fiber-optic sensors are spectrally resolved on the linear
calibration with the ASE light source.
CCD photodiode array. To perform a spatial shift of the signal
a precise stepper motor is implemented. It rotates the
B. Detection and control devices diffraction grating with 0.1deg accuracy.

The detection unit is based on a 512-pixel InGaAs CCD R ( )   f1 (n) f 2 (n   )

(G9204-512D – Hamamatsu Photonics) linear array with an  (1)
integrated low-noise charge-amplifier featuring high R ( max )  max
sensitivity, a low dark current and high stability in the 800-
1750 nm spectral range Two high-speed capacitive-based With cross-correlation function (1) the exact signal shift is
analog-digital converters (ADCs) transform the analog data calculated and any difference in signals is analyzed [6, 15]. In
from the CCD sensor into 16-bit corresponding digital values. this way the multiple spatially shifted measurement of an
The obtained data from the CCD array is filtered and identical signal may result in increased spectral measurement
further transmitted via USB interface to a personal computer. resolution.
The interrogation system is operated and configured by an On Fig.3a two measurements with spatial shift are
application using a Lab-View programming environment [8] presented. The cross-correlation function (R(τ)) and exact
which allows an individual settings for the parameters - shift are calculated. The optimal signal match is achieved at
integration time (τ), sensor sensitivity (s), conversion speed, maximum of R(τ) and the exact shift (τ) is calculated.
data communication speed (r)., has start, stop and pause
functions to be manually configured. Dark current can be
50 1.0
subtracted after averaging, reference and the current signal x10^3
Rmax=0.9923 Signal 1
0.9
tau=18 Signal 2
can be read, and the signal-to-reference ratio will be presented 40 0.8
Cross-correlation function
in dB on the screen [11]. 0.7
30 0.6
Intensity, I

0.5
20 0.4
0.3
10 0.2
τ 0.1
0 0.0
0 50 100 150 200 250 300 350 400 450 500
А Pixel Number, n

50
x10^3 Filtered Signal

40
Intensity, I

30

20

Fig. 2. Photographs of CCD Detection Unit and control device 10

The communication protocol between personal computer 0

0 50 100 150 200 250 300 350 400 450 500
and devices is command-based – via the LabView application B Pixel Number, n
a command is transmitted to the CCD detection unit and it
responds with corresponding packet of data or device status. If
command is intended to adjust or acquire status for fiber-optic Fig.3. Correlation analysis of two shifted spectrum measurements-A,
switches’ position or stepper motor position then the CCD Noise suppression with spatially shifted measurements-B
detection unit retransmits the command to the control device.

256
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

To filter any noise resulting from signal conversion or in The results from measurements of the transmission spectra
photodiode array and channel inequalities an averaging low- in Channel 1 are shown in Fig.5. Fig.6 presents linear change
pass filter with respect to the spatial signal shift is designed of spectrum change with two LPG sensors compared to the
(2). It averages N-shifted spectral measurements with the reference ASE broadband light source. The same results are
initial signal [7]. Since the correlation function is shown for Ch.2 and Ch.3 respectively in Fig.7-8 and Fig.9-10.
preliminarily calculated and the exact shift is acknowledged, The spectrum change in Ch.1 when sensors are under stress
the average signal is calculated and noises are filtered (bending) is presented on Fig.6, where two LPGs are
1 N simultaneously measured.
S ( n)   S k (n   k )
N  1 k 0 (2)
 k 0 0 0
Signal w/o Force
-5 Signal with Force
By introducing this filtering scheme the acquired spectrum
LPG1 LPG2
preserves any narrow minima in measured grating sensors but -10

Loss, a (dB)
also suppresses noises due to conversion or channel ∆=2.6nm

inequalities in CCD array. The resulting filtered signal is -15

presented on Fig.3b.
-20

-25 a=7.8dB
III. EXPERIMENTAL SET-UP AND L= 100mm
MEASUREMENT RESULTS -30
1480 1500 1520 1540 1560 1580 1600 1620 1640
Wavelength,  (nm)
The designed multi-channel broadband spectrum
measurement system is tested by analysing four different long
period grating sensors arranged in three channels. On Ch.4 Fig. 6. Channel 1 relative measurement with LPG sensor under stress
only a tuneable reflector was connected and this channel was
utilized as reference signal. A joint multiple channel graphic
is presented on Fig.4 with relative measurement to the x10^3
ASE Reference Signal
reference signal. 40
Channel 2 - LPG3
35
Multi-channel spectra measurement 30
Intensity, I

2
Channel 1 25
-3 Channel 2 20
Channel 3
15
Losses, a (dB)

-8
10
-13
5
-18 0
0 50 100 150 200 250 300 350 400 450 500
-23 Pixel number, n

-28
1485 1505 1525 1545 1565 1585 1605 1625 1645 Fig. 7. Channel 2 LPGs’ signal compared to reference spectrum
Wavelength,  (nm)

Fig. 4. Joint multi-channel spectral response measurement

-2 Signal with Force

35
x10^3 ASE Reference Signal
30 Channel 1 - LPG1 & LPG2 Signal w/o Force
-7
Loss, a (dB)

25
∆l=2nm
Intensity, I

20 -12

15
-17
10
L= 100mm
5
-22
0 1480 1500 1520 1540 1560 1580 1600 1620 1640
0 50 100 150 200 250 300 350 400 450 500 Wavelength, l (nm)
Pixel number, n

Fig. 8. Channel 2 relative measurement with LPG sensor under stress

Fig. 5. Channel 1 LPGs’ signal compared to reference spectrum

257
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

x10^3 forces on the fibers and also the ability of the demonstrated
ASE Reference Signal
40 system to detect and analyze those spectral changes.
Channel 3 - LPG4
35
30 ACKNOWLEDGEMENT
Intensity, I

25
20 The authors acknowledge the support if the Ministry of
15 Education of Bulgaria under research project VU EES 303/
10 07, the Natural Sciences and Engineering Research Council of
5 Canada, the Canada Research Chairs Program, and
0
NanoQuébec.
0 50 100 150 200 250 300 350 400 450 500
Pixel number, n
REFERENCES
Fig. 9. Channel 3 LPGs’ signal compared to reference spectrum [1] Murphy, D. F., Flavin, D. A., McBride, R., and Jones, J. D. C.,
Interferometric Interrogation of In-Fiber Bragg Grating Sensors
1 Without Mechanical Path Length Scanning, J. Lightwave
Techn., 19, 1004-1009 (2001).
-1 Signal w/o Force [2] Alves, J., Santos, J.L., Carvalho, A., and Lage, A. Fiber Bragg
-3
Sensor Interrogation System Based on a CCD Spectrometer,
Proceedings of IEEE, Sensors, 1, 909-913 (2003)..
Loss, a (dB)

Signal with Force

-5 ∆=27nm [3] J. Choi, Adaptive and Iterative Signal Processing in
Communications, Cambridge Press, 2006.
-7
[4] C.G. Askins, M.A. Putnam E.J. Friebele, Instrumentation for
-9 Interrogating Many-element Fiber Bragg Grating Arrays, SPIE
vol.2444 p.257-266, 1995.
-11 L= 100mm [5] K. Srimannarayana1, M. Shankar,R.S. Prasa, Fiber Bragg
-13
Grating And Long Period Grating Sensor For Simultaneous
1480 1500 1520 1540 1560 1580 1600 1620 1640
Measurement And Discrimination Of Strain And Temperature
Wavelength,  (nm) Effects, Optica Applicata, Vol. 38,No. 3, 2008
[6] S. Vaseghi, Advanced Digital Signal Processing and Noise
Reduction, 4th Edition, Wiley, 2009.
Fig. 10. Channel 3 relative measurements with [7] Farhang-Boroujeny, B., Adaptive Filters – Theory and
LPG sensor under stress Applications, Wiley, 1998.
[8] Bitter, R., Mohiuddin, T., Nawrocki, M., LabVIEW – Advanced
They are particularly selected not to interfere with each Programming Techniques, 2nd Edition, CRC Press, 2007.
others’ spectra. LPG1 shows significant depth change in its [9] Zahariev P., Hristov G., Iliev M., Tsvetkova I., “Extending the
transmission minimum at =1515nm, while LPG2 results in lifetime of wireless sensor networks by mechanisms for routing
minimum shift of Δ=2.6nm. the information in more energy efficient ways”, Mosharaka
International Conference on Communications, Computers and
Corresponding transmission minimum shifts for Ch.2 and
Applications MIC-CCA, Istanbul, Turkey, pp. 73 – 78
Ch.3 are presented on Fig.8 and Fig.10. The latter sensor is [10] Guan, Z.G., Zhang, A.P., Liao, R., “Wavelength Detection of
particularly sensitive resulting in a larger minimum shift of Coherence-Multiplexed Fiber-Optic Sensors Based on Long-
Δ=27nm. Period Grating Pairs”, IEEE Sensors Journ., 7, 36-37, 2007.
[11] Tamhane, A. C., “Statistical Analysis of Designed Experiments:
Theory and Applications”, Wiley, 2009.
IV. CONCLUSION [12] M. Mikhov, P. Nakov,”A Hybrid Step Motor Drive System with
High Positioning Accuracy”, Proceedings of the International
The reported automated multi-channel broadband analysis Conference on Electronic Devices and Systems, pp. 59-64,
system for measurement of fiber-optic grating sensors is Brno, Czech Republic, 2005, ISBN 80-214-2990-9.
capable of simultaneously monitoring up to 12 long period [13] M. Mikhov, P. Nakov, “Stepping Motor Drive for Precise
gratings in groups of three spectrally multiplexed sensors per Positioning Applications”, Proceedings of the International
Scientific Conference on Information, Communication and
channel. An advanced signal processing and analysis is
Energy Systems and Technologies, Vol. 1, pp. 227-230, Nish,
demonstrated using cross-correlation function and adaptive Serbia, 2008, ISBN 978-86-85195-59-4.
filtering techniques which effectively suppress noises while [14] Zahariev P., Hristov G., “Performance evaluation of data
preserving the spectral resolution. delivery approaches for wireless sensor networks”, World
It is also possible to extend significantly the measured Conference on Information Technology WCIT 2010, Istanbul,
spectrum range by rotating the diffraction grating with the Turkey; Procedia Computer Science, Volume 3, 2011, Elsevier,
precise stepper motor and the wide sensitivity range of the pp. 714 – 720
linear CCD photodiode array (800-1750 nm). [15] Tsvetkova I., Aleksandrov Y., Hristov G., Zahariev P., Iliev M.,
The measurement results demonstrate the high sensitivity of “Comparison of target tracking algorithms in hierarchical
WSNs”, World Conference on Information Technology WCIT
implemented long period grating sensors to external strain
2011, Antalya, Turkey

258
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Measurement of the Position by Using

Hybrid Pseudorandom Encoder
Dragan Denić1, Goran Miljković2, Jelena Lukić3, Miodrag Arsić4 and Milan Simić5
Abstract – This paper proposes the position measuring method drawback of error accumulation. There were attempts to
conducted with a hybrid pseudorandom encoder designed by compensate these drawbacks by using the optical calibration
combining the pseudorandom and the incremental code tracks. methods or special code marking of the specified locations
This combination of code tracks increases the measurement [2]. Unfortunately, the usage of these solutions is limited with
resolution. The proposed hybrid pseudorandom encoder contains
the number of reference points, i.e. for economic reasons.
the detector of the pseudorandom code reading errors, which
helps to reduce the frequency of the autocalibration and The result of the latest research in the field of digital
eliminates the possibility for the coarse error. The paper also position transducers for AGVs is the pseudorandom coding
describes the testing of the proposed encoder electronic block in method, which, for the absolute position determination,
the presence of incremental and code track contaminations. requires only one bit per quantization step, [3], [4]. The
longitudinal coding method was developed with the help of an
Keywords – position measurement, hybrid pseudorandom useful pseudorandom code property that the first n-l digits of a
encoder, automatically guided vehicles (AGV) pseudorandom code word are identical with the last n-l digits
of a previous pseudorandom code word. Therefore, unlike
transverse coding technique, which requires that for each
coding device sector a digital code is written in the transverse
I. INTRODUCTION
direction, this technique allows the absolute position
measurement with the use of just one code track. It is based on
The high accuracy measurement of the lengths and the
the „window property“ [3] of the pseudorandom binary
angles is often requested in modern industrial production
sequences (PRBS) {S(p)/p=0,1,...,2n-2}. Accordingly, any n-
processes. The accuracy of 0.1 µm is no rare and it is more
bit long code word {S(p+n-k)/k=n,...,l} provided by the
often requested. Such a request is a result of the high level of
scanning of the PRBS with the window of width n,
production automation and strict requirements regarding the
{x(k)/k=n,...,l}, is unique and may fully identify the absolute
positioning of the movable parts of the manufacturing units.
position of the window p with regard to the sequence start.
At the same time, the requirements for the position measuring
In the case when the high resolution of the measurement is
of the automatically guided vehicles (AGV), which move
required, the PRBS of the greater length need to be applied
independently on the factory floor as a production platforms
because the operating range of the AGV can be 300-400 m.
or transport units, are stricter. These vehicles have the
Therefore, due to the code conversion problem [4], for a given
increasing tendency towards the flexible movable systems
maximum velocity of the AGV, the overall measurement
which represent the combination of vehicles and robots [1].
resolution is limited. Even if there is no such a problem, the
These automatically guided vehicles can often be in the
code reading uncertainty limits the overall measurement
interaction with the other production units, such as robots,
resolution of the pseudorandom position encoder. However,
machine tools, charging and discharging units and etc.
the convenience of the pseudorandom coding application can
Therefore, the precise and highly reliable information about
be utilized for the case when the high precision of positioning
the position is necessary. For this reason the utilization of the
is required. The basic idea, which will be presented in this
classical absolute measuring systems is unacceptable for
paper, was derived from the fact that the pseudorandom
economic reasons, because of the large number of code tracks.
coding is very suitable for use with incremental position
Today, most of the systems rely on the incremental methods
measurement methods. PRBS-s can be used for encoding of
for the position determination, which in turn have the known
the reference markers used for the measuring system
autocalibration. This approach leads to the realization of the
1
Dragan Denić is with the Faculty of Electronic Engineering at hybrid measuring system. The aim of this paper is to
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E- emphasize the great advantage of the hybrid method for the
mail: dragan.denic@elfak.ni.ac.rs measuring of the AGV position.
2
Goran Miljković is with the Faculty of Electronic Engineering at From the measurement and the position determination
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E- points of view, the hybrid measuring system uses a
mail: goran.miljkovic@elfak.ni.ac.rs
3 measurement method that combines the absolute and the
Jelena Lukić is with the Faculty of Electronic Engineering at
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E- incremental methods. The hybrid measuring method includes
mail: jelenalukicpk@gmail.com the functional elements of both methods. The intention is to
4
Miodrag Arsić is with the Faculty of Electronic Engineering at retain the good qualities of these methods, and to eliminate
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E- their disadvantages. In 1983, Whitwell proposed [5] that the
mail: miodrag.arsic@elfak.ni.ac.rs track with conventional markers (measurement grid) needs to
5
Milan Simić is with the Faculty of Electronic Engineering at be placed along the several code tracks of some absolute
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E- scale. Using n code tracks for each sector of the width d the
mail: milan.simic@elfak.ni.ac.rs

259
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

digital code is written. At the same time, the conventional be fulfilled. For the case when the number of pulses is
incremental encoder with minimum two detectors is applied. quadrupled, as it is proposed here, the number of markers (or
Two signals phase shifted for π/2 are needed for the AGV reflecting areas, in the case when an optical reflection method
motion direction determination. is applied) in one sector, for the absolute position
Whenever the determination of the absolute position is determination, is g=2m-2. The (n+m)-bit UP/DOWN counter is
carried out (by direct reading of the digital code in the applied and thus the reached output resolution is (n+m)-bits.
transverse direction) the accumulated encoder errors are The code reading uncertainty problem is solved by using the
automatically eliminated. Further incremental position external synchronization method. Along the code track an
determining is performed with respect to the absolute value of additional synchronization track is placed. The code reading is
the determined position. done at the moment when the sensor head AUT detects the
Any error in the absolute value of the determined position transition between two adjacent sectors, [3], [4].
would represent a systematic error which can be eliminated
only during the next absolute position determination process.
However, this systematic error would be enormously large in
relation to the accuracy and resolution of the applied
incremental encoder. Except this drawback, the disadvantage
of a large number of code tracks also remains. For the case of
movable systems with a wide range of movement, even at
relatively low resolution of the absolute position
determination the number of code tracks is so large, that the
proposed hybrid system becomes economically unacceptable
for the implementation.

