Acta Universitatis Sapientiae

Electrical and Mechanical

Engineering
Volume 2, 2010

Sapientia Hungarian University of Transylvania

Scientia Publishing House
 Contents

Industrial Electronics & Control Systems

K. György, L. Dávid
Comparative Analysis of Model Predictive Control Structures .................... 5
Cs. Szabó, M. Imecs, I. I. Incze
Synchronous Motor Drive at Maximum Power Factor with Double
Field-Orientation.............................................................................................. 16
I. I. Incze, A. Negrea, M. Imecs, Cs. Szabó
Incremental Encoder Based Position and Speed Identification:
Modeling and Simulation ................................................................................ 27
D. Fodor
Aluminium Electrolytic Capacitor Research and Development
Time Optimization Based on a Measurement Automation System ............. 40

Computer Science

L. Haţegan, P. Haller
Framework for Modeling, Verification and Implementation of
Real-Time Applications ................................................................................... 51
M. Muji
Application Development in Database-Driven Information Systems ......... 63
A. Aszalos, J. Domokos, T. Vajda, S. T. Brassai, L. Dávid
Exambrev - Integrated System for Patent Application ............................... 73

3
Telecommunications

L. Huszár, Cs. Simon, M. Maliosz

Inter-Domain Traffic Engineering for Balanced Network Load ................. 87
L. Szilágyi, T. Cinkler, Z. Csernátony
Energy-Efficient Networking: An Overview ................................................. 99
V. Cazacu, L. Cobârzan, D. Robu, F. Sandu
Localization of the Mobile Calls Based on SS7 Information and
Using Web Mapping Service ......................................................................... 114

Signal Processing

L. F. Márton, L. Szabó, M. Antal, K. György

Analysis of Neuroelectric Oscillations of the Scalp EEG Signals ............. 123
Z. Germán-Salló
Nonlinear Filtering in ECG Signal Denoising ............................................. 136

Mechatronics & Industry Applications

D. Biró, S. Papp, L. Jakab-Farkas

Microstructural Modification of (Ti1-xAlxSiy)N Thin Film Coatings
as a Function of Nitrogen Concentration ..................................................... 146
D. Hollanda, M. Máté
On Some Pecularities of Paloid Bevel Gear Worm-Hobs ................. 159
Z. Forgó
Kinematic Analysis of a 6 DOF 3-PRRS Parallel Manipulator ................ 166

Acknowledgement .......................................................................................... 177

4
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 5-15

Comparative Analysis of Model Predictive Control

Structures
Katalin GYÖRGY, László DÁVID
Department of Electrical Engineering, Faculty of Technical and Human Sciences,
Sapientia University, Tg. Mureş,
e-mail: kgyorgy@ms.sapientia.ro, ldavid@ms.sapientia.ro

Manuscript received October 28, 2010; revised November 03, 2010.

Abstract: The industrial implementation of advanced multivariable control

techniques like Model Predictive Control (MPC) is complex, time consuming and
therefore it is expensive. Nowdays it is a popular research area to reduce the complexity
of the MPC algorithm while preserving the control performance. This problem could be
solvable with implementation of the MPC solution in a distributed way. The main idea
of this work is to develop simple software agents that can be easily implemented in low
cost embedded systems. Each one of these software agents solves the problem of
finding one of the control actions with parallel computational facilities. This paper
presents at first some general and theoretical considerations about centralized and
distributed model predictive algorithms. The comparison between these algorithms is
made using numerical simulation of these methods for a multiple input and multiple
output theoretical linear discrete-time system. The comparision is possible to be made
from the point of wiew of the normalized absolute reference tracking error. There is
described a possible implementation of the distributed MPC algorithm using Matlab
Simulink environment. It is important to notice that the algorithm to be solved by each
software agent while computing the control action is much simpler than the one to be
solved by the centralized algorithm.

Keywords: optimal control, cost function, model predictive control, distributed

control, prediction horizon, control horizon, Jacobi over-relaxation method, linear
constrained optimal programming problem.
6 K. György, L. Dávid

1. Introduction
The model predictive control (MPC), – also called receding horizon control
(RHC) – is the most important advanced control technique which has been very
successful in practical applications, where the control signal can be obtained by
solving a discrete-time optimal control problem over a finite horizon. The most
important advantage of the MPC algorithms is the fact that they have the unique
ability to take into account constraints imposed on process inputs, process state
variables and outputs, which usually determine the quality, the efficiency, and
safety of production. Implementation of centralized state space (SS) MPC
algorithms is becoming an important issue for different multivariable industrial
processes. The main idea of our work is to develop a multi-agent software that
can be implemented in low cost embedded systems, with parallel computational
facilities. These software agents are valid for a default model and can be
multiplied and customized according to the control horizon. Each one solves the
problem of finding one of the control actions. This procedure is repeated several
times before the control action values are delivered to the final control elements.
An agent, as an executive, has to know general information about the system
and some others which are specific of his own department. It is important to
notice that the algorithm to be solved by each agent while computing its control
action is much simpler than the one to be solved by the centralized solution.
Previous works on distributed MPC [2], [3], [4], [6] use a wide variety of
approaches, including multi-loop ideas, decentralized computation using
standard coordination techniques, robustness to the actions of others, penalty
functions, and partial grouping of computations. The key point is that, when
decisions are made in a decentralized fashion, the actions of each subsystem
must be consistent with those of the other subsystems, so that decisions taken
independently do not lead to a violation of the coupling constraints. The
decentralization of the control becomes more complex when disturbances act on
the subsystems making the prediction of future behavior uncertain.
We will analyze how the overall performance of a distributed system is
influenced if one or more agents – except the coordinating agent –, fail or
obviously underperforms from some reasons. The objective is to solve SS-MPC
problems with locally relevant variables, costs, and constraints, but without
solving a centralized SS-MPC problem. The coordinated distributed
computations solve an equivalent centralized SS-MPC problem. This means
that properties that can be proven for the equivalent centralized MPC problem
(e.g., stability, robustness) are valid to the above distributed SS-MPC
implementation. The significance of the proposed distributed control scheme is
that it reduces the computational requirements in complex large-scale systems
and it makes possible the development of fault tolerant control systems.
 Comparative Analysis of Model Predictive Control Structures 7

2. Centralized State Space Model Predictive Control

All the MPC algorithms possess common elements and different options can
be chosen for each one of these elements: prediction model, objective function
and algorithms for obtaining the control law. In this paper the process model is
a discrete input – state – output relationship:
x k +1 = Φ ⋅ x k + Γ ⋅ u k
, (1)
yk = C ⋅ xk

where xk is the state vector (n x 1), uk is the input vector (m x 1), yk is the output
vector (p x 1), and Φ, Γ and C are the matrices of the system. If these matrices
(parameters) are unknown, we have to implement a system identification
module in the control algorithm.
The centralized model predictive algorithm looks for the vector ∆Uk that
minimizes a cost function represented by the scalar J, defined as:

(
J (∆U k ) = Y k − Y ref
k )
T
(
⋅ Q ⋅ Y − Y ref
k )+ ∆U Tk ⋅ R ⋅ ∆U k , (2)

where Y ref
k is the vector with the future references, Y k is the vector of the
predictions of the controlled variables (output signals), ∆U k is a vector of
future variations of the control signal, Q is a diagonal matrix with weights for
set-point following enforcement, R is a diagonal matrix with weights for control
action changes. If the prediction horizon is N and the control horizon is Nc these
vectors and matrices are [1]:
y   y ref   ∆u k k 
 k +1 k   k +1 k 
 
Y k =    , Y ref k =   , ∆U k =   , (3)
y   ref   
 k + N k   y k + N k  ∆u k + N c −1 k 

Q1 0  0   R0 0  0 
0 Q  0  0 R1  0 
Q= 2
, R= . (4)
          
   
 0 0  Q   R N c −1 
N  0 0

An incremental state space model can be used if the model input is the
control increment ∆u k = u k − u k −1 . The following representation is obtained for
predictions:
8 K. György, L. Dávid