II. THE HYBRID PSEUDORANDOM ENCODER

The main drawback of the previously described "classical"
hybrid position encoder is that it has a high number of code
tracks applied for writing the digital code in the transverse
direction. This drawback can be eliminated with
pseudorandom position encoder employment for
determination of the AGV absolute position. The benefit of
combining pseudorandom and incremental coding methods
was indicated in [3]. This is done in the terms of resolution
increase by introducing a greater number of sensor heads
arranged according to the "vernier" method. This opened the
possibility for measuring system functioning according to the
incremental method. However, in this measuring system are
not included the main advantages of the incremental method,
such as: simplicity, small number of connecting wires and the
high density of measurement range partitioning. As noted
above, the hybrid encoder represents the direct combination of
the absolute and the incremental encoder [6]. The solution
proposed in this paper is shown in the Fig. 1. Along the
synchronization and the code track the measurement grid with
the conventional markers is added. The whole electronic
block, required for obtaining the information on the relative
position of the AGV in binary code, is included. It may be
Fig. 1. The hybrid pseudorandom position encoder
noted that the pseudorandom position encoder is simply
extended by adding a complete two-phase incremental Otherwise, the hybrid position encoder from Fig. 1., does
encoder. The number of pulses is quadrupled by combining A not need the additional VER head, which is used at
and B signals using the EXOR circuit and with edge detection pseudorandom encoders for determining of the movable
of thus obtained signal. To determine the direction of system movement direction, [3]. The AGV movement
movement one of the solutions described in [7] can be direction is now determined with the applied incremental
applied. Otherwise, depending on the detection technique and encoder. The marker, in relation to which the position is
specific detectors applied, the distance between y(1) and y(2) determined, is placed opposite the detector y(1). The n-bit
can be r=(e+1/4)d/g=(e+1/4)q, where e can take an arbitrary adder is no more needed as the correction element [3]. The
value from the set {0,1,2,3,...}, and g is the number of difference in the position of the AGV on the same transition,
markers per one sector width d. In general, if the resolution depending on the AGV movement direction, is now q/4. This
increase for m bits is needed, the requirement for g=2m should

260
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

correction for the least significant bit is automatically III. EXPERIMENTAL RESULTS
performed during the measuring system autocalibration.
The proposed measuring system operates according to the The development of an experimental system can be of great
incremental method. At the moment when the synchronization importance for the research of various solutions of position
head AUT detects the transition between two sectors, of width encoders. The realization of different encoder solutions
d, the pseudorandom code reading is done. In this case, the requires the code tracks manufacturing, which can be
control logic for LOAD signal obtaining is very simple and expensive and impractical because the obtained solutions need
can be realized by using the Schmitt circuit (or the the application of different code reading methods. Therefore,
comparator) and edge detector. Converting of the read code the simulation of the system movement and the testing of the
into the natural code gives the absolute position of the AGV. proposed encoder electronic block should be performed with
At the same time a pulse is obtained at the output of the delay the computer and the corresponding hardware application.
circuit and the absolute position is loaded as a new state of the In order to test the electronic block of the position encoder,
counter. These n bits of the AGV current position are an experimental system consisted of the computer, the parallel
respectively entered in the first n inputs of the UP/DOWN I/O card and the microprocessor development system based
counter starting from the input of the greatest weight. In the on the Intel 8031, is built (Fig. 2.). So, in this experimental
next m-1 counter stages the logic zeros are entered. At the system configuration the electronic block of the
least significant input logic one or logic zero is fed depending pseudorandom position encoder is realized by using the
on the AGV movement direction, Fig. 1. In this way the microprocessor development system based on the Intel 8031,
measuring system autocalibration is performed. It continues to while the computer is used for the simulation of the movable
work according to the incremental method and in relation to system movement. By starting the program for the simulation
the new counter state. of the movable system movement at the corresponding
This increase of this solution complexity in relation to the parallel outputs of the I/O card the signals, equal to those
pseudorandom position encoder complexity is insignificant which would be obtained at the output of sensor heads for the
comparing to the quality obtained. The proposed hybrid given movement route, are generated. These signals are fed to
position encoder is a cost-effective solution even for movable the port 1 of the Intel 8031microprocessor. After the testing is
systems with the highest requirements. It can be applied in finished, by using the tabular or graphical presentations the
systems with the wide range of movement, which require high given series of positions and the obtained series of positions
precision and high resolution of the position measuring can be compared. The tabular view provides the continuous
because of the mutual influence with the other production information about the measuring system working regime.
units. Since all the real situations can be simulated, including
However, the act of autocalibration represents a critical damaged or contaminated measuring tracks, there is no reason
moment in the functioning of the hybrid position encoder. to expect a different behavior of the measuring system in its
That is the reason why the hybrid position encoder has not concrete application. In other words, the test results fully and
found its application in practice until today. The word realistically characterize the proposed position encoder.
"calibration" implies the existence of a reference against
which the correction is made. The question is whether the
measured absolute position satisfies the criteria of one
reference. At first instant it does. The code is written on a
physical track and it can be read only at precisely determined
locations. Such information can be accepted as a reference,
but it is necessary to ensure the conditions that will guarantee
that no errors will happen during the code reading procedure.
Unfortunately, no one can claim that in real industrial
conditions such terms will be provided. Approaching to these
ideal conditions can be very expensive, but it does not solve
the problem. Therefore, in this paper we started with a new
approach. The additional information is introduced for Fig. 2. Testing of the position encoder electronic block
indicating the presence of an error, with the probability of
one, in the pseudorandom code reading. For the realization of The electronic block has been tested during a period of
the hybrid position encoder the detector of the pseudorandom several months and for over a thousand randomly selected
code reading errors is required. It confirms that the obtained AGV movement routes. After the measuring system entered
pseudorandom code word corresponds to a certain physical into the normal working regime, for the ideal measuring
reference, with which the current AGV position coincides. tracks located in an ideal environment case, there is no typical
Only in this case the measuring system autocalibration is AGV movement which would lead to an error in the position
done. Otherwise, if such confirmation does not exist, the determination or which would take the measuring system out
obtained pseudorandom code word will not be accepted as the of its normal working regime.
reference and the measuring system autocalibration will not The simulation of incremental and code track
occur. In this way the autocalibration frequency becomes contamination can be done with the program that has been
lower, but the possibility of the course error is eliminated. already developed for the simulation of the movable system

261
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

movement. The contamination level is specified in percentage.

As a consequence of the code track contamination the
frequency of the measuring system autocalibration decreases.
Fig. 3 gives the graphical representation of the 16-bit hybrid
encoder test results when the incremental track contamination
level is 9.4% and an additional code track contamination level
is 3%. Obviously, the position measurement error
considerably increases with wrong pseudorandom code bit
readings. In the best case, the maximal measurement error is
about ten times higher. The error is even higher when the code
reading synchronization is lost, which often happens. As it is
noted above, instead of 64 quantization steps, the next
measuring system autocalibration, in the worst case, does not
occur until 1280 quantization steps are ran over in the same
AGV movement direction. Specifically, in the case of the
performed testing, Fig. 3, there was the code reading
synchronization loss. The next measuring system
autocalibration was performed after 16 absolute quantization Fig. 3. Testing results of the encoder electronic block when the
steps d of the AGV movement. As it can be seen from the incremental track contamination level is 9.4% and the code track
error graphic, i.e. from the graphic of real and determined contamination level is 3%
position difference, the previous condition caused the
maximal error of one hundred increments. Normally, at the Of course, it should be noted that the adopted incremental
moment when the error is detected, the measuring system track contamination level during testing was much higher than
provides the information that it switched to the incremental it has been reported in practice.
working mode. The measuring system continues to function as
a classical two-phase incremental position encoder until the ACKNOWLEDGEMENT
next autocalibration of the measuring system is performed. It
should be mentioned again that the adopted contamination
Research activities presented in this paper, are supported by
level of the incremental track is much higher than it is usually
funds of the Ministry of Education and Science of the
reported in the practice, and that the considered case
Republic of Serbia, having the reference project number
represents the worst possible case. It is important to say that
TR32045.
during the long testing period of the encoder electronic block,
each simulated code reading error was detected.
REFERENCES
IV. CONCLUSION [1] J. Zygmont, ''Guided Vehicles Set Manufacturing in Motion'',
High Technology, pp. 16-21, December, 1986.
The proposed hybrid pseudorandom position encoder [2] T. Hongo, "An Automatic Guidance System of a Self-
solution represents a new approach in the field of position Controlled Vehicles'', IEEE Trans., Ind. Electron., vol. IE-34,
measurements. The hybrid pseudorandom encoder, which pp. 5-10, 1987.
combines the good properties of the incremental and the [3] E. M. Petriu, J. S. Basran, "On the Position Measurement of
absolute position measuring methods, is applicable in practice Automated Guided Vehicles Using Pseudorandom Encoding'',
due to the presence of the code reading error detector. For the IEEE Trans., Instrum. and Meas., vol. IM-38, no. 3, pp. 799-
first time, the reliability of the obtained position information 803, 1987.
[4] D. Denić, M. Arsić, D. Živanović and M. Pešić, ''Elektronski
is considered and the reliable detection of the possible errors sistemi za pozicioniranje automatski upravljanih vozila'',
is provided. The results obtained so far indicate the fully Zbornik radova SAUM, Kragujevac, Srbija, pp. 503-514, Jun,
functionality of the proposed encoder. The software 1992.
realization of the encoder electronic block is also possible by [5] A. L. Whitwell, ''Moire Techniques Ensure Unerring Positional
using the microprocessor. The realized encoder electronic Control'', Design Engng., pp. 45-48, November, 1973.
block has been tested by using an experimental system [6] D. Denić, M. Arsić, ''Precizno pozicioniranje automatski
developed for the AGV movement simulation, even in the upravljanih vozila primenom metoda hibridnog merenja
presence of damage or contamination of the measuring tracks. pozicije'', Zbornik radova SAUM, Novi Sad, Srbija, pp. 341-
After testing it was shown that the maximal error is one 345, Oktobar, 1995.
[7] P. Thajchayapong, M. Chinnakarn, ''Further Improvement in a
hundred increments when the incremental track contamination
Counting and Direction-Sensing Circuit”, International journal
level is 9.4%, when the code track contamination level is 3% of electronics, vol. 66, no. 6, pp. 935-938, 1989.
and in the presence of the code reading synchronization loss.

262
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Method for calculating the stability at moderate and big

heeling angles of a vessel
Mariya Eremieva1, Viktoriya Sabeva2, Mariya Nikolova3, Emiliya Koleva4
Abstract – Calculating the stability of a vessel at moderate and
big heeling angles by building Static Stability Curves. The
method introduces the notation Dynamic Height of the Mass
Centre (DHMC).

Keywords – Static Stability Curves, Dynamic Height of the

Mass Centre, Metacentric Height

I. INTRODUCTION
“Stability” is the ability of the vessel to return to its initial
position before capsizing after the inclining forces have
ceased to operate. Hence the stability issue is an issue of the
ship behavior in transverse and longitudinal angle inclination
caused by external effects. The change of the stability is
searched by the change of the metacentric height and the Fig.1. Stability at external effects
corresponding coefficients. The existing dependences for
evaluating the stability are based on the criterion of minimal Furthermore, some basic assumptions are usually taken into
starting metacentric altitude and guarantee basically the consideration, one of which is the principle of the geometric
starting stability. This report presents a new method for inverse which states that the vessel is always considered
evaluating the stability using “dynamic metacentric height” upright and the waterline WL is inclining at a certain angle
(DMCH). (  ). Assume that the point M is the cross point of the
directrix of two infinitely closed buoyancy forces. At
moderate and big angles of the heel, after applying the
II. METHOD FOR EVALUATING THE DMCH principle of geometric inverse, the point M doesn’t lie down
in the diametrical plane CL of the ship. This assumption
When sailing each vessel is exposed to external forces like
wind, heavy sea, surge and etc. Its safety depends on the makes pointless the using of GM and it’s necessary to
seaworthiness quality - stability. In basic aspects this means investigate the change of the arm GZ (the arm of the righting
that the ship must counteract the negative force effects. That is
to say not to heel to dangerous angles and to redress its initial moment) or the righting moment itself ( M r ). Hence
balance after the external influences are over. According to M r  GZ (1)
the magnitude of the heeling angle  (transverse inclination of
could be represented as a function of the heeling angle 
the vessel) the stability is classified in two types: initial
stability and stability at moderate and big angles of list (heel). GZ  f1 ( ) or M r  f 2 ( ) (2)
In both cases the counteraction is due to the couple forces: denoted as Static Stability Curve (SSC). (Figure 2)
buoyancy (Archimedes) force and the weight P . They form a
moment M r that is the bases of stability and is opposite to
the heeling moment M l . At small angles  is defined the

notation “initial metacentric height” GM . This altitude is

enough to evaluate the initial stability in details. (Figure 1)

1
Mariya Eremieva, 2Viktoriya Sabeva, 3Mariya Nikolova,
4
Emiliya Koleva are at the Engineering Faculty at Naval Academy
“Nikola Yonkov Vaptsarov”, 73 Vasil Drumev str, Varna 9026,
Bulgaria, Fig.2 – Dynamic areas in Static Stability Curve
e-mails respectively:
eremievam@abv.bg, viktoriya_sabeva@yahoo.com,
mpn@abv.bg, emiliya_f@abv.bg.