Yk = Φ * ⋅ x k + Γ * ⋅ u k −1 + G y ⋅ ∆U k , (5)
where
 C ⋅Γ   
    
 C ⋅Φ   N c −1   C ⋅Γ  0 
   
    ∑ C ⋅ Φ i
⋅ Γ 
 
N
  
.
 C ⋅Φ N c   iN=0   c 
Φ =
*
C ⋅ Φ N c +1  Γ *
= 
 ∑
c

⋅ Φ i
⋅ Γ


G y = ∑
 C ⋅Φ ⋅ Γ
i
 C ⋅ (Φ ⋅ Γ + Γ )
  C  i =0 
    i =0    
 N      N −1  N − Nc 
 C ⋅Φ   −   C ⋅Φ i ⋅ Γ C ⋅Φ ⋅ Γ 
N 1
∑
 C ⋅Φ ⋅ Γ 
i
∑  ∑ i

 i =0   i =0 i =0 
(6)

The cost function can be written as:

J (∆U k ) =
1
∆U k T ⋅ H ⋅ ∆U k + f T ⋅ ∆U k + const , (7)
2
where
(
H = 2 ⋅ G yT ⋅ Q ⋅ G y + R , )
f = −2 ⋅ G yT ⋅ Q ⋅ E k , (8)

k − Φ ⋅ x k − Γ ⋅ u k −1 .
E k = Y ref * *

For problems without constraints the centralized model predictive control

determines the vector ∆Uk that makes
∂J (∆U k )
∂ (∆U k )
= 0 ⇒ ∆U opt
1
k = ⋅ H +H
2
( T −1
) ( −1
)
⋅ f = G Ty ⋅ Q ⋅ G y + R ⋅ G Ty ⋅ Q ⋅ E k (9)

It is to be mentioned that only the first control action is taken at each instant,
and the procedure is repeated for the next control decision in a receding horizon
fashion.

3. Distributed State Space Model Predictive Control

The implementation of distributed model predictive control needs to search
for ∆u k / k , ∆u k +1 / k ,, ∆u k + N c −1 / k [5], [7], that makes the
∂J (∆U k )
=0 . (10)
∂ ( ∆u g / k )
for k ≤ g ≤ k + N c − 1 .
If the cost function is written as:
 Comparative Analysis of Model Predictive Control Structures 9

( )
J (∆U k ) = ∆U k T ⋅ G Ty ⋅ Q ⋅ G y + R ⋅ ∆U k − 2 ⋅ E Tk ⋅ Q ⋅ G y ⋅ ∆U k + E Tk ⋅ Q ⋅ E k
(11)
then the first order optimality condition can be determined in the following
way:
∂J (∆U k )
∂ ( ∆u g / k )
[
= 2 ⋅ GTy ⋅ Q ⋅ G y + R ]
g − k +1, g − k +1
⋅ ∆u g , k −

∑  [Q ⋅ G y ]Ti, g − k +1 ⋅ [E k ]i  +

N
− 2⋅ (12)
i =1

∑ (([GTy ⋅ Q ⋅ G y + R]Ti +1, g − k +1 + [GTy ⋅ Q ⋅ G y + R]g ,i +1 )∆u k +i,k )

N c −1
+
i =0
i≠ g −k
The variation of input signal is

( [ ]
∆u g / k = 2 ⋅ G Ty ⋅ Q ⋅ G y + R g − k +1, g − k +1 )
−1  N
∑[
⋅  2 ⋅  Q ⋅ G y

]Ti, g − k +1 ⋅ [E k ]i  −
 i =1 
 (13)

∑ (( ) )
N c −1
− ⋅Q ⋅Gy +
[G Ty ⋅ Q ⋅ G y + R] g ,i +1 ∆u k + i , k 
R]Ti +1, g − k +1 + [G Ty

i =0 
i≠ g −k 
The first value of every ∆ug/k is only an approximation since it depends on
the other ∆ui+k/k values ( i ≠ g − k ). It should be noticed that the computation
burden to obtain ∆ug/k is much smaller than the one needed to compute the
whole vector ∆Uk. As already discussed, in this distributed approach, the vector
∆Uk is determined by software agents using a combination of repeated
computation of ∆ug/k and exchange of information.
The equation (13) can be written in the following general form:

∑ (A j +1,i +1 ⋅ ∆u nk −+1i / k + B j +1 ),
N c −1
∆u nk + j / k = (14)
i =0
i≠ j
where 0 ≤ j ≤ N c − 1 , the matrices Ai,j have the dimension m × m and vector Bi
has the dimension m × 1, where m is the number of inputs. Matrix Ai,i is zero. A
centralized expression for ∆Uk using equation (14) can be written as:
10 K. György, L. Dávid

n n −1
 ∆u k k   ∆u k k   B1 
   0 A1, 2  A1, Nc   
A   ∆u k +1 / k   
 ∆u k +1 / k  0  A  B2 
 =
2,1 2, Nc 
  ⋅   +   (15)
          B 
∆u k + N c −1 k  

 ∆u k + N c −1 k 
 Nc 
   Nc ,1
A A Nc , 2 0     

which, in a compact form becomes
∆U nk = A ⋅ ∆U nk −1 + B . (16)
The convergence of the ∆Uk vectors to their true values has to be assured for
a reliable application. For unconstrained applications the results obtained in the
field of distributed computation can be used [5]. The Jacobi over-relaxation
approach is adopted here by recomputing ∆Uk as a linear combination of the
value computed using equation (16) and the value obtained in the previous
iteration,
∆U nk , filtered = ( I − diag (α)) ⋅ ∆U nk + diag (α) ⋅ ∆U nk −1 (17)
where α is a vector of the filter parameters. Applying the filter according to
equation (16), it results:
∆U nk = ((I − diag (α ) ) ⋅ A + diag (α ) ) ⋅ ∆U nk −1 + (I − diag (α ) ) ⋅ B =
(18)
= A(α ) ⋅ ∆U nk −1 + (I − diag (α ) ) ⋅ B
A sufficient condition for convergence of the iterative process is to have
A(α ) < 1 for α ∈ (0,1) . The search for a filter vector α which minimizes
A(α ) can be reduced to a linear constrained optimal programming problem.

4. Numerical simulation
This section presents the application of the centralized and distributed model
predictive algorithm to a multiple input and a multiple output theoretical system
which is characterized by following state space model:
 0.7 0 0.1 0   4 0
 0 − 0.5 0.2 0   3 9
x k +1 =  ⋅ xk +   ⋅u
 0 0.01 0.1 0  − 10 1  k
    (19)
0.01 0 0 − 0.5  0 2
1 0 0 0 
yk =   ⋅ xk
0 0 1 0 
where
 Comparative Analysis of Model Predictive Control Structures 11

[
x k = x1,k x 2, k x3,k x 4, k ]T , u k = [u1,k u2,k ]T and [
y = y1,k y2,k
k
]T .
For both algorithms the Simulink models have been built and the following
parameters were used for both simulations:
N = 4; N c = 3; R = 0.1 ⋅ I 2 Q = 10 ⋅ I 2 (20)
The Simulink diagram of the centralized predictive control is shown in
Fig. 1, where the “Centralized_MPC_control” subsystem contains one complex
S-function module for centralized control algorithm. The Simulink model of the
distributed predictive algorithm is shown in Fig. 2.
The “Distributed_MPC_control” subsystem is presented separately in Fig. 3,
where the three interdependent modules for calculating ∆u1 / k , ∆u 2 / k , ∆u 3 / k
can be observed. The structure of these modules is one and the same, just the
input signals and parameters are different.
3
2
u
2
2{2} y 1
1 y _ref 2
u u y(k)
y_ref(k) x 2 4
4 x_estim 2
4
Centralized_MPC_control x(k)
Process

Figure 1: Simulink diagram for numerical simulation of the

centralized predictive control.

y _ref (k)

y _ref (N*p) y ref y

Outputs/References (y/y_ref)

u u
x0 x_est
Reference_signal
Distributed_MPC_control Process States(x)

Control_signals(u)

Figure 2: Simulink diagram for numerical simulation of the

distributed predictive control.
12 K. György, L. Dávid

B1

B2 duk1 1
B3 B1
2
Aj1 A1 for { ... }
B2

duk1
Aj2 A2 3
du(k-1)
du(k-1) du(k-1) 1
Aj3 A3 duk B3 Bj du3(k)
Bj du2(k) Bj du1(k) duk1
du_0 u_calc 6 Aj
A_distr Aj Aj
A3 Modul_u_3 Modul_u_2 Modul_u_1
1 Distributed_duk 5
Concat_vect
yref A2

u0
y ref B1 duk1
4
duk0 duk A1 duk2
2 x0 B2
For
1:N
x0 Iterator duk3 du(k)
u0 B3 Initial value 2
For Iterator iter
duk
B_distr 7
1 duk0
u du_0

Figure 3: Subsystem diagram for distributed predictive algorithms (Nc=3).

The choice of alpha (α) provides all eigenvalues of matrix A(α) of equation
(18) smaller than 1, which is sufficient to assure that the iterative method
converges. These values were determined before the numerical simulation, and
the one optimal constrained problem was solved in Matlab environment. It
seems that the parameter tuning for the distributed algorithm does not need to
be exactly the same as the one used for the centralized version. For the same
amount of information exchange among agents, a faster reference filter
improves the response.
The results obtained by numerical simulation for the centralized control
algorithm using a variable reference signal are shown in Fig. 4. and results of
numerical simulation of the distributed algorithm after 500 iterations are shown
in Fig. 5.
Time variation of control signals Time variation of reference and output signals
30 15
u1 yref1
u2 yref2
20
10 y1
Amplitude

Amplitude

y2
10

5
0

-10 0
0 2 4 6 8 10 0 2 4 6 8 10
Time [sec] Time [sec]
(a) (b)
Figure 4: Time variation of the control signals (a) and outputs signals (b) in case of the
centralized algorithm.
 Comparative Analysis of Model Predictive Control Structures 13

Time variation of control signals Time variation of reference and output signals
25 14
u1 yref1
20 12
u2 yref2
10 y1
15
y2
Amplitude

Amplitude
8
10
6
5
4
0 2

-5 0
0 2 4 6 8 10 0 2 4 6 8 10
Time [sec] Time [sec]
(a) (b)
Figure 5: Time variation of the control signals (a) and outputs signals (b) in case of
distributed algorithm after 500 iterations.

It is noticeable that control signal obtained with the distributed algorithm is

smoother than the control signal obtained with the centralized algorithm.
In order to have an idea on the number of information interchange iterations
between agents needed for a certain performance, an analysis has been made
based on the error betveen the outputs and reference signals. For both outputs
(i=1,2) it was computed the error at every sample time k =1,…, Nt:
( yi ,k − yiref
,k )
ei ,k = ref
(21)
yi , k
A normalized absolute reference tracking error was computed using
following relationship:
1 2  1 t 
N
eabs =
2 i =1  N t k =1
ei ,k 
 ∑ (22) ∑

There was estimated also the simulation time using the Matlab functions
clock and etime for different numbers of iteration in case of the distributed
model predictive control algorithms. Fig. 6.a presents the variation of the
average errors and Fig. 6.b presents the estimated simulation time versus the
number of iterations in case of the distributed algorithms. The simulation time is
more important to be calculated for both algorithms (centralized version and
distributed version) with different prediction horizon (N) values. This
comparision is presented at Fig. 7 in case of a fixed control horizon (Nc=3).
14 K. György, L. Dávid

Normalized absolute error amplitude Simulation time

2 15

1.5
abs

10
Amplitude of e

T [sec]
1

5
0.5

0 0
1 2 3 4 [n] 1 2 3 4 [n]
n
Number of iteration [10 ] Number of iteration [10n]
(a) (b)
Figure 6: Variation of the reference tracking error (a) and of the simulation time (b)
versus the number of iterations in case of the distributed algorithm.

Figure 7: Comparison of the simulation times represented in function of the prediction

horizon (N), obtained in case of the centralized respectively in case of the distributed
algorithm.

5. Conclusion
The performance of the distributed control applied for the example
discussed in the paper is comparable to that obtained with the centralized model
predictive control. The computation power needed to solve the distributed
problem is smaller than that is needed for the centralized case. This fact may
allow the utilization of the model predictive control executed in distributed
hardware with low computational power. The size of the centralized problem
grows considerably with the number of inputs/outputs while the size of the
 Comparative Analysis of Model Predictive Control Structures 15

distributed problem remains the same for the same control horizon. One point to
mention is that unlike in the case of the presented example, most of the
multivariable problems do not have a complete interaction. In case of the
distributed algorithms the problem is to choose the convenient sample time and
the correct filter parameters’ vector. The choice of the filter should be done off-
line and the condition presented is enough to ensure the convergence of the
algorithm. Future developments are needed to provide the best filter option
(assuring the fastest convergence with robustness) and to introduce some
constraints in the model predictive applications. The main benefit expected in
case of the distributed MPC control is the improvement of the system’s
maitainability and the ’apparent’ simplicity to the user.

References
[1] Camacho, E. F., “Model Predictive Control”, Springer Verlag, 2004.
[2] Camponogara, E., Jia, D., Krogh, B. H. and Talukdar, S. N., “Distributed model predictive
control”, IEEE Control Systems Magazine, vol. 22, no. 1, pp. 44–52, February 2002.
[3] Venkat, A. N., Rawlings, J. B. and Wright, S. J., “Implementable distributed model
predictive control with guaranteed performance properties”, American Control Conference
Minneapolis, Minnesota, USA, June 14-16, 2006, pp. 613-618.
[4] Mercangoz, M. and Doyle, F. J, “Distributed model predictive control of an experimental
four tank system”, Journal of Process Control, vol. 17, no. 3, pp. 297–308, 2007.
[5] Plucenio, A., Pagano, D. J., Camponogara, E., Sherer, H. F. and Lima, M., “A simple
distributed MPC algorithm”, Rio de Janerio, Brasil.
[6] Maestre, J. M., Munoz de la Pena, D. and Camacho, E. F., “Distributed MPC: a supply
chain case study”, IEEE Conference on Decision and Control, Shanghai, China, December
16-18, 2009, pp. 7099 – 7104.
[7] Venkat, A. N., Hiskens, I. A., Rawlings, J. B. and Wright, S. J. “Distributed MPC
Strategies With Application to Power System Automatic Generation Control”, IEEE
Transactions on Control Systems Technology, vol. 16, no. 6, pp. 1192-1206, November,
2008.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 16-26

Synchronous Motor Drive at Maximum Power Factor

with Double Field-Orientation

Csaba SZABÓ, Maria IMECS, Ioan Iov INCZE

Department of Electrical Drives and Robots, Faculty of Electrical Engineering,
Technical University of Cluj-Napoca,
e-mail: csaba.szabo@edr.utcluj.ro, imecs@edr.utcluj.ro, ioan.incze@edr.utcluj.ro

Manuscript received Oct 1, 2010; revised Oct 15, 2010.

Abstract: The paper presents a vector control structure for a wound-excited salient-
pole synchronous motor, fed by a voltage-source converter, working at unity power
factor. The variable exciting current is ensured by a DC chopper. Due to this additional
intervention possibility the motor may have three degrees of freedom from the control
point of view, and three control loops will be formed instead of two: one for the control
of the mechanical quantities, and two for the magnetic ones. The three prescribed
references are the rotor angular speed, the stator flux (both directly controlled by using
PI regulators) and the power factor that is only imposed at its maximum value. In the
control structure two types of orientation procedure are used: stator-field-orientation for
power factor control, and rotor-orientation for computation of the voltage-control
variables and self commutation. There is also presented a speed-computation procedure
used in practical implementation regarding the signal processing of the incremental
encoder position. The method is based on the derivation with respect to time of both
sine and cosine functions of the rotor position. The angular speed is obtained then by
computing the module of the two resulted sinusoidal signals. This method avoids the
division by zero related issue that occurs at every zero crossing if the angular speed is
computed by dividing the time based derivative of one signal with the other one. For
validation of the presented control strategy simulations were carried out in
Matlab/Simulink® environment.