263
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The SSC examination, used to evaluate the stability, is

For the arm of the form ( KN ) i ,n are used the KN - curves for
related to the way the external effects are enforces – static or
the specific ship for   10 ,   20 ,   30
dynamic [1]. In practice dominates the dynamic enforce. 0 0 0
and often
That’s why the given evaluation is expressed in determining
  40 . (Fig. 3) Next step is to calculate the areas S 00 300 ,
0
the work of the righting moment i.e. the magnitude of the area
under SSC to certain heeling angles. The requirement of the S 00 400 and S300 400 (Fig. 2) for each SSC and compare them
International Maritime Organization (IMO) concerning the
with the ones required in IMO. If the three areas are the same
minimal values of the areas and angle ranges are lied down in
as the required or one of them is the same and the others are
resolution A749(18) as a guarantee for certain aspect of
stability. They are the following [3]: bigger, than the value (KG) max is maximum acceptable
1) Area of diagram S1 for   0 up to   30 ,
0 0
altitude for positioning the mass centre according to this
S1  0.055 (but not smaller than 0.055)m.rad displacement. The graph of all KG max  f ( i ) shows that
2) Area S 2  0.09m.rad up to the utmost angle the mass centre of the ship should not be above it when
constructing a cargo plan [2].
 , which is interpreted as the smallest of the The method is applied in the paper “An algorithm and a
three variants: program module for calculating the border height of the mass
a) flooding angle  f ; centre of a vessel” using documentation from the Naval
Academy’s training ship “Nikola Yonkov Vaptsarov”.
b) heeling angle corresponding to GZ  m ;
0
c) angle equal to 40 . III. CONCLUSIONS
3) Area S 3 between   30 and    ,
0

1) The evaluation of the stability using the dynamic height

S3  0.03m.rad of the mass centre (DHMC)
The direct building of the diagrams under the terms of IMO
is too hard for the command staff because of the many
KG j max  f ( i ) (6)
calculations. Therefore the suggestion is to introduce in ship’s is better than the existing similarly dependences
papers a diagram of the dynamic height of the mass centre KG  f (d )  f1 () , ( d - draught) (7)
(DHMC) of the vessel that meets the required norms of IMO
and represents the upper limit for placing the mass centre in or GM  f (d )  f 2 () . (8)
height. That way in designing the cargo plan is estimated They are based on the criterion of minimal starting
whether the obtained mass centre satisfies the requirements metacentric altitude and guarantee basically the starting
for dynamic stability. stability i.e. they should not be considered as reliable
In creating the shown diagram (fig.2) are examined all the evaluation of stability.
SSC in the range from “empty” to “full” ship with given 2) The suggested DHMC removes the necessity of building
realistic height tolerance of the mass centre. It means that the SSC when realizing a cargo plan. The results are as authentic
as the more complex classical SSC.
arm of the righting moment GZ is considered as a function
of the displacement  i , the altitude of the mass centre KG j
ACKNOWLEDGEMENT
and the accepted angle interval  n . (Fig. 3) Since the
indicated argument define the arm of the form The authors would like to acknowledge Assoc. Prof.
P. Metodiev for the special help and support.
KN i ,n  f ( i ,  n ) (3)
and the arm of the height
REFERENCES
(lG ) j ,n  KG j * sin  n , (4)
[1] M. Zhelyazkov, P. Kaloyanchev, “Book of problems based on
hence (GZ ) i , j ,n  KN i ,n  ( KG) j * sin  n . (5) ships theory”, Military publisher, 2001.
[2] P. Petkov, V. Petkova, I. Draganov, “Theory of the ships and
ships construction”, Steno, 2008.
[3] Resolution А 749 (18) of IMO-requirements for SSC.
[4] STCW Module 17-Introduction to Ships.
[5] STCW Module 7-Ship Construction, Stability and Damage
Control.

Fig.3 – Arm of the form and arm of the height

264
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Design of a high – sensitive capacitive sensor for wireless

monitoring of bulk material’s level
Teodora Trifonova 1, Valentina Markova 2, Valentin Todorov3 and
Ventseslav Draganov 4
Abstract - The design of high sensitive capacitive sensor achieved Upon reaching certain, predefined ratio of the both
by reducing the destabilizing impact of temperature and supply frequencies, control signal is generated at the output of the
voltage is reported. The possible influence of the generators’ comparator [6].
frequencies on the capacitance measurement is analyzed. Stability and
SENSOR
sensitivity of several types of stabilized quartz generators is also S
investigated. Based on results a new intelligent capacitive sensor for
measuring the extremes level of bulk materials is created. The Cx G1 fx
developed sensor is suitable for use in wireless sensor network. Out
DC
Keywords – Сapacitive sensor, sensor network, quartz
G2 fr
generator.

I.INTRODUCTION Fig.1. Scheme to monitor the capacity variation

of capacitive sensor
In recent years there has been an explosion in sensor
The sensitivity of the sensor is determined by the ratio of the
technology. The selection of sensor for a given application
frequencies of both generators. It can be changed in wide ranges
depends on the nature of physical objects that must be observed,
by setting the digital comparator [7].
such as temperature, pressure, humidity or level of materials [1].
To increase the sensitivity of the capacitive sensor it is
Traditionally, environmental monitoring is achieved through
necessary to reduce the instability of the two generators’
expensive sensors with high accuracy. Creating a wireless sensor
frequencies, caused by alteration of the temperature and supply
network provides an alternative solution by deploying a larger
voltage changes of both generators.
number of sensor nodes with less precision. Network as a whole,
There exist a great variety of quartz generators in the
however, provides better spatial resolution of the area and users can
literature. Only several schemas of stabilized quartz generators -
have immediate access to data [2].
with an active element transistor, TTL or CMOS integrated
The capacitive sensor that controls the extremes of bulk
circuit fulfill the requirements of our development.
materials with very low permittivity has been developed by
A great number of investigations on selected schemes have
authors in [3-5]. The sensor provides high sensitivity in spite of
been made. It was found different variation of the frequencies of
the influence of destabilizing factors like variation of
each generator for changes in the ambient temperature and
temperature and supply. Changing the capacity of the sensor
alteration of the supply voltage.
leads to a change in the frequency of quartz stabilized generator
In this publication, due to the limited number of pages only the
to which the sensor is plugged.
best results obtained for the two of analyzed schemes are presented.
The aim of this work is to further enhance the sensitivity of the
Studies have shown that two basic patterns of quartz stabilized
sensor by choosing appropriate generators and quartz resonators,
generator - with TTL integrated circuit (Fig. 2) and with CMOS
and its adaptation for inclusion in the wireless sensor network.
integrated circuit (Fig. 3) are the most appropriate for our sensor.
II. SCHEMATIC DIAGRAM OF THE CAPACITIVE Both schemes are similar. Primary capacitive converter with
CONVERTER capacity CX is connected to the stabilized quartz generator serially
linked with quartz resonator. The frequency of the generator can be
The block diagram of developed capacitive sensor is shown adjusted in small ranges.
on fig.1. A capacitive converter S with capacity CX is
connected to quartz stabilized generator G1. The frequency of III.EXPERIMENTAL RESULTS
the generator is varying within a certain range in capacity
adjustment CX. The output signal with frequency fX from the For greater accuracy and precision of the sensor it is
digital comparator compares the frequency fr of the second necessary to use two identical generators. The experiments were
supporting quartz stabilized generator G2. performed with two pairs of generators. It was used different
T. Trifonova, V. Markova, V. Todorov, V. Draganov are with the
types of quartz resonators with serial resonance frequency FS
Faculty of Electronics, Technical University of Varna, 1 Studentska varying from 1 MHz to 10 MHz in dependence on the lowest
str., 9000 Varna, Bulgaria. E-mails: 1t_t_trifonova@abv.bg, temperature fluctuation.
2
valliq@abv.bg, 3todorov_88@mail.bg, 4draganov_vd@abv.bg .

265
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Cx Cx As the output of the sensor (fig.1) is given control signal upon

reaching a predefined ratio of the frequencies of the two identical
generators, it is more important to compare the temperature

1
2

1
2
X1A X1A dependencies of the frequencies ratio of both generators. The results
X1 X1 for two quartz generators with frequencies FS1.1 = FS1.2 = 5 MHZ and
FS2.1 = FS2.2 = 10 MHZ are presented, respectively on fig.6 and fig.7.
1
2
F1.1 / F1.2 = ϕ (T )

1
2
Y1 C2
Y1 C2 1,000001

1,0000005
R1 R2
R1 R2
1

F1.1/F1.2
D1A D1B
C1 D1A C1 D1B 0,9999995
1 4 Out Out
3 6 1 2 3 4
2 5 0,999999

0,9999985
Fig.3. Capacitive converter 0,999998
Fig.2. Capacitive converter with with CMOS integrated circuit 20 30 40 50 60 70 80
TTL integrated circuit T [o C]
Fig.6. Dependence F1.1 / F1.2= φ (Т), where FS = 5 MHz
IV.EXPERIMENTAL RESULTS F2.1 / F2.2 = ϕ (T )
1,000018
1,000016
For greater accuracy and precision of the sensor it is 1,000014

F2.1/F2.2
necessary to use two identical generators. The experiments were 1,000012
performed with two pairs of generators. It was used different 1,00001
types of quartz resonators with serial resonance frequency FS 1,000008
varying from 1 MHz to 10 MHz in dependence on the lowest 1,000006

temperature fluctuation. 1,000004

20 30 40 50 60 70 80
o
A. Investigations of capacitive converter with TTL integrated T [ C]
circuit Fig.7. Dependence F2.1 / F2.2= φ (Т), where FS = 10 MHz
Temperature dependences of the generators’ frequency FX for Obviously, the deviation of ratio of both generators’ frequencies
frequencies of the quartz resonators, respectively FS1.1=FS1.2 = 1 for variation of environmental temperature with 45°С is much less
MHz and FS2..1=FS2..2 = 5 MHz are1.2
display on Fig.4 and Fig.5. – below 2.10-6/°C for first generators and below 10-5/°C for others.
1.1
5,0006 This corresponds to instability,
F1.1 , less
F1.2than
=ϕ 10-7(U)
/°C.
f1.1,MHz
f1.2,MHz 1,000012
5,00055
F1.1, F1.2 [MHz]

1,000011
F1.1, F1.2 {MHz]

5,0005
1,00001
5,00045 1,000009
1,000008
5,0004
1,000007
5,00035
1,000006 F1 [MHz]
5,0003
F2 [MHz]
1,000005
20 30 40 50 60 70 80 3 3,5 4 4,5 5 5,5 6
o
T [ C] U [V]
Fig.4. Dependence Fx = φ (T), where FS = 5 MHz Fig.8. Dependence F1.1 = φ (U), F1.2 = φ (U)), where FS = 1MHz
9,99725 5,0007
f2.1,MHz
f2.2,MHz
F2.1, F2.2 [MHz]

9,9972 5,0006
F2.1, F2.2 [MHz]

9,99715 5,0005
9,9971
5,0004
9,99705
5,0003
9,997
5,0002 F1 [MHz]
9,99695
F2 [MHz]
5,0001
9,9969
3 3,5 4 4,5 5 5,5 6
20 30 40 50 60 70 80
U [V]
T [oC]
Fig.9. Dependence F2.1 = φ (U) F2.2 = φ (U), where FS = 5 MHz
Fig.5. Dependence Fx = φ (T), where FS = 10 MHz
Fig.8 and fig.9 illustrate the dependences of the generator’s
It is clearly seen from results that the deviation of the frequencies frequency FX for change of the voltage UX if the frequencies of quartz
is extremely small – bellow 10-4 for FS = 5 MHZ and bellow 2.10-5 resonators are FS1.1 = FS1.2 = 1 MHz and FS2.1 = FS2.2 = 5 MHz. The
for FS = 10 MHZ with changes in ambient temperature with 45° C. figures show that the frequency is changed in very small range – of
This corresponds to instability, less than 10-5 / °C (FS = 5 MHZ) and the order of 5.10-6/V for the first resonator and below 10-4/V for the
less than 10-6 / °C (FS = 10 MHZ). second in case of voltage alteration U from 4 to 6V.

266
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Next two figures (fig.10 and fig.11) are connected with Dependencies of the frequencies of the generators FX.Y by the
investigation the impact of the variation of the voltage on the ratio of variation of the ambient temperature T, given frequencies of the quartz
the generators’ frequencies, respectively, for frequencies of both resonators, respectively FS = 5 MHz and FS = 10 MHz are presented on
quartz generators FS1.1 =FFS1.2 =/ 1FMHZ=and FS2.1 = FS2.2 = 5 MHZ.
ϕ (U) fig.12 and fig.13. Temperature instability under these conditions is less
1.1 1.2 ϕ
1,0000047 than 10-5/°С for FS = 5 MHz and bellow 10-6/°С for FS = 10 MHz.
1,0000046 As mentioned in the previous section for proposed schemes (fig.2,
1,0000045
fig.3) the temperature dependence of the ratio of both generators’
F1.1 / F1.2

1,0000044
1,0000043 frequencies is more importantly. It can be seen from fig.14 and fig.15
1,0000042 that the temperature instability of the ratio F1.1 / F1.2 is less than 10-
1,0000041 6
/°С if FS1.1 = FS1.2 = 5 MHZ, respectively for F2.1 / F2.2 the instability is
1,000004
1,0000039 bellow 3.10-7/°С for FS2.1 = FS2.2 = 10 MHZ.
1,0000038
1,000007
3 3,5 4 4,5 5 5,5 6
U [V] 1,000006

1,000005
Fig.10. Dependence F1.1 / F1.2= φ (U), where FS = 1 MHz
2.1 2.2

F1.1 / F1.2
1,000004
1,000005
1,000003
1
1,000002
0,999995
F2.1 / F2.2

1,000001
0,99999
1
0,999985 0,999999
0,99998 20 30 40 50 60 70 80
o
T [ C]
0,999975
0,99997
Fig.14. Dependence F1.1 / F1.2= φ (Т), where FS = 5 MHz
3 3,5 4 4,5 5 5,5 6
1,000018
U [V] 1,000016
1,000014
Fig.11. Dependence F2.1 / F2.2= φ (U), where FS = 5 MHz 1,000012
F2.1/F2.2

1,00001
As it was expected variation of voltage U led to alteration in
1,000008
ratio of the frequencies of both generators bellow 5.10-7/V for FS 1,000006
= 1MHz and less than 10-5/V for FS = 5 MHz. 1,000004
1,000002
B. Investigations of capacitive converter with CMOS 1
integrated circuit 20 30 40 o
T[ C]
50 60 70 80

The results of experimental studies of the scheme on fig.3 are Fig.15. Dependence F2.1 / F2.2= φ (Т), where FS = 10 MHz
presented on fig.12 ÷ fig.19. The last group of studies are related to determining influence of
5,0008
F1.1 [MHz]
F1.2 [MHz]
changes in voltage on the frequency of generator, given frequencies of
5,00075
quartz resonators, respectively F
FS1.1
1.1= FS1.2
,F 1.2==1MHz and FS2.1 = FS2.2 = 5MHz.
ϕ (U)
5,0007
1,000012
F1, F2

5,00065 1,000011
F1.1, F1.2 [MHz]