Keywords: Vector control, unity power-factor, voltage-controlled drive, rotor speed

identification, voltage-source inverter, stator-field orientation.
 Cs. Szabó, M. Imecs, I. I. Incze 17

1. Introduction
For high performance dynamic applications the most suitable solution is the
vector controlled AC drive fed by a static frequency converter (SFC). The
wound-excited synchronous motor (Ex-SyM) is the only machine capable to
operate at unity or leading power factor (PF). The structure of the vector control
system is determined by the combination between the type of the SFC used
including the pulse width modulation (PWM) procedure, the orientation field
and its identification method [2], [8], [9].
The rigorous control of the PF can be made only with the resultant stator-
field orientation. If the PF is maximum, there is no reactive energy transfer
between the armature and the three-phase power source.
Some motor-control-oriented digital signal processing (DSP) equipments
present on the market don’t dispose over implementation possibility of the
current-feedback PWM, suitable for current-controlled VSIs, consequently in
the control structure it is necessary the computation of the voltage control
variables from the current ones, imposed or directly generated by the
controllers.
The proposed control structure is based on both types of orientation. The
stator-field orientation is used for control of the unity power factor and stator-
flux, and also for generation of the armature-current control variables. The
orientation according to the rotor position (i.e. exciting-field orientation) is
applied for self-commutation and for generation of the armature-voltage control
variables for the inverter control. The transition between the two orientations is
performed by using a coordinate transformation block (CooT), which rotates the
stator-field oriented reference frame with the value of the load angle (δ =λs−θ ).

2. The mathematical model of the Ex-SyM

The mathematical model (MM) of the Ex-SyM is suitable for simulation and
also in implementation because, the computation of the control and feedback
variables in the control structure is performed also based on the MM. Usually
the equations are written in rotor-oriented rotating reference frame ( dθ − qθ ),
where θ is the rotor position. The state equations based on the quasi-flux model
may be written, as follows:
18 Synchronous Motor Drive at Maximum Power Factor with Double Field-Orientation

 dΨ sdθ
 dt = u sdθ − Rs ⋅ isdθ + ω ⋅Ψ sqθ ;

 dΨ sqθ
 dt = u sqθ − Rs ⋅ isqθ − ω ⋅Ψ sdθ ;

 dΨ e = u − R ⋅ i ;
 dt e e e

 dΨ (1)
 Ad
= u Ad − R Ad ⋅ i Ad ;
 dt
 dΨ
 Aq
= u Aq − R Aq ⋅ i Aq ;
 dt

 dω = p ⋅  3 z (Ψ 
z
  2 p sdθ ⋅ isqθ −Ψ sqθ ⋅ isdθ ) − mL ,
 dt J tot  

The integration of the state equations is made directly from the derivatives of
the angular rotor speed and fluxes, then the currents are computed from the
fluxes expressed according to the longitudinal dθ rotor axis:
 1 1 1
isdθ = '' Ψ sdθ − '' Ψ Ad − '' Ψe
 Lsd Lm ( sdθ − Ad ) Lm ( sdθ − e )

 1 1 1
i Ad = − '' Ψ sdθ + '' Ψ Ad − '' Ψe (2)
 Lm ( sdθ − Ad ) LAd Lm ( e − Ad )

ie = − 1 1 1
Ψ sdθ − '' Ψ Ad + '' Ψ e
 ' '
Lm ( sdθ − e ) Lm ( e − Ad ) Le

and according to the quadrature qθ rotor axis:
  L 
isqθ = 1 Ψ sqθ − mq Ψ A 
'  q 

 L'sq LAq
 
 (3)
 1  Lmq 
i Aq = − '' Ψ Aq − Ψ sqθ 


 LAq  Lsq 
As state-variables were chosen the fluxes (i.e. the direct and quadrature axis
components of the stator flux (Ψ sdθ and Ψ sqθ ) and of the damper winding flux
(Ψ Ad and Ψ Aq ), the exciting winding flux (Ψ e ) and the rotor electrical angular
speed ( ω ).
 Cs. Szabó, M. Imecs, I. I. Incze 19

3. Double field-oriented control of the Ex-SyM

In the proposed control structure, presented in Fig. 1, the salient pole
Ex-SyM is fed by a voltage-source inverter (VSI). Three control loops are
formed: two magnetical (for flux and PF control) and a mechanical one (for
speed). The flux and the speed are controlled directly by PI controllers, and the
PF is controlled indirectly. The PF is maximum, if the stator voltage and stator
current are in phase. Consequently, the stator-current space phasor is results
perpendicular onto the stator-flux vector Ψs [2], [8].
The perpendicularity can be achieved by canceling the stator-field-oriented
longitudinal armature reaction ( isdλs = 0 ). The reference armature-current
components are oriented according to the direction of the resultant stator-flux
Ref
phasor. The longitudinal component isdλs is an imposed value, and it is
Ref
cancelled, while the quadrature component isqλs results at the output of the
speed controller. However the stator-current control is recommended to be
made in exciting field- (i.e. rotor-position-) oriented (dθ–qθ) rotating reference
frame, because the self-commutation of the motor, based on the rotor position,
is made inherently by means of the reverse Park transformation block. The re-
orientation (from the resultant stator field to the exciting one) is made by means
of a reverse CooT block, which rotates the stator-field-oriented reference frame
with the value corresponding to the load angle δ = λs−θ [2], [8], [9].
The voltage-control variables are generated by the two current controllers in
the active and the reactive control loops. In the voltage-computation block UsC
the electromagnetic cross-effect is taken into account, realizing the re-coupling
of the two decoupled control loops, the active and the reactive ones, by means
of the rotating EMF components [2], [8], as follows:
u sdθ
Ref
= vsdθ
Ref
− ωΨ sqθ ;

 Ref (4)
u sqθ = vsqθ + ωΨ sdθ
Ref


The inverter control is made by feed-forward voltage-PWM procedure with

simple on-off controllers [5], based on the following PWM logic:
− 1, if ucr < u Ref ;

m log =  (5)
1, if ucr > u
Ref
.

The exciting current is controlled also with voltage PWM by means of a DC

chopper.
 20
Synchronous Motor Drive at Maximum Power Factor with Double Field-Orientation

Figure 1: Vector control system of the adjustable excited synchronous motor fed by a static frequency converter with feed-
forward voltage-PWM and double field orientation, operating with controlled stator flux and imposed unity power factor.
 Cs. Szabó, M. Imecs, I. I. Incze 21

In the third control loop the resultant stator-flux is directly controlled with a
PI controller, which outputs the ims magnetizing current, necessary for the
computation of the excitation current in the IeC block [2], [8].

4. Angular speed computation

For the self-control of the Ex-SyM it is important an accurate information
about the rotor position. This was realized using an incremental encoder,
mounted on the motor shaft. The mounting is realized in a manner, that the
encoder index signal is synchronized with respect to the rotor position. The
encoder generates a number of pulses proportional to the angular position of the
shaft. It gives information also about the sense of the rotating motion: positive
values for direct and negative ones for reverse running. The counter resets it to
zero at every full rotation. The amplitude of this signal will be equal to the
number of the increments/revolution of the encoder. This position signal θ enc
provided by the encoder is processed in order to obtain a position signal θ
between [0, 2π ] for direct, and [0, − 2π ] for reverse rotation respectively,
based on the following expression:
θ enc
θ= 2π , (6)
Nr
where Nr is the number of increments per revolution of the digital encoder.
The electrical angular speed of the rotor is also required in the UsC block for
the computation of the voltage control variable. The angular speed can be
determined using the sine and cosine functions of the rotor position. The
amplitude of these functions is equal to 1. The method presented in [1] is based
on the derivation of either the sine or the cosine function of the position, and
uses the following relation:
dθ d (sin θ )
ω= = (cos θ )−1 (7)
dt dt
The drawback of this method is, that division by zero occurs at every zero
d (sin θ )
crossing of the sinusoidal function, because the is shifted by 90°, and
dt
is in phase with the cos θ . The resulting angular speed is inaccurate and
presents infinite peaks at these moments, that cannot be processed correctly, as
is shown in Fig. 2. In order to avoid this situation different methods were
implemented. A known method is to filter the obtained signal, but this method
leads to inappropriate results especially in transient operation when the speed is
not constant. Another approach is based on avoiding the division by zero by
22 Synchronous Motor Drive at Maximum Power Factor with Double Field-Orientation

introducing a discontinuity in the sine function around zero, assigning instead of

this a very small, but nonzero value.
In this paper a different approach is used, which consists in the derivation of
both, the sine ( S = sin θ ) and cosine ( C = cos θ ) functions of the angular
position θ, where S = f (θ ) , C = f (θ ) , and θ = f (t ) [4], [15]:
dS d (θ )
= cos θ (8)
dt dt

dC d (θ )
= − sin θ (9)
dt dt

The angular speed is determined by the following relation [15]:

2  d (θ )  2  d (θ ) 
2 2 2 2
 dS   dC  dθ
ω=   +  = cos θ   + sin θ   = (10)
 dt   dt   dt   dt  dt

Using this procedure the zero crossing is avoided, and leads to an accurate
result, as shown in Fig. 3.
-sin(theta)
cos(theta),
cos(theta)

1 1
cos(theta)
0 0
-sin(theta
-1 -1
0.9 0.95 1 1.05 1.1 1.15 0.9 0.95 1 1.05 1.1 1.15
t(s) t(s)
dS/dt, dC/dt

500 500
dS/dt
dS/dt

0 0 dC/dt

-500 -500
0.9 0.95 1 1.05 1.1 1.15 0.9 0.95 1 1.05 1.1 1.15

t(s) t(s)
500
w (rad/s)
w (rad/s)

500
0
0

-500 -500
0.9 0.95 1 1.05 1.1 1.15
0.9 0.95 1 1.05 1.1 1.15
t(s)
t (s)

Figure 2: The rotor angular speed Figure 3: The rotor angular speed
computed with the classical method, computed using the proposed method
leading to an inaccurate result avoiding the division by zero issue.
The sign of the position signal θ gives the direction of the rotation. The
computation of the angular speed is based on the following expression:
 d (sin θ )   d (cos θ ) 
2 2
ω = ω sign(θ ) =   +  sign(θ ) , (11)
 dt   dt 
 Cs. Szabó, M. Imecs, I. I. Incze 23

and its computation may be processed with the structure presented in Fig. 4 [4].

Figure 4: Block symbol and structure for computation of the rotor angular speed based
on the encoder position signals.

5. Simulation results
Based on the structure from Fig. 1 simulations were performed in Matlab-
Simulink® environment. The rated data of the simulated salient pole Ex-SyM
are: UsN = 380 V, IsN = 1.52 A, PN = 800 W, fN=50 Hz, nN = 1500 [rpm], cosφ=
0.8 (capacitive).
400 1
Power factor

200 0.5
w rad/sec

0 0

-200 -0.5

-1
-400
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
t (s) t (s)
10
Stator flux (Wb)
m e , m L (Nm)

1
0

0.5
-10

-20 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
t (s) t (s)

Figure 5: Electrical angular speed (w), Figure 6: The power factor and the stator-
electromagnetic (me) and load torque (mL) flux amplitude versus time.
24 Synchronous Motor Drive at Maximum Power Factor with Double Field-Orientation

400 4

300 3

200 2

100 1
w (rad/s)

isd (A)
0 0

-100 -1

-200 -2
w=f(m )
-300 e
-3
w=f(mL)
-400
-15 -10 -5 0 5 10 -4
4 3 2 1 0 -1 -2 -3 -4
m , m (Nm) i (A)
e L
sq

Figure 7: Mechanical characteristics: Figure 8: The trajectory of the armature-

speed versus torque of the motor w=f(me) current space-phasor in natural stator-
and of the mechanical load w=f(mL). fixed coordinate frame.
4
i 1
sdls
3
isqls 0.8
2 0.6

0.4
isdls, isqls (A)

1
psisd (Wb)

0.2
0 0

-0.2
-1
-0.4
-2
-0.6

-3 -0.8

-1
-4
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1
t(s)
psisq (Wb)

Figure 9. The armature-current two-phase Figure 10. The trajectory of the stator-
components ( isdλs and isqλs ) in stator-flux- flux space-phasor in natural stator-fixed
coordinate frame.
oriented coordinate frame.

After the starting process the motor runs at the rated speed value
corresponding to a frequency of 50 Hz. At t = 1 s under the full rated load a
speed reversal is applied. The mechanical load has reactive character and it is
linearly speed-dependent.
The simulation results show that the proposed control structure from Fig. 4
is a viable one with improved performances with respect to the conventional
vector control systems [2].
The results show a good performance of the drive also in transient operation
at starting, and also at speed reversal (Fig. 5). The power factor is maximum
also during the speed reversal, when the drive is in regenerative running for a
short period of time, as is shown in Fig. 6. Unity power factor is realized by
canceling the stator-field-oriented longitudinal armature reaction, as in Fig. 9.
 Cs. Szabó, M. Imecs, I. I. Incze 25

6. Conclusion
The presented control structure uses two types of orientations: resultant
stator-field and exciting-field, i.e. rotor-position orientation.
For a rigorous control of the power factor, stator-field orientation was
applied, and in order to achieve unity power factor operation in the reactive
control loop the stator-flux oriented longitudinal armature reaction was
cancelled.
In order to obtain improved control performances, the computation of the
control variables were made in rotor-oriented reference frame, so the self-
control of the motor and the synchronization of the inverter trigger signal are
made based on a directly measured value of the rotor position.
In the control structure of the voltage-controlled Ex-SyM drives the dual
field-orientation combines the advantages of the two types of field-orientation
procedure, on the one hand of the stator-field orientation suitable for power
factor control and on the other hand of the exciting-field orientation for
computing feedback- and control-variables based on the rotor-position-oriented
classical MM, in order to ensure sophisticated calculus due to the geometry
characteristics, i.e. two-axis symmetry of the salient-pole rotor.
The applied computation procedure of the synchronous angular speed
corresponding to the rotating the stator-flux avoids the division by zero and
gives an accurate result.
The applied computation procedure of the rotor angular speed, which avoids
the division by zero, gives accurate results also in computation of the
synchronous angular speed of the rotating resultant orientation flux in any field-
oriented vector control structure, including induction motor drives, too.
The presented control structure was validated by simulation in
Matlab/Simulink®, and the obtained result shows the reliability of this method.
The practical implementation was realized on an experimental rig based on
the dSpace DS1104 controller board. The results were published in [8].

References
[1] Kelemen, Á., and Imecs, M., “Vector Control of AC Drives, Volume 1: Vector Control of
Induction Machine Drives”, OMIKK-Publisher, Budapest, Hungary, 1991.
[2] Kelemen, Á., Imecs, M., “Vector Control of AC Drives, Volume 2: Vector Control of
Synchronous Machine Drives” Ecriture Budapest, Hungary, 1993.
[3] W. Leonhard, “Control of Electrical Drives”, Springer Verlag. Berlin, Heidelberg, New
York, Tokyo, 1985.
[4] Szabó, Cs., “Implementation of Scalar and Vector Control Structures for Synchronous
Motors (in Romanian)”, PhD Thesis, Technical University of Cluj-Napoca, Romania,
2006.
26 Synchronous Motor Drive at Maximum Power Factor with Double Field-Orientation

[5] Kazmierkowski, M. P., Tunia, H., “Automatic Control of Converter-Fed Drives”, Elsevier,
Amsterdam, 1993.
[6] Vauhkonen, V., “A cycloconverter-fed synchronous motor drive having isolated output
phases”, in Proc. International Conference on Electrical Machines, ICEM ’84, Lausanne,
Switzerland, 1984.
[7] Imecs, M., Szabó, Cs., Incze, I. I., “Stator-field-oriented control of the variable-excited
synchronous motor: numerical simulation”, in Proc. 7th International Symposium of
Hungarian Researchers on Computational Intelligence HUCI 2006, Budapest, Hungary,
2006, pp. 95-106.
[8] Szabo, C., Imecs, M., Incze, I. I., “Vector control of the synchronous motor operating at
unity power factor”, in Proc. 11th International Conference on Optimization of Electrical
and Electronic Equipment, OPTIM 2008, Brasov, Romania, 2008, vol. II-A pp. 15-20.
[9] Szabo, C., Imecs, M., Incze, I. I., “Synchronous motor drive with controlled stator-field-
oriented longitudinal armature reaction”, in Proc The 33rd International Conference of the
IEEE Industrial Electronics Society, IECON 2007, Taipei, Taiwan, 2007, CD-ROM
[10] Davoine, J., Perret, R., Le-Huy H., “Operation of a self-controlled synchronous motor
without a shaft position sensor”, Trans. Ind Applications, IA-19, no. 2, pp. 217-222,
March/April 1983.
[11] Shinnaka, S., Sagawa, T., “New optimal current control methods for energy-efficient and
wide speed-range operation of hybrid-field synchronous motor,” IEEE Trans. Ind
Electronics, vol. 54, no. 5, pp. 2443-2450, Oct. 2007.
[12] Imecs, M., Incze, I. I., Szabó, Cs., “Double field orientated vector control structure for
cage induction motor drive”, Scientific Bulletin of the „Politehnica” University of
Timisoara, Romania, Transaction of Power Engineering, Tom 53(67), Special Issue, pp.
135-140, 2008.
[13] Imecs, M., Incze, I. I., Szabó, Cs., “Dual field orientation for vector controlled cage
induction motors”, in Proc. of the 11th IEEE Internat. Conference on Intelligent
Engineering Systems, INES 2009, Barbados, 2009, pp 143-148.
[14] Imecs, M., Szabó, Cs., Incze, I. I., “Stator-field-oriented vector control for VSI-fed
wound-excited synchronous motor”, in Proc. International Aegean Conference on
Electrical Machines and Power Electronics, ACEMP-ELECTROMOTION Joint
Conference, Bodrum, Turkey, 2007, pp. 303-308.
[15] Wallfaren, W, “Method and apparatus for determining angular velocity from two signals
which are function of the angle of rotation”, US Patent No.4814701, Mar. 21 1989.
[16] Bose K. B., “Modern Power Electronics and AC Drives”, Prentice-Hall PTR. Prentice-
Hall Inc., Englewood Cliffs, New Jersey, USA, 2002.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 27-39

Incremental Encoder Based Position and Speed

Identification: Modeling and Simulation
Ioan Iov INCZE, Alin NEGREA, Maria IMECS, Csaba SZABÓ
Department of Electrical Drives and Robots, Faculty of Electrical Engineering,
Technical University of Cluj-Napoca,
e-mail: ioan.incze@edr.utcluj.ro, Cornel.NEGREA@edr.utcluj.ro,
imecs@edr.utcluj.ro, csaba.szabo@edr.utcluj.ro

Manuscript received Oct 1, 2010; revised Oct 15, 2010.

Abstract: Electrical drives frequently use incremental encoders as position sensor.

The paper deals with the modeling and simulation of an incremental encoder and
associated units for processing the information provided by the encoder. A
mathematical model of the incremental encoder is presented. Based on encoder signals
the direction of the rotation, the position and the speed are identified. The described
procedure for determination of the direction of the rotation is able to identify the
direction changing in all cases during a rotation equal to the minimal detectable
rotation-angle-increment. The computing of the position is based on the algebraic
summing of the number of the generated encoder signals. For the speed determination
different methods are modeled and simulated: for high speed region the frequency
measurement is used and for low speed domain the period measurement is appropriate.
In case of a large speed variation the minimal-error-based switching between the two
methods is suitable. Matlab-Simulink® simulation structures were realized for the
encoder signals based on the identification of the direction of the rotation, for the
position and speed computation. Experimental results performed on a DSP-based set-up
(under development) are given. The presented simulation subsystems of the encoder,
position and speed computation may be integrated in any Matlab-Simulink® structure.

Keywords: Angle transducer, position sensor, incremental encoder modeling,

incremental encoder simulation, angular speed identification, electrical drive.

1. Introduction
The incremental encoder is a device which provides electrical pulses if its
shaft is rotating [1], [2], [4]. The number of the generated pulses is proportional
28 I. I. Incze, A. Negrea, M. Imecs, Cs. Szabó

to the angular position of the shaft. The incremental encoder is one of the most
frequently used position transducers. The principle of an optical incremental
encoder is presented in Fig. 1. Together with the shaft there is rotating a
transparent (usually glass) rotor disc with a circular graduation-track realized as
a periodic sequence of transparent and non-transparent radial zones which
modulates the light beams emitted by a light source placed on one side of the
disc on the fix part (stator) of the encoder. On the opposite side the modulated
light beams are sensed by two groups of optical sensors and processed by
electronic circuits. Each of the two outputs of the encoder (noted A and B) will
generate one pulse when the shaft rotates an angle equal to the angular step of
graduation θp, i.e. the angle according to one successive transparent and non-
transparent zone. The number of pulses (counted usually by external electronic
counters) is proportional to the angular position of the shaft. Due to the fact that
the light beams are placed shifted to each other with an angle equal to the
quarter of angular step of graduation θp/4, the pulses of the two outputs will be
also shifted, making possible the determination of the rotation sense. A third
light beam is modulated by another track with a single graduation. The output
signal (named Z) associated to this third beam provides a single pulse in the
course of a complete (360°) rotation. The shaft position corresponding to this
pulse may be considered as reference position. Fig. 2 shows the output pulses of
the encoder.
Usually for counter-clockwise (CCW) direction θ is considered as positive,
and for clockwise (CW) direction it is considered negative.

Encoder CCW direction

axis A
dθ θ
ω=
dt B
LED A θ
LED Z Z θp /4
Encoder LED B θ
disc θp 2π - θp 0 θp
θp 4 CW direction
A
4 −θ
θp B
θ Rotating −θ
axis
A Z
Graduation Z
Fixed −θ
B
reference - (2π - θp ) 0 - θp
Photosensors axis θp θp / 4 θp /4

Figure 1: Construction principle of the Figure 2: Diagram of the output

incremental encoder: the gray surfaces are signals for counter-clockwise (CCW)
optically transparent. and clockwise (CW) rotation.
 Incremental Encoder Based Position and Speed Identification: Modeling and Simulation 29

2. Incremental encoder modeling

The input signal of the incremental encoder is the angular position θ of its
shaft with respect to the fixed reference axis. The output signals are the two
pulses shifted by a quarter angular step A(θ) and B(θ), respectively the marker
signal Z(θ). If θp is the angular step of the encoder, the outputs may be described
by the following equations [2]:
( )
1 if 0 ≤ θ modθ p ≤ θ p /2;
A(θ ) = 
( )
0 if θ p /2 < θ modθ p ≤ θ p ;
(( ) )
1 if 0 ≤ θ − θ p / 4 modθ p ≤ θ p /2;
B(θ ) = 
(( ) )
0 if θ p /2 < θ − θ p / 4 modθ p ≤ θ p ;
(1)

1 if θ mod(2π ) = 0;
Z (θ ) = 
0 if θ mod(2π ) ≠ 0.
During a rotation angle of the shaft, equal to the angular step of graduation
θp, there are four switching events in the output pulses; therefore the minimal
rotation-angle-increment detectable by the encoder is θp/4 [3]. The number of
pulses, generated by the encoder in the course of a rotation, is equal with the
number of angular steps of the graduation on the circular track on the rotor.
2π
Nr = (2)
θp
Based on (1) a Matlab/Simulink® a simulation structure shown in Fig. 3 was
built. The outputs A, B and Z are computed by Simulink® function blocks. θp
is defined by a constant block. The structure is saved as a subsystem. The
simulation structure of the incremental encoder may be integrated in any other
Simulink® structure.