5,0006 1,00001
1,000009
5,00055
1,000008
5,0005
1,000007
20 30 40 50 60 70 80
o F1 [MHz]
T [ C] 1,000006
F2 [MHz]
1,000005
Fig.12. Dependence F1.1 = φ (Т), F1.2 = φ (Т), where FS = 5 MHz
3 3,5 4 4,5 5 5,5 6
U [V]
9,9975
9,99745
F2.1 [MHz] Fig. 16. Dependences F1.1 = φ (U), FS1.2 = φ (U), for FS = 1 MHz
F2.2 [MHz]
9,9974 5,0007
9,99735
F2.1, F2.2

F2.1, F2.2 [MHz]

5,0006
9,9973
9,99725 5,0005
9,9972
5,0004
9,99715
9,9971 5,0003
9,99705 F1 [MHz]
5,0002
9,997
F2[MHz]
20 30 40 50 60 70 80 5,0001
o 3 3,5 4 4,5 5 5,5 6
T [ C]
U X[V]
Fig.13. Dependence F2.1 = φ (Т), F2.2 = φ (Т), where FS = 10 MHz
Fig.17. Dependences F2.1 = φ (U), F2.2 = φ (U), for FS = 5 MHz

267
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

From the resulting dependences, shown on fig.16 and fig.17 2. Using a common stabilized power source of both
follows that the instability of the generators’ frequencies under the generators (fig. 20), with instability of the output voltage below
relevant conditions (F1.1, F1.2, F2.1, F2.2 ) to alteration the voltage is less 0,125 V, also enhanced the sensitivity of the capacitive sensor
than 10-5/V for FS = 1 MHz and bellow 10-4/V for FS = 5 MHz. over 10 times.
Finally, the ratio of frequencies of both generators dependences The proposed integrated stabilizer gives one additional
to variation of voltage, respectively, for frequencies of both quartz advantage of the sensor – the option of its power to be turned on
generators FS1.1 = FS1.2 = 1 MHz and FS2.1 = FS2.2 = 5 MHz are and off for a predefined period of time. This allows the insertion
displayed on fig.18 and fig.19. The frequency instability of both of the sensor in wireless sensor network.
generators is less than 10-6/V. 6
D1
2
Vout
C2 C+ Vout
1,0000047
R1 + C3
1,0000046 5
1,0000045 C- 1
Shutdow n PGOOD
F1.1 / F1.2

1,0000044 7
SHDN
1,0000043 3 8
1,0000042 Vin Vin SELECT
C1 GND
1,0000041 +
1,000004 MCP1252 4
1,0000039
1,0000038
3 3,5 4 4,5 5 5,5 6
U [V] Fig.20. Source of stabilized voltage with control options
Fig.18. Dependence F1.1 / F1.2= φ (U), for FS = 1 MHz
VI.CONCLUSION
1,000005
1
It has been made a number of investigations related to
0,999995
F2.1 / F2.2

0,99999
reduction of the destabilizing influence of ambient temperature
0,999985 and supply voltage to the frequencies of two quartz generators
0,99998 used in capacitive sensor. Based on results was developed high-
0,999975
0,99997
sensitive intelligent capacitive sensor for monitoring extremes
3 3,5 4 4,5 5 5,5 6 of bulk materials. The proposed sensor can be used as a node of
U [V]
a wireless sensor network.
Fig.19. Dependence F2.1 / F2.2= φ (U), for FS = 5 MHz
ACKNOWLEDGEMENT
V. ANALYSIS
This work was supported by the National Science Funds, under Grant
As it is well known the value of frequency instability varies No НП3/2012 “Изследване на комуникационни системи с
depending on particular scheme and used quartz resonator. It използване на цифрови сигнални процесори”.
can be concluded from made experiments that for proposed
schemes of capacity sensor the influence of the variation of the REFERENCES
ambient temperature and voltage to generated frequencies is [1] Dargie W., Ch. Poellabauer. Fundamentals of Wireless Sensor
extremely small. Networ:Theory and Practice, Wiley, 2010. Proc. of The VII Int.
In addition for further increasing the sensitivity of the Conf. ELECTRONICS’98, Book 3, pp. 108-113, 1998
designed capacitive sensor it is necessary to be ensured the [2] J. Yick, B. Mukherjee, D. Ghosal, Wireless sensor network survey,
Computer Networks, Vol. 52, Issue 12, pp.2292-2330, Aug. 2008
equal working conditions for both (the measuring G1 and the
[3] Драганов В., Д. Драганов, Н. Ненков, П. Балабански
supporting G2) generators of the scheme shown in Fig.1. Устройство за измерване на ниво” Авт. свидетелство №
This was achieved in following ways: 45168, МПК – С 01 G 23/26, 08.01.1988
1. To ensure a very small difference in operating temperatures of [4] Draganov V., Il. Tanchev. Possibilities for Registering the Changes of
both generators it was proposed constructive decision using Capacity of a Capacity Convecter. Proc. of the Int. Conf
common integrated circuit for both generators. The quartz Electronics’2004, Sozopol, 22-24 Sep 2004, Book 3, pp.195-199.
resonators were mounted much close to each other. Their [5] Draganov V., Il. Tanchev. Device for Measuring the Level of Bulk
temperature was aligning by an additional thermal connection with Materials in Bunkers, Proc. of the Int. Conf. ICEST’2005, Nish,
heatsink with heat-conveying paste. Serbia, Vol. 1, pp 379-381, June 2005.
Thus, it was assured the temperature difference substantially [6] Markova V., R. Dimova, V. Draganov ,An architecture design of
a monitoring level sensor system”, Annual Journal of
below 1°C, which leads to increase of the capacitive sensor’s Electronics, Vol. 5, N 2, 2011, pp. 37-39, ISSN 1313-1842
sensitivity more than 10 times, as seen from experimental results. [7] Draganov V., Т. Тrifonovа, V. Мarkovа, R. Dimova, D.Kaneva
„Precise measurement opportunities of very small capacity c

268
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Different Implementations of Serial

Pseudorandom/Natural Code Converters
Dragan Denić1, Goran Miljković2, Jelena Lukić3, Miodrag Arsić4 and Dragan Živanović5
Abstract – The duration of the pseudorandom/natural code moment defining [4, 5], and error detection methods [3],
conversion is critical for the absolute position measurement cycle which increase reliability of encoder. One more functional
when pseudorandom position encoders are applied. Because of part of encoder, but no less important than previous ones, is
their simplicity, serial code converters have advantages in pseudorandom/natural code conversion. Pseudorandom binary
implementation. This paper presents serial pseudorandom/natural
code is not suitable for direct application in digital electronics.
code converters and proposes one new faster converter. Concrete
examples for the proposed solution are also presented. There are different methods for pseudorandom/natural code
conversion, and they can be separated on three distinct
Keywords – position measurement, pseudorandom position groups: parallel [6], serial [4] and serial–parallel code
encoder, serial pseudorandom/natural code converter conversion [4]. Parallel solution for code conversion is fast,
but expensive and impractical for long PRBS. Serial code
conversion is developed as one simple and cheap way for
conversion of long PRBS. However, conversion time is
I. INTRODUCTION critical for one absolute position measurement cycle. Through
development of different solutions of serial code converters
The pseudorandom position encoders, their main the main goal is reducing of conversion time. Serial–parallel
advantage, have only one code track regardless to the code conversion is one compromise solution, which combines
resolution. This solution for the absolute position serial and parallel conversion techniques. During mounting on
measurement is based on property of n-bit pseudorandom the shaft pseudorandom encoder provides possibility of direct
binary sequence (PRBS) that each sliding window of length n, zero position adjustment without a significant change of
which passed along a sequence, will extract unique code word hardware and software, but only when serial code conversion
in every moment [1]. Also, the last (n-1) bits of the current is used [7].
code word are equivalent to the first (n-1) bits of the In the first part of the paper existing serial
subsequent code word. The PRBS is useful type of periodic pseudorandom/natural code converters are explained, and then
signal and also has the following properties: the signal is on new faster serial converter is proposed. This new solution
periodic and bipolar, signal exhibits a uniform power spectral employed different feedback configuration of logic gates. The
density over a wide frequency band, signal is deterministic presented solutions are detailed explained using appropriate
repeatable, etc. Pseudorandom binary sequences are also used concrete example.
in cryptography, bit-error-rate measurements, wireless
communication systems, audio applications, etc. The PRBS
generator can be implemented using discrete electronics (shift II. THE SERIAL PSEUDORANDOM/NATURAL CODE
register with D flip-flop cells), using a microprocessor CONVERTERS
(flexible implementation), using a FPGA-based
implementation (flexible and very fast), using virtual The simple solution for pseudorandom/natural code
instrumentation concept [2], etc. conversion is the serial or sequential pseudorandom/natural
The main functional parts of pseudorandom position code conversion method [4], but in the case of high resolution,
encoder are the code reading system [3, 4], where different the conversion time becomes a limiting factor. This method
solutions are developed (with one, two or more heads), code finds the actual value of the position ‘p’ simply by counting
scanning methods in the sense of reliable code reading the steps that the shift register with inverse feedback needs
until it reaches the initial state by successive shifting from the
1
Dragan Denić is with the Faculty of Electronic Engineering at read pseudorandom n-bit word. Serial pseudorandom/natural
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E- code conversion process for n = 7 is shown in Fig. 1.
mail: dragan.denic@elfak.ni.ac.rs Pseudorandom code on the code track is read using of only
2
Goran Miljković is with the Faculty of Electronic Engineering at one code reading head x(7) [4]. In the code conversion
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E- process one Fibonacci generator with inverse feedback
mail: goran.miljkovic@elfak.ni.ac.rs configuration is applied (Fig. 1). The Fibonacci
3
Jelena Lukić is with the Faculty of Electronic Engineering at
implementation consists of a shift register in which a
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E-
mail: jelenalukicpk@gmail.com
exclusive-OR (XOR) gates for modulo-2 sum of the binary-
4
Miodrag Arsić is with the Faculty of Electronic Engineering at weighted taps are used for feedback configuration. The states
University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, E- of the shift register are actually sequential code words of
mail: miodrag.arsic@elfak.ni.ac.rs pseudorandom binary sequence until it came to the state that
5
Dragan Živanović is with the Faculty of Electronic Engineering corresponds to the initial code word. The forbidden state is
at University of Niš, Aleksandra Medvedeva 14, 18000 Niš, Serbia, usually referred to be 0000000, because when all the flip-flop
E-mail: dragan.zivanovic@elfak.ni.ac.rs

269
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 1. Serial pseudorandom/natural code converter based on Fibonacci implementation

values are 0, the XOR will reveal a 0 regardless of the generator is generally faster than the Fibonacci in hardware
location of the taps. Therefore, the fed back value is always 0, due to the reduced number of logic gates in the feedback loop.
and the shift register stays in the 0000000 state. If one of the Now, the total propagation delay in the feedback
feedback values are inverted, or XNOR instead XORs are configuration is equal to the propagation delay of only one
used, the forbidden state may be altered. logical gate. The order of the Galois weights is opposite that
Also, in the serial pseudorandom/natural code converter of the Fibonacci weights, for given identical feedback set. The
one 7-bit counter is added that counts steps and also the logic pseudorandom/natural code converter based on the Galois
for the initial state identification. The basic advantage of this generator is shown in Fig. 2. It is added a logic that the read
code converter is simplicity, and the disadvantage is the serial code word converts to the appropriate content of the shift
connection of logical elements (XOR gates) in the feedback register. When the code word is read, this logic provides the
configuration, which increases the total propagation delay and equivalent state of the shift register that is loaded in that shift
thus conversion time is limited. On the other hand, the register. This logic does not participate further in the code
conversion time limits the maximum rotation speed of the conversion process and thus negligibly influences to
encoder. The table which contains for maximal length conversion time. Furthermore, the steps counted that are
pseudorandom sequences feedback sets for different shift needed for the shift register with the determined and the
register sizes is given in [2, 8]. written state come from the known initial state of the shift
One way of reducing the code conversion time of the register. The obtained number is the result of the conversion,
previous method approximately two times is based on the idea the same as in the case of the known serial code converter
that, thanks to PRBS cycling property, the initial state could (Fig. 1). In the second part of the paper the proposed logic for
be reached using feedback sets that are used for either “direct” translation of the read code word in the appropriate content of
or “inverse” PRBS generating [9]. Depending on the previous the shift register is explained in detail, without which it would
position of the movable system it is decided which PRBS not be possible to realise the new code converter.
generating low (“direct” or “inverse”) would be used for
current code conversion.
Another solution for code conversion process is using the
III. LOGIC FOR INITIAL ADJUSTMENT OF READ
Galois implementation of PRBS generator, which consists of a PSEUDORANDOM CODE WORD
shift register, the content of which is modified at every step by
a binary-weighted value of the output stage, using XOR gates. The read n-bit pseudorandom code word (assigned as x =
The pseudorandom binary sequence generator with a parallel xnxn−1xn−2. . . . . . x2x1) in real time is not identical to the n-bit
feedback logic configuration (Galois generator of current content of the shift register (assigned as X =
pseudorandom binary sequence) is known as a faster XnXn−1Xn−2. . . X2X1), which corresponds to the position of this
pseudorandom binary sequence generator [10]. The Galois code word in the generated pseudorandom binary sequence.
For each n-bit code word of the pseudorandom binary

270
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 2. Faster serial pseudorandom/natural code converter based on Galois implementation

sequence there corresponds exactly one state of the shift The relations for the parallel feedback logic configuration are
register with a parallel feedback logic configuration and it is well known, based on the known serial feedback logic
possible to design a simple logic composed from XOR gates configuration of the n-bit shift register [8]. According to basic
(Fig. 2). For n = 7 will be shown process of logic design, and mathematical relationships X7’ = X6, and X7’ = x6, can be
such a procedure is applicable to any other pseudorandom concluded
binary code word of arbitrary length. On the Fig. 3 is shown
passing through the states of the 7-bit shift register with a X 6 = x6 (2)
parallel feedback logic, which is known as a ‘Galois’ shift Then, after the second clock pulse the content of the shift
register [8, 10]. On the start, suppose that the initial content of register becomes {X7’’, X6’’, X5’’, X4’’, X3’’, X2’’, X1’’}.
the shift register is {X1, X2, X3, X4, X5, X6, X7} and the According to previous principle the following relations are
pseudorandom code word {x1, x2, x3, x4, x5, x6, x7} corresponds obtained:
to that content. The direct generation law of PRBS and the
moving direction from X1 to X7 is applied. The pseudorandom X7’’ = X6’
bit output is always identical to the state X7, and X6’’ = X5’
X5’’ = X4’
X 7 = x7 (1)
X4’’ = X7’ ⊕ X3’
After the first clock pulse, the content of the shift register X3’’ = X7’ ⊕ X2’
becomes {X7’, X6’, X5’, X4’, X3’, X2’, X1’}, where, in X2’’ = X7’ ⊕ X1’
accordance with the direct generation law of pseudorandom X1’’ = X7’
binary sequences for n = 7:
Since X7’’ = x5 and X7’’ = X6’, plus from the previous clock
X7’ = X6 pulse the valid relation is X6’ = X5, there is obtained
X6’ = X5
X5’ = X4 X 5 = x5 (3)
X4’ = X7 ⊕ X3 With the identical procedure for the next clock pulse or by
X3’ = X7 ⊕ X2 writing the relations on the same principle and using the
X2’ = X7 ⊕ X1 relations from the previous clock pulse, and also properties of
X1’ = X7 modulo-2 sum, the following dependences are obtained:

271
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Fig. 3. Contents of the ‘Galois’ shift register through 7 clock periods

X 4 = x4 (4) adjustment of the read code word into a appropriate state of
X 3 = x3 ⊕ x7 (5) the shift register, without which it would not be possible to
X 2 = x2 ⊕ x6 ⊕ x7 (6) realise the code converter proposed here.
X 1 = x1 ⊕ x5 ⊕ x 6 ⊕ x7 (7)
These seven relations define relations between the content ACKNOWLEDGEMENT
of the register and the appropriate pseudorandom binary code
word, and also define the logic for initial adjustment of the Research activities presented in this paper, are supported by
read pseudorandom binary code word shown in Fig. 2. funds of the Ministry of Education and Science of the
Now compare the conversion of the read pseudorandom Republic of Serbia, having the reference project number
code word {0, 1, 1, 0, 0, 0, 1} for the case of applying the first TR32045.
explained serial code converter and the proposed faster serial
code converter. The initial code word is {1, 1, 1, 0, 0, 1, 0}.
According to the algorithms described in [4], the REFERENCES
pseudorandom/ natural code conversion is accomplished
sequentially after loading it into a shift register having a [1] S. Engelberg and H. Benjamin, “Pseudorandom sequences and
reverse feedback equation X(1) = X(4) ⊕ X(5) ⊕ X(6) ⊕ X(7) the measurement of the frequency response”, IEEE Instrum.
(Fig. 1). In the given example there are 10 shifts of register, Meas. Mag., vol. 8, no. 1, pp. 54–59, 2005.
and the counter state is p = 10 at the end, which is actualy the [2] G. Miljković, I. Stojković and D. Denić, “Generation and
value of the current position of the movable system. Let us application of pseudorandom binary sequences using virtual
instrumentation”, Facta Universitatis, Series: Automatic Control
now look at the new code converter, which is shown in Fig. 2.
and Robotics, vol. 10, no. 1, pp. 51–58, 2011.
The code track is the same as shown in Fig. 1. Now, the read [3] M. Arsić and D. Denić, “New pseudorandom code reading
code word is not written directly to the shift register, but feds method applied to position encoders”, Electron. Lett., vol. 29,
the input of logic for the initial adjustment of the read code no. 10, pp. 893–894, 1993.
word (Fig. 2). By the application of relations (1), (2), (3), (4), [4] E.M. Petriu, “Absolute position measurement using
(5), (6), (7) the code word {0, 1, 1, 0, 0, 1, 1} is obtained as pseudorandom binary encoding”, IEEE Instrum. Meas. Mag.,
output. It is now directly saved in the shift registry. Now, the vol. 1, no. 3, pp. 19–23, 1998.
shift register sequentially passes through the following states: [5] D. Denić and G. Miljković, “Code reading synchronization
{1, 1, 0, 0, 1, 1, 0}, {1, 0, 0, 0, 0, 1, 1}, {0, 0, 0, 1, 0, 0, 1}, method for pseudorandom position encoders”, Sens. Actuators
A, vol. 150, pp. 188–191, 2009.
{0, 0, 1, 0, 0, 1, 0}, {0, 1, 0, 0, 1, 0, 0}, {1, 0, 0, 1, 0, 0, 0},
[6] E.M. Petriu, J.S. Basran and F.C.A. Groen, “Automated guided
{0, 0, 1, 1, 1, 1, 1}, {0, 1, 1, 1, 1, 1, 0}, {1, 1, 1, 1, 1, 0, 0}, vehicle position recovery”, IEEE Trans. Instrum. Meas., vol. 39,
and {1, 1, 1, 0, 1, 1, 1} when the stop will be. The state {1, 1, no. 1, pp. 254–258, 1990.
1, 0, 1, 1, 1} is the content of the register, which corresponds [7] D. Denić, I. Ranñelović and G. Miljković, “Recent trends of
to the initial pseudorandom code word {1, 1, 1, 0, 0, 1, 0}. So, linear and angular pseudorandom encoder development”. Int.
at the end of conversion the counter state is p = 10, which is Symp. on Power Electronics, Electrical Drives, Automation and
exactly the same value as in the case of the first serial code Motion (SPEEDAM), Taormina, Sicily, Italy, pp. 746–750,
converter. 2006.
[8] “Linear feedback shift register: Implementation, M-sequence
properties, feedback tables”, New Wave Instruments’, January
IV. CONCLUSION 2004 [Online] Available:
http://www.newwaveinstruments.com/resources/articles/m_sequ
ence_linear_feedback_shift_register_lfsr/
During development of this faster serial
[9] D. Denić, I. Ranñelović and M. Rančić, “Pseudorandom
pseudorandom/nature code converter the goal was to reduce position encoder and code conversion problems”, ICEST, Ohrid,
the conversion time. It is achieved by the reduction in the Macedonia, pp. 437-440, 16-19 June 2004.
number of serial connected gates in the feedback logic, which [10] S. Bourdel, E. Campo, P. Melet and L. Andrieux, “From
provides less propagation delay. For implementation of code modelling of a CDMA transceiver in indoor environment to an
converter, the parallel feedback logic configuration is applied ASIC circuit synthesis”, J. Telecommun. Inf. Technol., vol. 3,
and there has also been designed a simple logic of initial pp. 33–45, 2001.

272
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Cloud systems for environmental telemetry - A case study

for ecological monitoring in agriculture
George Suciu1, Octavian Fratu1, Cristian Cernat1, Traian Militaru1, Gyorgy Todoran1 and
Vlad Poenaru1
Abstract – Large telemetry systems have several hundreds of duration of any installed software. Slapgrid daemon receives
RTUs that are sending data to be processed by intelligence requests from a central scheduler the SlapOS Master which
algorithms and stored in a database that is accessible via Web collects back accounting information from each process.
interface on the Internet. In this paper we present the way in SlapOS Master follows an Enterprise Resource Planning
which SlapOS, an open source provisioning and billing system (ERP) model to handle at the same time process allocation
for distributed cloud computing, is used to gather centrally optimization and billing. SLAP stands for “Simple Language
environmental information from different sensors at remote for Accounting and Provisioning”.
observation points.
This structure has been implemented for cloud-based
automation of ERP and CRM software for small businesses
Keywords – Cloud, Telemetry, Sensors, Remote monitoring,
RTU.
and aspects are under development under the framework of the
European research project “Cloud Consulting” [3]. We will use
our platform hosted on several servers running Ubuntu Linux –
Apache – MySQL template with current software release. On
I. INTRODUCTION our cloud testing environment we provide the platform for
processing information from hundreds different sensors,
enabling the analysis of environmental data through a large
In this paper we develop a test platform for environmental
sample of RTUs.
telemetry and use it as a case study for monitoring ecological
In previous approaches RTUs were implemented in most
parameters in agriculture. We use different types of RTUs
cases on a local server and no company could aggregate
(Radio Transmission Units) and Sensors that monitor and
enough sensor data to consider automating the treatment
transmit important information such as temperature,
process.
precipitation, wind speed and leaf wetness from selected
locations.
The RTUs will transmit sensor data over GSM/GPRS to our II. CLOUD ARCHITECTURE FOR TELEMETRY
cloud platform where we can conveniently process the site-
specific weather and soil data in near real-time, display it in
our web-based visualization application and get detailed A. Cloud Architecture
recommendations when and where to spray and how much to
irrigate - resulting in optimized yield, quality and income. SlapOS is an open source Cloud Operating system which
Our system can also help keeping track of pathogen was inspired by recent research in Grid Computing and in
development, optimize treatments to hit a disease dead on, particular by BonjourGrid [4]–[5] a meta Desktop Grid
warn of frost, and to produce crops as environmentally friendly middleware for the coordination of multiple instances of
as possible and to improve agricultural risk management. Desktop Grid middleware. It is based on the motto that
Falling producer prices and rising costs of production are ”everything is a process”.
increasingly forcing agricultural businesses to optimize SlapOS is based on a Master and Slave design. In this
production costs [1]. Therefore "precision farming", the chapter we are going to provide an overview of SlapOS
selective use of inputs such as water, fertilizers or chemicals, is architecture and are going in particular to explain the role of
now indispensable in modern agriculture. The growing Master node and Slave nodes, as well as the software
environmental awareness of consumers further accelerates this components which they rely on to operate a distributed cloud
process and promotes the usage of remote automatic for telemetry applications.
monitoring system for field information such as the one we
Slave nodes request to Master nodes which software they
developed [2].
We will introduce in this article SlapOS, the first open should install, which software they show run and report to
source operating system for Distributed Cloud Computing. Master node how much resources each running software has
SlapOS is based on a grid computing daemon called slapgrid been using for a certain period of time. Master nodes keep
which is capable of installing any software on a PC and track of available slave node capacity and available software.
instantiate any number of processes of potentially infinite Master node also acts as a Web portal and Web service so that
end users and software bots can request software instances
1
which are instantiated and run on Slave nodes. Master nodes
The authors are with the Faculty of Electronics, are stateful. Slave nodes are stateless. More precisely, all
Telecommunications and Information Technology at Politehnica
information required to rebuild a Slave node is stored in the
University of Bucharest, Bd. Iuliu Maniu, nr. 1-3, Bucharest 060042,
Romania, E-mails: george@beia.ro, ofratu@elcom.pub.ro, Master node. This may include the URL of a backup service
cernatcristi@gmail.com, gelmosro@yahoo.com, which keeps an online copy of data so that in case of failure of
todoran.gyorgy@gmail.com, vlad.wing@gmail.com.

273
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

a Slave node, a replacement Slave node can be rebuilt with the sensor data and Google Maps integration for RTUs
same data. localization, a Wiki, etc. The conditions can include price,
It is thus very important to make sure that the state data region (ex. China) or specific hardware (ex. 64 bit CPU).
present in Master node is well protected. This could be Conditions are somehow called Service Level Agreements
implemented by hosting Master node on a trusted IaaS (SLA) in other architectures but they are considered here
infrastructure with redundant resource. Or - better - by hosting rather as trading specifications than guarantees. It is even
multiple Master nodes on many Slave nodes located in possible to specify a given computer rather than relying on the
different regions of the world thanks to appropriate data automated marketplace logic of SlapOS Master.
redundancy heuristic. We are touching here the first reflexive By default, SlapOS Master acts as an automatic
nature of SlapOS. A SlapOS master is normally a running marketplace. Requests are processed by trying to find a Slave
instance of SlapOS Master software instantiated on a node which meets all conditions which were specified.
collection of Slave nodes which, together, form a trusted SlapOS thus needs to know which resources are available at a
hosting infrastructure. In other terms, SlapOS is self-hosted, given time, at which price and under which characteristics.
as seen in Fig. 1. Last, SlapOS Master also needs to know which software can
be installed on which Slave node and under which conditions.

C. SlapOS Slave

SlapOS Slave nodes are relatively simple compared to the

Master node. Every slave node needs to run software
requested by the Master node. It is thus on the Slave nodes
that software is installed. To save disk space, Slave nodes
only install the software which they really need.
Each slave node is divided into a certain number of so-
called computer partitions. One may view a computer
partition as a lightweight secure container, based on Unix
users and directories rather than on virtualization. A typical
barebone PC can easily provide 100 computer partitions and
can thus run 100 RTU web portals or 100 sensors monitoring
Fig. 1. SlapOS Master – Slave Architecture sites, each of which with its own independent database. A
larger server can contain 200 to 500 computer partitions.
B. SlapOS Master SlapOS approach of computer partitions was designed to
reduce costs drastically compared to approaches based on a
SlapOS master nodes keep track of the identity of all parties disk images and virtualization. As presented in Fig. 2, it does
which are involved in the process of requesting Cloud not prevent from running virtualization software inside a
resources, accounting Cloud resources and billing Cloud computer partition, which makes SlapOS at the same time
resources. This includes end users (Person) and their company cost efficient and compatible with legacy software.
(Organisation). It includes suppliers of cloud resources as well
as consumers of cloud resources. It also includes so-called
computer partitions which may run a software robot to request
Cloud resources without human intervention. It also includes
Slave nodes which need to request to SlapOS master which
resources should be allocated. SlapOS generated X509
certificates for each type of identity: X509 certificates for
people like you and me who login, an X509 certificate for
each server which contributes to the resources of SlapOS and
an X509 for each running software instance which may need
to request or notify SlapOS master. A SlapOS Master node
with a single Slave node, a single user and 10 computer
partitions will thus generate up to 12 X509 certificates: one
for the slave, one for the user and 10 for computer partitions.
Any user, software or slave node with an X509 certificate Fig. 2. SlapOS Slave Node
may request resources to SlapOS Master node. SlapOS Master
node plays here the same role as the backoffice of a SlapOS Slave software consists of a POSIX operating
marketplace. Each allocation request is recorded in SlapOS system, SlapGRID, supervisord and buildout [3]. SlapOS is
Master node as if it were a resource trading contract in which designed to run on any operating system which supports
a resource consumer requests a given resource under certain GNU's glibc and supervisord. Such operating systems include
conditions. The resource can be a NoSQL storage, a virtual for example GNU/Linux, FreeBSD, MacOS/X, Solaris, AIX,
machine, an ERP with web-portal interface for displaying etc

274
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

D. SlapOS kernel In the case of private Cloud, IPv6 replaces existing corporate
tunnels with a more resilient protocol which provides also a
SlapOS relies on mature software: buildout and wider and flat corporate addressing space. IPv6 addressing
supervisord. Both software are controlled by SLAPGrid, the helps allocating hundreds of IPv6 addresses on a single server.
only original software of SlapOS. SLAPGrid acts as a glue Each running process can thus be attached to a different IPv6
between SlapOS Master node (ERP5) and both buildout and address, without having to change its default port settings.
supervisord, as shown in Fig. 3. SLAPGrid requests to Accounting network traffic per computer partition is
SlapOS Master Node which software should be installed and simplified. All this would of course be possible with IPv4 or
executed. SLAPGrid uses buildout to install software and through VPNs but it would be much more difficult or less
supervisord to start and stop software processes. SLAPGrid resilient. The exhaustion of IPv4 addresses prevents in
also collects accounting data produced by each running practice allocation of some many public IPv4 addresses to a
software and sends it back to SlapOS Master. single computer. After one year of experimentation with IPv6
Supervisord is a process control daemon. It can be used to in Romania, using IPv6 native Internet access (more than 50%
programmatically start and stop processes with different users, of worldwide IPv6 traffic), we found that IPv6 is simple to
handle their output, their log files, their errors, etc. It is a kind use and creates the condition for many innovations which
of much improved init.d which can be remotely controlled. would else be impossible.
Supervisord is lightweight and old enough to be really mature
(ie. no memory leaks).
Buildout is a Python-based build system for creating,
III. RESEARCH RESULTS AND FUTURE DESIGN
assembling and deploying applications from multiple parts,
some of which may be non-Python-based. Buildout can be In order to collect the information from the RTUs we
used to build C, C++, ruby, java, perl, etc. software on Linux, developed the following test platform as shown in Fig. 4. The
MacOS, Windows, etc. Buildout can either build applications usage of GSM/GPRS data transmission can be extended in
by downloading their source code from source repositories areas where there is no coverage by using a UHF bridge
(subversion, git, mercurial, etc.) or by downloading binaries operating in the fixed frequency range 430 – 440 MHz
from package repositories (rpm, deb, eggs, gems, war, etc.). connected to a gateway that has access to the Internet.
Buildout excels in particular at building applications in a way
which is operating system agnostic and to automate
application configuration process in a reproducible way.

Fig. 4. General Architecture of Telemetry System

The case study was done on 2 grape yards in Romania

(Bucharest and Blaj) with the following sensors, as seen in
Fig. 5

Fig. 3. SlapOS Kernel and User Software

Every computer partition consists of a dedicated IPv6

address, a dedicated local IPv4 address, a dedicated tap
interface (slaptapN), a dedicated user (slapuserN) and a
dedicated directory (/srv/slapgrid/slappartN). Optionally, a
dedicated block device and routable IPv4 address can be
defined.
SlapOS is usually configured to use IPv6 addresses.
Although use of IPv6 is not a requirement (an IPv4 only
SlapOS deployment is possible) it is a strong
recommendation. IPv6 simplifies greatly the deployment of
SlapOS either for public Cloud applications or for private
Cloud applications. In the case of public Clouds, use of IPv6
helps interconnecting SlapOS Slave Nodes hosted at home
without having to setup tunnels or complex port redirections. Fig. 5. General Structure of RTU and Sensors

275
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

The total quantity of rain reported by the system during the IV. CONCLUSION
months of May – September 2011 was of 222 l/sqm with the
following monthly distribution: May – 33 l/sqm, June – 116 Our system for environmental telemetry can be adapted
l/sqm, July – 49 l/sqm, August - 5 l/sqm, September – 8 l/sqm. also to other applications besides agriculture and meteorology.
Other climatic parameters such as Precipitation, Leaf Knowing how the weather will be is important but knowing
Wetness, Temperature and Relative Humidity can be seen on how the environmental parameters are right now is just as
Fig. 6. much important for power plants, airports, wind and solar
parks, incinerators and landfills - they all need wind,
temperature, radiation data, etc. reliably and up to date.
Even though IPv6 is used to interconnect processes globally
on a SlapOS public or private Cloud, we found that some
existing software on RTUs are incompatible with IPv6.
Reasons varry. Sometimes, IP addresses are stored in a
structure of 3 integers, which is incompatible with IPv6.
Sometimes, IPv6 URLs are not recognized since only dot is
recognized as a separator in IP addresses. For this reason, we
decided to provide to each computer partition a dedicated,
local, non routable IPv4 address.
We hope in the future that Microsoft Windows will also be
Fig. 6. Results climatic parameters during a week (26.09 – supported as a host (Microsoft Windows is already supported
02.10.2011) as a guest) through glibc implementation on Windows and a
port of supervisord to Windows.
Another important parameter we studied is the
accumulation of thermal energy over time, known as degree- ACKNOWLEDGEMENT
days or heat units. The growth and development of plants,
insects, and many other invertebrate organisms is largely This paper is presented as part of the project “Valorificarea
dependent on temperature. In other words, a constant amount capitalului uman din cercetare prin burse doctorale
of thermal energy is required for the growth and development (ValueDoc)” Project co-financed from the European Social
of many organisms, but the time period over which that Fund through POSDRU, financing contract
thermal energy is accumulated can vary. Many organisms POSDRU/107/1.5/S/76909 and part of the project “Cloud
slow or stop their growth and development when temperatures Consulting” and “TELE GREEN”.
are above or below threshold levels. Degree-days and other
heat unit measurements have been used for determination of
planting dates, prediction of harvest dates, and selection of REFERENCES
appropriate crop varieties.
[1] Tokihiro Fukatsu, Tomonari Watanabe, Haoming Hu, Hideo
Yoichi, Masayuki Hirafuji, “Field monitoring support system
for the occurrence of Leptocorisa chinensis Dallas (Hemiptera:
Alydidae) using synthetic attractants, Field Servers, and image
analysis”, Computers and Electronics in Agriculture, vol. 80,
January 2012, pp. 8–16, 2012
[2] Jiang, J.A.; Tseng, C.L.; Lu, F.M.; Yang, E.C.; Wu, Z.S.; Chen,
C.P.; Lin, S.H.; Lin, K.C.; Liao, C.S., “A GSM-based remote
wireless automatic monitoring system for field information: A
case study for ecological monitoring of the oriental fruit fly,
Bactrocera dorsalis (Hendel)”, Computers and Electronics in
Agriculture vol. 62, Issue 2, July 2008, pp. 243–259, 2008
[3] George Suciu, Octavian Fratu, Simona Halunga, Cristian
George Cernat, Vlad Andrei Poenaru, Victor Suciu, “Cloud
Consulting: ERP and Communication Application Integration in
Open Source Cloud Systems”, 19th Telecommunications Forum
Fig. 7. Heat units graphs (daily degree-days and total) - TELFOR 2011, IEEE Communications Society, pp. 578-581,
2011
The calculation methods available for heat unit include: [4] Heithem Abbes, Christophe C´erin, and Mohamed
Averaging, Standard, GDD (Growing Degree-Days), Single Jemni.Bonjourgrid as a decentralised job scheduler. In APSCC
Triangle, Double Triangle, Single Sine, Double Sine and Near 08.Proceedings of the 2008 IEEE Asia-Pacific Services
Real-Time. As shown in Fig. 7 we used the Averaging ComputingConference, pages 89–94,Washington, DC,
Method and the maximum heat unit (26,3 degree days) was USA,2008. IEEE Computer Society.
[5] Heithem Abbes, Christophe C´erin, Mohamed Jemni: A
calculated on the date of 10.07.2011 and the total accumulated
decentralized and fault-tolerant Desktop Grid system for
thermal energy by the crop on 01.10.2011 was 3.484,2 degree- distributed applications. Concurrency and Computation:
days. Practice and Experience 22(3): 261-277 (2010)

276
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

High-quality Primary School Education in the Field of

Electrotechnics and Informatics - Beginning of the
Development of Successful Engineers
Sonja Cvetkovic1 and Zoran Stankovic2
Abstract – Taking into consideration that education of future II.IMPLEMENTATION OF ELECTRICAL AND
electrical and IT engineers starts in primary school through
acquisition of the initial knowledge in the field of electrotechnics INFORMATICS KNOWLEDGE ACQUISITION IN
and informatics, the presented abstract is dedicated that segment SERBIAN PRIMARY SCHOOLS
of education. It shows that present implementation level of
knowledge acquisition and education support in this field by the Acquisition of the first terms and basic knowledge from the
current educational plans and programmes for primary schools field of electrotechnics and informatics starts in the first four
in the Republic of Serbia. It shows the outcome of the grades of primary school within the school subjects (courses)
undertaken researches based on a poll conducted amongst the World around us and Nature and Society. In the table I is
pupils and the questionnaire referred to the preference of shown which topics in the field of electro-technics and
subjects, teaching and methodical units and other activities informatics are taught, within which school subjects and how
related to knowledge acquisition in the field of electrotechnics many classes according to the current curriculum and syllabus
and informatics. It also explains possibilities, available methods
for primary schools in Serbia.
and strategies for pupil education quality improvement in these
fields which are supported by the current educational plans and TABLE I. ACQUISITION OF THE INITIAL KNOWLEDGE IN THE FIELD OF
programmes for primary schools in the Republic of Serbia ELECTROTECHNICS AND INFORMATICS
Topics related to eletrotechnics and
Keywords – Electrical engineers education, Primary school Grade Course name
informatics and teaching hours
education, Education quality
Communication devices (1) Computer
1th World around us
and computer components (1)
2th World around us Computer and information devices (4)
I. INTRODUCTION Electrical properties of materials, Current,
3th Nature and society
Circuit (4)
The education of future electrotechnics engineers starts in
Electrical and magnetic properties of
primary school through acquisition of the initial knowledge in materials, Electrostatics, Current, Circuit,
4th Nature and society
the field of electrotechnics and informatics. Therefore the Experiments in the filed of electrostatics,
quality of teaching and knowledge acquisition of the primary current, circuit and magnetism (6)
school pupils in these fields is extremely important for future In the upper grades, electrotechnics and informatics topics
successful education development of quality experts – are mostly covered by the compulsory courses: Physics and
engineers who will be able to face business challenges of the Technical and informatics education and by elective course
modern society in which technical development happens very Information technology and computer technique which scores
fast. The first part of work is dedicated to the implementation a high percentage in most schools in the Republic of Serbia.
of the knowledge acquisition in the field of electrotechnics General aim of the Physics is for the pupils to familiarise
and informatics within the new curriculum and syllabus for themselves with natural occurrences and principal laws of
primary schools in Serbia [1]. It was shown in which school nature, to get basic scientific literacy, to recognize physical
subjects (courses) and with how many classes the education of phenomena through research and to apply physics laws in
the mentioned fields was carried out. In the second part of the their everyday life and work. Table II shows the curriculum
paper were shown the results of the researches conducted for this course per classes in accordance to the current
amongst the primary school pupils with the goal to estimate teaching plans and programmes for primary schools regulated
how satisfied they are with the quality of the acquired by the Ministry for Education and Science of the Republic of
knowledge in the field of electronics and informatics which Serbia [1].
are determined with the current curriculum and syllabus, and
also to estimate generally how much they are interested for TABLE II. PHYSICS - CURRICULUM
the further education within these fields. The last part of the Course name: Physics
paper is dedicated to the abilities, methods and strategies for 6th grade Course duration in teaching hours: 72 per school years
the improvement of the education quality of pupils in the Introduction (2), Kinematics (14), Forces (14), Measurement (15), Mass
mentioned fields which are supported by the curriculum for and density (15), Pressure (12)
primary schools in Serbia [1-14]. 7th grade Course duration in teaching hours: 72 per school years
1
Forces and motion (25), Friction forces (12), Body balance (11),
Sonja Cvetkovic is with the Primary school "Cele Kula" Radnih Mechanical work, energy and power (15) Thermal phenomena (9)
brigada 28, 18000 Nis, Serbia, E-mail: sonjacvetkovic61@gmail.com 8th grade Course duration in teaching hours: 68 per school years
2
Zoran Stankovic is with the Univeristy of Nis, Faculty of
Oscillatory motion and waves (8), Liht waves and phenomena (15),
Electronic Engineering, Aleksandra Medvedeva 14, 18000 Nis, Electrical field (10), Current Electricity (19), Magnetic field (6), Basic of
Serbia, E-mail: zoran.stankovic@gmail.com atomic and nuclear physics (8), Physics and modern word (2)

277
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

You may notice that topics which in the narrower sense acquired knowledge and skills in the everyday life. In table III
belong to the core field of electro-technics are more present in is given the curriculum for this school subject signed by the
the 7th class. As this is a more comprehensive subject it Ministry of Education and Science of Serbia [1]. The school
requires a solid background in mathematics and other subject Technical and informatics education has grown out
fundamental areas of physics. Almost the entire 8th class from the previous school subject Technical education which
curriculum is dedicated to the electrotechics (Fig. 1). The was studied according to the old curriculum. With the new
teaching plans and programmes include laboratory practical curriculum this school subject leaves more room for studying
drills and experiments with interactive work with pupils as a informatics which is specifically done by introducing some
support in the process of knowledge acquisition of the creative elements in the teaching process, which will be more
relevant fields of physics. Learning quality critically depends talked about in the section IV [1]. Besides the regular classes,
on the level of the school teaching aids equipment. With there are also complementary classes held for the pupils who
regards to the electro-technics teaching, the physics laboratory have dispositions for this subject and for those who want to
at primary school “Cele Kula”, Nis is an example of a well- enter the competitions in this field [2].
equipped teaching aids school which include: complete The goal of the school subject Information technology (IT)
electrostatics experiments, influent machine, mathematical and computer technique is to enable pupils to use computers,
and electrical pendulum, Faraday’s cage, electrical car to gain informatics literacy as well as to enable them to apply
models, alternator, voltmeter, various resistors and the use of computers and IT in everyday life and work (Table
conductors, magnetic field, various magnets, magnetic needle, IV). According to the current curriculum for primary schools,
compass, Ersted’s experiment. This is similar in most Serbian this is an optional subject. Besides the classes which are held
schools. Teachers of physics believe that teaching of physics as regular classes, there are also complementary classes held
will be more qualitative, interesting and attractive to pupils if for the pupils who have dispositions for this subject and for
the schools had larger numbers of teaching aids, but teaching those who want to enter the competitions in this field [2].
aids expansion is limited by the budget which is regulated by TABLE IV. INFORMATION TECHNOLOGY AND COMPUTER TECHNIQUE –
the authorised state institutions. CURRICULUM
Apart from the regulated number of regular class hours,
knowledge in this subject is also gained from supplement Course name: Information Technology and Computer Technique
class hours aimed at less successful pupils and from 5th grade Course duration in teaching hours: 36 per school years
complementary class hours in the format of young physicist Computer operating system (14), Text editing (14), Introduction to
groups. These groups are aimed at pupils with special interest multimedia (8)
in physics who would like to promote their high level of 6th grade Course duration in teaching hours: 36 per school years
acquired knowledge by the means of taking part in Text editing (10), Internet (4), Computer graphics (10), Computer
animation (3) Elective modules (9)
competitions whish are supported by the Ministry of
Education and Science of the Republic of Serbia [2]. 7th grade Course duration in teaching hours: 36 per school years
Internet (6), Audio processing (4), Video processing (6), WWW
TABLE III. TECHNICAL AND INFORMATICS EDUCATION - CURRICULUM presentation design (10) Elective modules (10)
Course name: Technical and informatics education 8th grade Course duration in teaching hours: 68 per school years
5th grade Course duration in teaching hours: 62 per school years Computer spreadsheets (10), Elective modules (16), Software
programming (14), Computer networks and WWW presentations (14),
Introduction (2), Traffic (8), Graphic communications (16), From idea to Development of individual projects in the filed of information technology
realization (8), Materials and technologies (12), Energy systems (4), and computer technique (14)
Constructive modeling (12)
6th grade Course duration in teaching hours: 72 per school years Figure 1 shows the percentage of hours per subject Physics
Introduction to the architecture and construction (4), Technical drawing and Information technology and computer technique, which is
(8), Information technology (16), Building materials (4), Energy systems directly related to the study in the field of electrotechnics and
(4), Technical resources in construction (4), Traffic systems (2), Housing
culture (4), Constructive modeling (22), Technical resources in
informatics. It can be seen that the presence of of these
agriculture (4) teaching hours increases from lower to higher grades.
7th grade Course duration in teaching hours: 72 per school years 120%
Introduction to mechanical engineering (2), Technical drawing in
mechanical engineering (8), Information technology (14), Materials (2), 100%
Measurement and control (2), Materials processing technology (4),
Machines and Mechanisms (16), Robotics (2), Energy systems (6), 80%
Constructors modeling - Modules (16) Phisics
8th grade Course duration in teaching hours: 68 per school years 60%
Information technology (16), Electrical materials and installation (10), Technical and IT
Electrical machines and devices (14), Digital electronics (12) 40% education
From idea to realization - Modules (16)
20%
The general goal of the school subject (course) Technical
0%
and informatics education is to get the pupils acquainted with 5th grade 6th grade 7th grade 8th grade
the technical and technologically developed surrounding
through the acquisition of new technical and informatics Fig. 1. The percentage of teaching hours in which pupils learn
knowledge and skills, as well as to enable them to apply the lessons that directly relate to electrotechnics and informatics