3. Encoder based position identification

In an incremental-encoder-based system the angular position θ is measured
with respect to a fixed reference axis (θ=0 rad.) and it is obtained by algebraic
counting of the number of the generated encoder pulses according to the CCW
(ΣNi pulses) and CW direction (ΣNj pulses) and multiplying it with the angular
step θp of the encoder [2]. Mathematically:
30 I. I. Incze, A. Negrea, M. Imecs, Cs. Szabó

 CCW CW 
θ = θp
 ∑ Ni − ∑ N j  = θ pN .

(3)
 i j 
In order to compute the algebraic number of pulses it is necessary to know
the direction of the rotation.

A. Identification of the rotation direction

The two θp/4 shifted output signals of the encoder contain implicitly also the
direction information which may be obtained in different ways.

A
Modulo Relational A
θp Operators Logic AN

B
θ Modulo Relational B
θp BN
Operators Logic

Z
Modulo Zero Z
2π Detect Logic ZN
IE
Subsystem

Figure 3: Simulation structure of the incremental encoder.

(Note: The subscript “N” denotes the negated logical variable.)
Taking into account the all four possible combinations of A and B signals for
the reversals, it is possible to detect the direction changing in all cases during a
rotation of the minimal detectable rotation-angle-increment θp/4. Table 1 shows
the all combinations of signals which detect the reversal of the rotation sense.

Table 1. Combination of signals which detect the reversal of rotation.

From CCW to CW From CW to CCW
Q=1 to Q=0 Q=1 to Q=0
(Triggered by R) (Triggered by S)
Occurs if Occurs if
A B A B
0 0→1 0→1 0
0→1 1 0 1→0
1→0 0 1 0→1
1 1→0 1→0 1
Note: The 0→1 denote the raising-edge and the 1→0 the falling -edge
of associated logic variable.
 Incremental Encoder Based Position and Speed Identification: Modeling and Simulation 31

A trivial solution of the problem may be sampling at every rising edge of the
B output pulses of the A output logic value. The resulted logic value will be 1
for counterclockwise (CCW) direction of rotation, and 0 for the clockwise (CW)
direction. The method detects the direction changing only after a time interval
according to a rotation of 3 θp/4 – 5 θp/4.

B. The position-identification structure

Based on (3) and the above presented direction identification method, a
Simulink® subsystem was built for position computation. Its structure is
presented in Fig. 4.
Poz Subsystem En
CCW Out θp
A
In
Counter ΣN CCW
CCW Res
B Logic S Q θ
FF
CW R Q En +
Logic
In
CW Out - N

Z
Counter
ΣN CW
1st Res
Pulse S

Figure 4: The simulation structure of the position computing subsystem.

The left side of the structure identifies the direction of the rotation. The
direction signal S enables the appropriate (CCW or CW) counter. The structure
has to be provided with the “1st Pulse” block (broken line in Fig. 4) in order to
extend the position measurement to more rotations.

4. Encoder based speed identification

The signals generated by incremental encoder provide also information
regarding the speed of rotation. A few basic methods are known which are
discussed below.

A. Speed identification based on frequency measurement

The frequency of the encoder pulses is proportional to the angular speed.
The number of pulses ΔN is counted during a known fix sampling period Ts
(see Fig. 5) [5]. The angular speed is determined by the expression:
32 I. I. Incze, A. Negrea, M. Imecs, Cs. Szabó

dθ dθ 2π∆N θ p ∆N
ω= ≅ ≅ = (4)
dt Ts N rTs Ts
Due to the lack of synchronization between the sampling period and encoder
pulses a quantization error occurs. The relative error of the procedure is given
by
1 2π
εf = (5)
ω N rTs
and depends on the reciprocal of the speed, the measuring interval and the
resolution of the encoder [5].
∆ N pulses
A
or
t
B
Ts
measurement period

Figure 5: The principle of the speed identification based on the frequency measurement.
The speed calculation structure based on frequency measurement is
presented in Fig. 6. In order to enhance the precision, the “Logic x4” block
multiplies by 4 the frequency of the encoder signals. Two, alternatively resetted
and enabled counters count the number of pulses. The content of the just
disabled counter is used for speed computation.
The speed identification based on frequency measurement produces
relatively small errors at high speed because the number of pulses from the
encoder in the measurement-time interval is high.

Speed-f Subsystem
In θp Ts
A f
Counter Out
∆N′ ∆N′
Logic 4f
B x4 or
En
∆N′′ ω
Logic
Speed
Ts In and
Ts Counter Out Error
Generator T s En
∆N′′ Computing εf

Figure 6: The simulation structure of the speed computing subsystem

based on frequency measurement.
 Incremental Encoder Based Position and Speed Identification: Modeling and Simulation 33

B. Speed identification based on period measurement

The speed identification based on frequency measurement at low speed is no
longer an option. A better solution is the speed identification based on period
measurement. The method consists of counting the pulses of a high frequency
clock signal (having period Thf) during an encoder period, as is shown in Fig. 7
[5].
A
or
t
B

HF
Clock t
T hf

n pulses

Figure 7: The principle of the speed identification based on the period measurement.
In this case the expression of the angular speed is
dθ dθ 2π
ω= ≅ ≅ (6)
dt nThf N r nThf
where n represents the counted number of the high frequency pulses. The
relative error is increasing with the rotation frequency and is given by [5]
N rωThf
εp = (7)
2π
The speed calculation structure based on period measurement is presented in
Fig. 8.
Speed-p Subsystem
In θp Thf
HF Clock Counter Out
n′ n′
Generator or
En
n′′ ω
Logic
Speed
A En In and
Counter Out Error
B Logic
En n′′ Computing εp
En

Figure 8: The simulation structure of the speed computing subsystem

based on the period measurement.
The structure resembles the previous one; the main difference is that the high
frequency pulses are counted during an encoder signal period.
34 I. I. Incze, A. Negrea, M. Imecs, Cs. Szabó

C. Combined method for speed identification

In order to minimize the speed identification error it is suitable to use in low
speed region the period measurement method u (t ) and as the speed is
increasing to switch to the frequency measurement one. The moment of
switching is given by the equality of the errors ( ε p = ε f ), that happens when the
speed riches the value
2π 1
ωs = ⋅ (8)
Nr Ts ⋅ Thf

The subsystem presented in Fig. 9 computes the speed based on above

described method.
Speed Subsystem
ωf
A ω
ωp
B Speed_f

εp
Speed_p εp >
< εf
εf

Figure 9: The simulation structure of the combined speed computing subsystem.

The compensation of the errors caused by the sampling times may enhance
the precision of the measurement [6].

5. Simulation results
The structure of the interconnected functional units for simulation of
position computing is shown in Fig. 10. The reference angular position θref (the
input signal of the encoder block “IE”) is generated by a user programmable
“Function generator” block. The encoder generates the A, B and Z signals.
Based on these, the block “Poz” computes the position θ and the block “Speed”
provide the computed angular speed. In order to test the structure, the function
generator was programmed in order to start the simulation generating a positive
ramp-reference angle, which is the input signal for the “IE” encoder block. At
0.2 s the ramp is switched to negative (equivalent to a reversal from CCW to
CW), decreasing in time until 0.8 s, when it is again switched to positive.
 Incremental Encoder Based Position and Speed Identification: Modeling and Simulation 35

A
Function θ ref B
IE
Generator
Z

Poz Speed

θref A B Z N θ S ω ωf εf ωp εp

To Workspace

Figure 10: The structure of the interconnected functional units for simulation
of the position and speed computation:
IE – incremental encoder, Poz – position computing block,
Speed – speed computing block.
The time profile of the generated reference angle is presented in Fig. 11 a)
(top trace). The block “Poz”, using the encoder output signals, determines the
direction of the rotation (in Fig. 11 a) bottom trace) and computes the position
(shown in Fig. 11 a) middle trace). The computed position follows very well the
reference one. Fig. 11 b) presents an enlarged detail of superposed reference
and computed angle before and after the reversal at 0.2 s. The incremental
character of the computed position is evident.
0.1 0.105
thetaref [rad]