278
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Performing quality teaching in the filed of IT and Computer

90.0% 78.7%
Technique requires that each primary school is equipped with 80.0%
specialized IT rooms. This issue will be discussed in more 70.0%
57.7%
60.0%
details later in section IV. 50.0%
46.3% 50.2%
39.0% 34.6% 34.7%
40.0%
III. SOME SURVEY RESULTS AS AN SUPPORT TO THE 30.0%
21.3%
14.6% 15.1%
EVALUATION OF THE PRIMARY CURRICULUM IN THE 20.0%
7.7%
10.0% 0.0%
FIELD OF ELECTROTECHNICS AND INFORMATICS 0.0%
5th grade 6th grade 7th grade 8th grade
There was a survey conducted amongst the primary school
Interesting Partially interesting Uninteresting
pupils in the region of Nish in order to give an approximate
evaluation how satisfied they are with the way and quality of Fig. 2. Pupils’ answers to the question of how interesting for
the acquired knowledge in the field of electro-technics and them are the teaching units in the field of electrotechnics and
informatics which are determined by the current curriculum informatics
for primary schools in Serbia, and also to find out how much 70.0% 58.7% 56.1%
interested they are in the further education in these fields. 60.0% 48.6%
50.0% 42.3%
There were two schools from the closer urban region of Nish 40.0% 30.6% 29.3% 25.0% 28.0%
chosen for this survey, “Cele kula” and “Dusan Radovic”, and 30.0% 23.1% 18.8%
one school from the wider rural region of Nish, “Ivan Goran 20.0%
8.0% 9.7% 9.6% 4.6%
10.0% 2.7% 4.9%
Kovacic” in Nish Spa. In the survey participated 407 pupils in 0.0%
total from 5th to 8th grade, 218 of them from “Cele kula”, 102 5th grade 6th grade 7th grade 8th grade
from “Dusan Radovic” and 87 pupils from “Ivan Goran
Electrotechnics, informatics and computer technique
Kovacic” (observed according to the grades: V-75 pupils, VI-
Only electrotechnics
41 pupils, VII-52 pupils and VII-239 pupils). The pupils
Only informatics and computer technique
answered seven questions which referred to education in the I do not want to listen anything about it
field of electrotechnics and informatics.
In figure 2 are shown the pupils’ answers to the question Fig. 3. Pupils’ answers to the question of what they want to
how interesting for them are the teaching units in the field of listen in the courses of future studies
electrotechnics and informatics. Only a small number of the 46.7% 46.2% 46.9%
46.3%
pupils answered that these units weren’t interested which 50.0% 36.6%
44.2% 40.1%
40.0% 32.0%
speaks a lot about the pupils’ interests for these fields. In 30.0% 21.3% 17.1%
figure 3 the pupils answers show that the number of pupils 20.0% 9.6% 13.0%
10.0%
who want to continue to attend the lectures on electrotechnics 0.0%
and informatics is much larger than the number of those who 5th grade 6th grade 7th grade 8th grade
don’t want that. There is a prevailing wish for attending IT
I want to be an electrical, computer science or informatics engineer
and informatics. Picture 4 shows how big is the interest of
pupils for choosing professions such as engineer of
electronics, informatics and computers. At least 10% of the I do not want to be an electrical, computer science or informatics engineer but I would like
pupils would like that their future profession is engineer of to use in my profession the knowledge I learned about electrotechnics, informatics and
computer technique
electronics, informatics or computers, and the majority of I do not wish that my future profession has anything to do with electrical engineering,
pupils would like to use the knowledge acquired from the informatics or computer science
fields mentioned above in their future jobs. Answers in figure
Fig. 4. Pupils’ answers to the question of which profession
5 show that the majority of the pupils think that there should
they want to choose after completing their education
be more practical work in the field of electrotechnics and
informatics. That indicates that current equipment and 80.0% 65.3%
56.1%
teaching aids aren’t in accordance with the novelties in the 60.0%
40.4% 38.5% 45.2%
curriculum and syllabus. So, the teaching aids should be 40.0%
24.0% 9.4% 22.0% 17.0% 19.2% 26.4% 19.2%
modernized to enable the good quality practical work of 20.0%
1.3% 4.9% 1.9% 9.2%
pupils. Figure 6 shows that a significant majority of pupils
0.0%
chooses elective course IT and computer technique which
5th grade 6th grade 7th grade 8th grade
again speaks about the popularity of IT and computer science.
In figure 7 can be seen that almost one half of the 6th grade Completely satisfactory
pupils attends complementary classes or are members of Partially satisfactory, because it has to be strengthen
science (physics) section, while this number significantly Unsatisfactory because it is uninteresting
declines in the 8th grade. This is explained with the increasing Unsatisfactory because it is too extensive and difficult
number of pupils’ obligations in higher grades. Figure 8
shows that there is an evident interest of pupils for Fig. 5. Pupils’ answers to the question how the practical work
competitions in the field of electrotechnics and informatics, (work in laboratories and workshops, work with computers,
but it should be worked even more on making these conducting experiments) in the field of electrotechnics,
competitions more participated by pupils. informatics and computer technique is satisfactory

279
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

90.0% 80.0% 79.1%

relevant working mechanism which open doors for different
80.0%
67.3% possibilities for pupils’ primary education improvement in the
70.0%
60.0%
58.6% field of electrotechnics and informatics. Amongst the
50.0% 41.4% Yes activities with significant importance we would like to
40.0% 32.7% No mention the following:
30.0%
20.0%
20.0% 20.9%  Harmonizing of educational plans and programmes
10.0% with the EU standards and modern tendencies in the
0.0% field of electrotechnics and informatics
5th grade 6th grade 7th grade 8th grade
 Improving of the teaching aids in the field of
Fig. 6. Percentage of pupils who chose elective course electrotechnics and informatics
Information technology and computer technique  Introducing modern information-communication
90.0% 84.1%
technologies (ICT) to support new methodologies in
80.0% education
70.0%
60.0%
58.5% 59.6%
 Professional development of teachers in the field of
50.0%
40.0%
electrotechnics and informatics through attending
26.9%
30.0%
19.5% 12.1%
regular lessons for continuous professional
20.0% 12.2%
10.0%
9.7% 7.7% 5.8%
2.5% 2.5%
development and enabling them to participate in
0.0% professional conferences and seminars
 Organising seminars, lectures and other activities for
6th grade 7th grade 8th grade

Attend supplementary clasess and members phisics section pupils with an aim to compliment the education in the
Attend supplementary clasess only
fields of electrotechnics and informatics and increase
Members of phisics section
their interest in further education in these particular
Not attend supplementary clasess and not members of phisics section
fields
Fig. 7. Percentage of pupils who attend supplementary classes  Improving the organisation and increasing the general
in physics or who are members of physics section technical and educational level of competition in the
7.0% area of electronics and informatics in order to attract
5.9% 6.4%
6.0% more pupils to participate in these competitions
5.0% 4.4% 4.4% 4.2%
Today’s rapid development of electrotechnics and
4.0%
3.0% 2.5% 2.5% informatics and their increasing presence in all areas of social
1.7% 1.7%
2.0% 1.2% 1.0%1.2%
living requires continuous updating of primary school
1.0% curriculum with the latest news in these fields. Modern plans
0.0% and programmes in high schools across the EU have the
School level Municipal level Regional level Republic level
fastest dynamics of adapting to those changes. One of the
Phisics popular solutions to adapt the plans and programmes to
Technical and informatics education changes in primary schools is their coordination with relevant
Information Technology and Computer Technique programmes in high schools [6]. National Education Council
in the Republic of Serbia has a vital role in continuous
Fig. 8. Percentage of pupils who participated at different monitoring and analyzing of the state of education on all
levels of competition (overview is given by the subjects and levels and its coordination with the European principles and
levels of competition) values. In accordance with this the council makes decisions on
changes and coordination of the primary schools plans and
IV. OPPORTUNITIES FOR IMPROVING THE PRIMARY programmes in educational areas of interest [3]. Primary
EDUCATION QUALITY IN THE FIELD OF schools participation in TEMPUS projects, financed and
approved by the EU, is a great opportunity to get help and
ELECTROTECHNICS AND INFORMATICS support for the process of modernizing primary schools plans
and programmes through their coordination with high schools
Beside the Ministry of Education and Science of the programmes [6].
Republic of Serbia which is the main and responsible The improvement of quality of education in primary
institution for implementing of all levels of education, in schools is unthinkable without introduction of certain
Republic of Serbia today there are a few more important state educational standards which are in coordination with the EU
institutions whose work have an influence on pupils’ primary standards. During the last few years the Republic of Serbia
education. They are [3-5]: has been making huge efforts to incorporate and implement
- National education council of the Republic of Serbia such standards in primary education. As a direct result of
- Center for improving education and schooling those efforts, in 2009, National Education Council has passed
- Center for assessment of education quality and schooling an act “the educational standards for the end of compulsory
The work of these institutions is coordinated and they work education”. Those standards originated from working on
very closely with each other. These institutions have the project by the Ministry of Education and s Sport of the
Republic of Serbia named: “Development of schooling in the

280
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

Republic of Serbia” and its project component “Standard

development and assessment” brought to life by the Center for
assessment of education quality and schooling [5]. The basic
problem that occurred after adopting these standards was the
fact that they only covered 10 subjects in primary education.
As far as the subjects containing electrotechnics and
informatics were concerned these standards included only the
physics while excluding other subjects such as technical and
informatics education and information technology and
computer technique. In the light of this obvious shortage,
today there have been efforts to develop adequate standards of
achievements for these subjects so they can give a significant
contribution in increasing the quality of education in these
subjects [7].
Due to its specific nature implementation of high quality
education in the area of electrotechnics and informatics is Fig 9. Modern digital cabinet in primary school “Cele kula“ in
difficult to achieve if primary schools do not have modern Nis realised through a project “Digital School” [9]
teaching aids required for this particular area. Significant
number of schools in the Republic of Serbia is facing a Implementation of digital cabinets in primary schools is
problem of using dated and inadequate teaching aids needed widely opening doors for introduction of
to achieve educational targets in the area of electrotechnics. information/communication technologies in teaching which is
Constant repairing and upgrading of the old equipment and essential for increasing education to a higher level particularly
buying of the new one which will be in accordance with new in the area of computing and informatics. Working within
modern curricula, requires continuous funding. This is computer networks, access to modern internet services, using
particularly relevant for the equipment used in laboratory multimedia and other information/ communication
experiments and training. Finding significant financial means technologies are opening possibilities for implementation of
in the time of financial crisis and limited and restricted methodical innovations in teaching such as learning based on
primary schools’ budgets may be an impossible task. One of working on project and problem orientated teaching, also
the ways to resolve this problem, which has been given more known as “problem solving teaching” [11]. These innovations
attention lately, is to introduce the concept of digital school, are very suitable for improving teaching in subjects such as
where teaching is delivered in the space of digital cabinets technical and information education and information
(classrooms equipped with computers and other ICT technology and computer technique because the knowledge
equipment) [9]. According to this concept, some adequate required is based on active participation of pupils in solving
additional interactive multimedia books will be used to the real life problems through team work on chosen subjects.
support the increase of quality of schooling in physics, On the other side, the implementation of ICT in teaching has
technics and informatics [10]. In accordance with that, instead reduced the problem of shortage of adequate literature on
of buying a complex and expensive equipment for practical problem solving teaching as the information is available on
training in electrotechnics, an adequate and cheaper the Internet [11]. With the objective to implement internet
interactive programme will be used to simulate the same services to the majority of schools the Ministry of Education
experiment on the computer screen. Using this method will and Science of the Republic of Serbia has signed a contract
make it possible for larger number of experiments to be done with the one of the biggest internet providers in Serbia,
without the need to expand the infrastructure of the Telecom Serbia, to install ADSL in primary and secondary
laboratories as practical training will be delivered in the schools. The contract aims at installing ADSL in all primary
virtual lab. The situation is quite similar when it comes to schools in Serbia until 2013 with the speed of 16Mbps which
training in the area of informatics, however we will talk about is sufficient for the quality internet access.
this a bit later in the part relating to the introduction of ICT in Primary teachers’ training to deliver schooling based on the
primary schools teaching. new plans and programmes and application of new methods as
At the beginning of 2011 huge support was given to the well as capability of constant improvement of their knowledge
concept of digital school, introduction of ICT in teaching and with new scientific resources results in their chosen subject is
increasing the quality of education in primary schools in one more important reason that influences the quality of
general by stating the project named “Digital School” which is education in electrotechnics and informatics. According to the
managed and funded by the Ministry of Telecommunications legal regulation about continuous professional development,
and Information Society of the Republic of Serbia. The aim of teachers’ training qualification, tutor and expert advisor,
this project is to fully equip 2910 schools (83%) with digital accepted by the Ministry of Education and Science of the
cabinets in the Republic of Serbia [9]. Including March 2012 Republic of Serbia, a teacher is legally obliged to attend 100
digital cabinets were fully established in 2808 schools (80%) hours programme over 5 years while at least 60 hours are
in the Republic of Serbia. One of the schools with fully dedicated to the compulsory and up to 40 hours to elective
equipped digital cabinet is primary school “Cele kula“ in Nis curriculum [12]. Every year Center for improvement of
(Fig. 9). education and teaching approves a catalogue of programmes
for professional development for teachers [4]. Taking into

281
I C E S T 2012 28-30 JUNE, 2012, VELIKO TARNOVO, BULGARIA

consideration that the Minister of Education and Science of school aids should be modernized in accordance with the
the Republic of Serbia has given priority to information – curriculum and syllabus. Serbian Government is making great
communication technologies in the last 3years, programmes effort with its ministries to enable such modernization of
for professional development in the field of technology and school aids in primary schools. An example for this is the
informatics are significantly more present in the catalogues project “Digital School”. As the quality of education depends
and are available to teachers. Yet another important way of on how much the teachers are trained to transfer modern
teachers’ professional development is their participation in knowledge to pupils, special attention is paid to teachers’
professional conferences and seminars. One example of training in the field of electrotechnics and informatics so that
professional conference which is important for primary school they are constantly involved in special trainings, attending
teachers’ development who are teaching electrotechnics and obligatory classes and being able to participate conferences
informatics is International Conference on Technics and and seminars. Today there are evident efforts being made in
Informatics – TIO organised by the Faculty of Technical order to adopt appropriate educational standards for all school
Science in Cacak [13]. subjects (courses) which are in accordance with EU standards
With the aim to make electrotechnics and informatics more which will lead to the further improvement in quality of
popular among primary school pupils, as well as enable them primary school education. Special attention is paid to
to widen their knowledge in this area, they are given an coordination of syllabus and curriculum of primary schools
opportunity to attend professional seminars, summer schools, with those in secondary schools and faculties. In the end, it is
lectures and workshops. “Summer school of science” hugely very important to point out that there are great efforts made to
popular among primary school pupils in Serbia, supported by organize competitions, seminars, lectures and other activities
UNESCO, takes place in Petnica and it can offer rich and for the pupils who want to complement the education in the
interesting programmes in the field of electronics and field of electrotechnics and informatics, which surely leads to
computing [14]. Primary school “Cele Kula” in Nis as well as an increase in pupils’ interest for further education in these
significant number of other schools in Serbia organise a very fields.
popular “Science Fair”where numerous experiments and REFERENCES
examples of practical teaching are demonstrated to pupils.
Also, the numbers of joint workshops for primary and [1] Education Programme for Primary Schools (School year
secondary pupils have been increasing in order to motivate the 2011/2012), Serbian Ministry of Education and Science, 2011.
[2] Competitions and Festivals Calendar for Primary School Pupils
primary school pupils choose some of the electrotechnics
(School year 2011/2012), Serbian Ministry of Education and
modules in their secondary education. Science, 2011, http://www.mpn.gov.rs/prosveta/page.php?page=156.
Improvement of organisation and increase of the general [3] National Education Council of the Republic of Serbia
technical and educational level of competition in the field of http://www.nps.gov.rs
electrotechnics and informatics is yet another way of [4] National Institute for Education and Upbringing Improvement,
widening pupils’ knowledge which has the impact on the http://www.zuov.gov.rs/
increase of the quality of education in general. Ministry of [5] National Institute for Education and Upbringing Quality
Education and Science has been making efforts to find Evaluation, http://www.ceo.edu.rs
additional finances for improving the organisation of [6] TEMPUS Project JP 516762-2011: "Harmonization and
Modernization of the Curriculum for Primary Teacher
competitions and adequately rewarding the achievements in
Education", http://www.tempus.ac.rs/projects-tempus/view/797/49/
order to increase the number of pupils – competitors. [7] Education Standards for End of Compulsory Education, Serbian
Engaging financial sponsors outside of education can help Ministry of Education and Science, National Institute for
resolving this problem. Education and Upbringing Quality Evaluation, Belgrade, 2009
http://www.ceo.edu.rs/images/stories/publikacije/Obrazovni%20standar
di%202009.pdf
V. CONCLUSION [8] M. Sanader, "Educational Standards of Achievements in the
New educational plans and programmes for primary Course on Technology and Informatics in Primary Schools", 3rd
schools which are regulated by the Ministry of Education and International Conference on Technics and Informatics in
Science of the Republic of Serbia have been coordinated with Education - TIO 2010, Conference Proceedings, pp.176-192,
Cacak, Serbia, 2010.
modern tendencies in the field of electrotechnics and
[9] Project ”Digital School”, Serbian Ministry of Culture, Media
informatics. They offer acquisition of the latest knowledge in and Information Society, Serbian Digital Agenda,
these fields which are developing very fast today and a special http://www.digitalnaagenda.gov.rs
attention is paid to the units referring to computers and [10] Kvark Media–Multimedia workshop, http://www.kvarkmedia.co.rs
informatics. A huge interest of pupils for studying and [11] M. Paroskaj, M. Mitrovic, " Use of ICT in Modernize and
acquiring knowledge from the field of electro-technics and Bringing Up Quality of Technical and Informatical Education in
informatics, which is shown in the conducted survey, justifies Primary School", 3rd International Conference on Technics and
these changes in the primary schools curriculum. These Informatics in Education - TIO 2010, Conference Proceedings,
curriculum and syllabus have established a foundation for pp.468-474, Cacak, Serbia, 2010.
[12] Rulebook of continuous professional training and acquiring of
acquiring high-quality knowledge in the mentioned fields
teacher, educator and professional associate titles, Sl. glasnik
which represents a beginning of education for successful RS (br. 14/2004, 56/2005)
engineers. The survey shows that a huge number of pupils [13] International Conference on Technics and Informatics in
demand the increase of practical work in the fields of electro- Education – TIO, http://www.tfc.kg.ac.rs/tio2012/
technics and computers, which inevitably leads to the fact that [14] Petnica Science Center, http://www.petnica.rs

282
 AUTHOR INDEX
A D
Acevska V. ............................ 387, 447
Damnjanovic M.............................163
Acevski I................................ 387, 447
Demirev V.......................................21
Acevski N. .....................................235
Denić D................................. 259, 269
Aćimović S. ...................................407
Denishev K....................................495
Agatonovic M. ..............................311
Despotović V......................... 573, 577
Aleksandrova M.................... 495, 509
Devic S.................................. 463, 467
Aleksieva V. ..................................439
Dimitrijevic T...................................29
Alexiev V.........................................89
Dimitrov B. ........................... 544, 548
Altimirski E......................................37
Dimitrov D. ...................................243
Andonova A. .................................541
Dimitrov K................ 55, 307, 599, 603
Angelov K.................................. 37, 51
Dimitrov V. ........................... 283, 567
Angelov K........................................85
Dimitrova E...................................571
Angelov P.............................. 521, 525
Dimkina E. ....................................371
Antolović I. ...................................212
Dimov A........................................485
Antonov S. ....................................375
Dimova R. .....................................283
Apostolov P. .................................117
Dimovski T. .....................................43
Aprahamian B. ..............................564
Djosic S. ........................................163
Arnaudov R...................................255
Djugova A. ......................................59
Arsić M. ................................ 259, 269
Djukovic M....................................463
Asenov O. ............................. 359, 428
Dobrev D. .....................................307
Atanasov I............................. 327, 418
Dobrikov G....................................509
Atanasovski M. .............................243
Dochev I. ......................................595
Atlagic B. .............................. 463, 467
Docheva L. ............................ 367, 595
Avramova N..................................403
Dochkova-Todorova I. ...................206
B Dokoski G. ....................................471
Dončov N.............................. 299, 311
Babic D. ........................................135 Đorđević G....................................105
Bakardjieva T. ...............................167 Đorđević M...................................212
Balabanov G. ..................................89 Drača D...........................................77
Balzhiev P. ....................................255 Draganov I. ........................... 143, 155
Bankov K.........................................71 Draganov V. ..................................265
Banković B. ........................... 552, 556 Dzhakov R.....................................121
Békefi Á. .........................................39 E
Bekjarski A............................ 367, 603 Eftimov T. .....................................255
Blagojević D. .................................303 Eremieva M. ......................... 159, 263
Bock W. ........................................255
Bodurov G. ...................................509 F
Bogdanović N................................303 Fahlberg-Stojanovska L. ................447
Bojchev D. ....................................471 Fehér A...........................................39
Bonev B. ................................... 25, 37 Fratu O. ...........................1, 5, 93, 273
Brodić D.........................151, 573, 577 Furkov G. ......................................232
Brusev T........................................513 G
Bucsa I. .............................................5
Gadjeva E.............................. 228, 505
C Gajić D. ................................. 216, 190
Gajic M. ........................................463
Cernat C................................ 175, 273 Galabov M. ...................................435
Cherneva G....................371, 493, 571 Gechev M. ......................................51
Cholakova I. .......................... 499, 509 Georgiev M.....................................71
Čičević S........................................407 Georgieva T. .................................343
Ćirić D................................... 113, 131 Georgieva V. .................................247
Craciunescu R. .............................. 1, 5 Gerasimov K. ................................239
Cvetkovic S. ..................................277 Gesheva K.....................................509
Cvetković T. ..................................299 Gjorgjievska S. ..............................194
Goleva R. ........................................89 Lozanovska A. ...............................171
Gorecan Z. ............................ 463, 467 Lubich L. .......................................287
Gradinarova B...............................167 Lukić J................................... 259, 269
Guliashki V....................................208
M
H
Malecic A......................................585
Hadjidimitrov A.............................247
Halunga S......................................1, 5 Malenko M. ..................................194
Hristov G............................... 351, 355 Manev S................................ 319, 321
Manojlović P. ................................295
I Marinov A.....................................548
Ilarionov R. ...................................451 Marinova G........................... 479, 482
Iliev G. ..........................................323 Markova G....................................399
Iliev I................................47, 319, 321 Markova V. ...................................265
Iliev Ivo.........................................502 Markovic D. ..................................331
Iliev M. ................................. 351, 355 Marković I.....................................178
Ivanov H. ........................................55 Martinovic L..................................467
Ivanov H. ......................................435 Matijasevic J. ................................431
Ivanova E. .......................................97 Mićić Z. .........................................295
Mihajlović V.......................... 212, 220
J Mihić D. ........................................212
Jakimovska D. ...............................194 Mihov G........................................383
Jakimovski G. ................................194 Mihov Y. .......................................101
Jakšić B. ........................................315 Mijoski K.......................................235
Janković D.....................................178 Mijoski T. ......................................235
Janković M....................................113 Mikarovski G.................................411
Janković S. ....................................407 Mikhov M. ....................................251
Jankulovska M. .............................387 Milenković A. ................................178
Jelenković M.................................131 Milić D. ................................. 315, 335
Jevtic M. .......................................163 Milijic M. ......................................291
Jokovic J.................................. 29, 299 Milinković S. .................................407
Jordanova L. ......................... 599, 307 Militaru T.............................. 175, 273
Jovanovic Z. ..................................331 Milivojević D. .................151, 573, 577
Milivojević Z..................................151
K Miljković G............................ 259, 269
Kanev J. ..........................................74 Miljković V. ...................................577
Karailiev H. ...................................451 Milosavljević A..............................220
Karailiev V.....................................451 Milosavljevic S. .............................335
Karova M. ............................. 403, 475 Milovanović B. .................29, 299, 311
Katsov R........................................493 Milovanović D...............................303
Kirilov L.........................................208 Milovanovic I. ...............................291
Kitov C. .........................................347 Milutinović V. ...............................299
Kolev G. ................................ 509, 495 Mironov R............................. 139, 603
Kolev N. ..........................................33 Mirtchev S. .....................................89
Kolev S.................................. 319, 321 Mitrevski P......................................43
Koleva E................................ 159, 263 Mitrović N............................. 552, 556
Koleva P........................................359 Mitrović S. ....................................407
Kopta A.........................................295 Mitsev T....................... 25, 33, 55, 599
Kostić V................................. 552, 556 Mladenović S. ...............................407
Kostov N. .......................395, 479, 482
Kotevski A.....................................411 N
Kovačević M.......................... 198, 220
Kovacheva M. ...............................224 Nagy L.............................................59
Krastev G. .....................................422 Nagy S. ...........................................39
Krasteva I......................................485 Necov B. .........................................71
Nedelchev M. .................................47
L Nedelkovski I. ...............................387
Lehtinen V. ...................................135 Nenov I.........................................493
Nenova M.....................................339 Roganović M. ............................... 198
Nikolić B. ......................................105 Ruzin I.......................................... 171
Nikolov B. ............................. 395, 482
S
Nikolov G........................529, 544,581
Nikolov N......................................425 Sabeva V. .............................. 159, 263
Nikolova B. ....................513, 529, 581 Sadinov S. .................................74, 85
Nikolova K.....................................125 Sechkova T................................... 143
Nikolova M. .......................... 159, 263 Sekulović N. ................................... 77
Novakov P.....................................283 Serafimov N. ................................ 513
Shtarbakov V................................ 564
O
Shupak M....................................... 25
Obradović D..................................295 Simeonov I. .................................. 121
Simić M........................................ 259
P
Sirakov E. ..................................... 129
Pacheco C. ......................................63 Sit L.............................................. 311
Pacheco de Carvalho J.....................63 Slavov M. ..................................... 109
Panagiev O.................................. 9, 13 Smiljakovic V................................ 533
Panajotović A..................................77 Spalević P..................................... 315
Pandiev I............................... 224, 517 Spalevic Z..................................... 431
Pargovski J. ...................................171 Spasova V. ................................... 502
Pavlov G. .............................. 371, 493 Stančić I. ...................................... 295
Pavlov M............................... 573, 577 Stanimirovic A.............................. 202
Pavlović N.....................................407 Stankovic A. ................................. 186
Penchev P.....................................109 Stanković D. .......................... 182, 186
Pencheva E. ..................................418 Stanković R. ................................. 190
Penev I.................................. 403, 475 Stankovic Z................... 277, 291, 311
Perić D............................................81 Stavru S........................................ 485
Perić M. ..........................................81 Stefanov T.................................... 455
Pesovic U. .....................................331 Stefanova M................................. 428
Petkov E........................................459 Stefanović Č. ................................ 182
Petkov P. ......................................239 Stefanovic D................................. 335
Petkova Y......................................403 Stefanovic H................................. 335
Petković M....................................105 Stoimenov L. ................................ 202
Petronijević M. ..................... 552, 556 Stojanović D. ................................ 198
Petrov A........................................525 Stojanović M. ............................... 560
Petrova-Antonova D......................485 Stoyanov G................................... 125
Petrović M. ...................................315 Stratev A. ..................................... 232
Pleshkova-Bekiarska S...................375 Streblau M. .................................. 564
Poenaru V............................. 175, 273 Suciu G.................................. 175, 273
Popova A. .....................................147 Šunjevarić M. ................................. 81
Popović M.......................................81
Poulkov V......................................359 T
Predić B. .......................................198
Tabakov S..................................... 502
Tahrilov H. .................... 544, 548, 564
R
Tasić D. ........................................ 560
Radev D. .........................................97 Tasić V. ................................. 573, 577
Radic J. ...........................................59 Tatić D. ................................. 182, 186
Radmanović M..............................216 Temelkovski I. .............................. 315
Radonov R. ...................................593 Tentov A. .............................. 194, 471
Raev R. ...........................................97 Todoran G............................. 175, 273
Rančić D.........................212, 220, 431 Todorov M.. ................................. 529
Randjic S.......................................331 Todorov V.. .................................. 265
Rankovska V. ................................489 Todorova M.. ............................... 425
Reis A. ............................................63 Todorović B.................................... 81
Ristić A..........................................560 Tomic D................................. 463, 467
Tomić S.........................................131 Videkov V.............................. 232, 593
Tomov Y. ......................................323 Videnovic-Misic M.......................... 59
Trifonov T. ....................................121 Vulović D...................................... 202
Trifonov V.....................................418 Vulpe A. ......................................... 93
Trifonova T. ..................................265 Y
Trpezanovski L. .............................243
Tsankov B. ....................................101 Yordanov H. ................................... 17
Tsenov A.........................................67 Yordanova S. ................. 395, 479, 482
Tsvetkova I............................ 351, 355 Z
Tаrpov I. .......................................493
Zahariev P. ............................ 351, 355
V Zdravković J. ................................ 131
Zhilevski M................................... 251
Valchanov H..................................443
Zhivomirov H......................... 129, 379
Valkov G. ......................................228
Živanović D................................... 269
Varbanova N...................................85
Zivanovic Z. .................................. 533
Vasić B..........................................105
Zwick T......................................... 311
Veiga H. ..........................................63