0
0.1
-0.1 theta
0 0.2 0.4 0.6 0.8 1
0.095
theta, thetaref [rad]

0.1
theta [rad]

0 0.09
thetaref
-0.1
0 0.2 0.4 0.6 0.8 1 0.085
1
Direction

0.08
0.5

0
0.075
0 0.2 0.4 0.6 0.8 1 0.16 0.18 0.2 0.22 0.24 0.26
time [s] time [s]

a) b)
Figure 11: The simulation results representing the reference angle and computed
angle during reversal.
a) From top to bottom: reference angle θref, computed angle θ, direction signal S.
b) Detail of the reference angle θref and computed angle θ versus time.
There was analyzed the reversal process. The simulated results are presented
in Fig. 12. a)–d). The parameters of the function generator were selected in such
a manner, that all possible combinations of signals A and B at reversal
(presented in Table 1) were captured. In all cases the sensing of the reversal is
done in a quarter of angular step, as is shown in Fig. 12.
36 I. I. Incze, A. Negrea, M. Imecs, Cs. Szabó

1 1
A output

out A
0.5 0.5

0 0
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
1 1
B output

out B
0.5 0.5

0 0
a) 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
1 1
Direction

Direction
0.5 0.5

0 0
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
time [s] time [s]

1 1
out A

out A
0.5 0.5

0 0
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
1 1
out B

out B
0.5 0.5

0 0
b) 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
1 1
Direction

Direction

0.5 0.5

0 0
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
time [s] time [s]

1 1
out A

out A

0.5 0.5

0 0
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
1 1
out B

out B

0.5 0.5

0 0
c) 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
1 1
Direction

Direction

0.5 0.5

0 0
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
time [s] time [s]

1 1
out A
out A

0.5 0.5

0 0
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11

1 1
out B
out B

0.5 0.5

0 0
d) 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14

1 1
Direction
Direction

0.5 0.5

0 0
0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
time [s] time [s]

Figure 12: The simulation results showing all combinations of the reversal process.
Left column: Reversal from CCW to CW, Right column: Reversal from CW to
CCW, top trace: output A, middle trace: output B, bottom trace: direction signal.
Reversal occurs at: a) A=0, B=0; b) A=0, B=1; c) A=1, B=0; d) A=1, B=1.
 Incremental Encoder Based Position and Speed Identification: Modeling and Simulation 37

Fig. 13 presents the A, B and Z signals at crossing the reference position

(θ = 0) in CW and CCW direction.
1 1
out A

out A
0.5 0.5

0 0
0.795 0.8 0.805 0.81 0.39 0.395 0.4 0.405
1 1
out B

out B
0.5 0.5

0 0
0.795 0.8 0.805 0.81 0.39 0.395 0.4 0.405
1 1
out Z

out Z
0.5 0.5

0 0
0.795 0.8 0.805 0.81 0.39 0.395 0.4 0.405
time [s] time [s]

a) b)
Figure 13: The A, B and Z signals of the encoder at crossing the reference position:
a) in CCW direction, b) in CW direction.
In order to test the speed identification, the function generator was
programmed for a linearly increasing and decreasing speed profile. Fig. 14
shows the theoretical speed profile and the computed speed. In Fig. 15 is
presented the variation of the errors εf and εp. The switching between the two
methods occurs at moment 0.2 s and 0.8 s, respectively.
50 0.35
speedref ωref, speed ω, [rad/sec]

0.3
40 εf
ωref 0.25
30
ε f, ε p, [%]

0.2
ω
0.15
20 εp
0.1
10
0.05

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
time[s] time[s]

Figure 14: The simulation results Figure 15: Variation of the error versus
showing reference and calculated speed. time of the two speed calculation
methods.
The structure presented in Fig. 10 may be integrated in the simulation
structures of electrical drives [2], [5]. In this case the input signal of the encoder
– i.e. the angular position – will be provided by the mathematical model of the
electrical machine. The computed position and speed is used as position feed-
back signal by the control system of the drive.
The conditions used in simulations are: N=500, Thf = 4 μs, Ts = 6 ms, the
simulation step was taken 1 μs.
38 I. I. Incze, A. Negrea, M. Imecs, Cs. Szabó

6. Experimental results
In order to investigate different position and speed determination algorithms
an experimental set-up is under construction (see Fig. 16). The incremental
encoder (type 1XP8001-1) is mounted on the shaft of an induction motor driven
by a static frequency converter. (Micromaster, Siemens). The encoder signals
are processed by an experimental board built around a DSP based development
board from Spectrum Digital.
Storage
Micromaster
PC Scope
Frequency Converter
TDS3014

A,B,Z Ua,b,c
Experimental
Board
Incremental Induction
IE IM
Encoder Motor

Figure 16: Block scheme of the experimental rig.

Fig. 17 a) shows the captured encoder signals for CCW rotation, and Fig. 17
b) represents the A and B signals during the CW to CCW reversal process. The
reversal occurs for A=0 B=0.

a) b)
Figure 17: Captured encoder output signals
a) for CCW direction versus time;
Top: Signal A, Middle: Signal B, Bottom: Marker signal Z;
b) for direction reversal from CW to CCW direction versus time;
Top: Signal A, Bottom: Signal B.
A comparison of the above figures to Fig. 13 a) and Fig. 12 a) shows that
the captures are very closed to the simulated results.
 Incremental Encoder Based Position and Speed Identification: Modeling and Simulation 39

7. Conclusion
The information provided by the incremental encoders is inherently digital.
The angular position of the encoder shaft is obtained by algebraic summing
of the number of pulses provided by the encoder according to CCW and CW
rotation.
The direction of the rotation may be determined by a digital decoding
scheme using the two quadrature signals. The direction changes are detected in
an angular interval equal to a quarter of the angular step of the graduation.
The frequency of the pulses generated by the encoder is proportional to the
speed of the rotation. The error of the measurement is inversely proportional to
the speed, therefore the procedure is appropriate for high speed region.
At low speeds the measurement of the period of the encoder pulses is
recommended. The measurement error is decreasing with the decreasing of the
speed.
In case of large speed variations – in order to minimize the errors – a
switching between the two described methods is suitable.
The presented simulation structure of the incremental encoder, position and
speed computation may be integrated in any Matlab-Simulink® structure.

References
[1] Incze, J. J., Szabó, Cs., and Imecs, M., “Modeling and simulation of an incremental encoder
used in electrical drives”, in Proc. of 10th International Symposium of Hungarian
Researchers on Computational Intelligence and Informatics CINTI 2009, Budapest,
Hungary, 2009, pp. 97-109.
[2] Incze, I. I., Szabó, Cs., and Imecs, M., “Incremental encoder in electrical drives: modeling
and simulation” in Studies in Computational Intelligence Editors: I. J. Rudas, J. Fodor, J.
Kacprzyk, Springer Verlag, Germany, under press.
[3] Koci, P., and Tuma, J., “Incremental rotary encoders accuracy”, in Proc. of International
Carpathian Control Conference ICCC 2006, Roznov pod Rashosten, Czech Republic, 2006,
pp. 257-260.
[4] Lehoczky, J., Márkus, M., and Mucsi, S., „Szervorendszerek, követő szabályozások”,
Műszaki Kiadó, Budapest, Hungary, 1977.
[5] Petrella, R., Tursini, M., Peretti, L., and Zigliotto, M., “Speed measurement algorithms for
low resolution incremental encoder equipped drives: comparative analysis”, in Proc. of
International Aegean Conference on Electrical Machines and Power Electronics, ACEMP-
ELECTROMOTION Joint Conference, Bodrum, Turkey, 2007, pp. 780-787.
[6] Miyashita, I., and Ohmori, Y., “A new speed observer for an induction motor using the
speed estimation technique”, in Proc. of European Power Electronics Conference EPE΄93,
Brighton, United Kingdom, 1993, vol. 5, pp. 349-353.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 40-50

Aluminium Electrolytic Capacitor Research and

Development Time Optimization Based on a
Measurement Automation System
Dénes FODOR
Pannon University, Veszprém, Institute of Mechanical Engineering, Department of Applied
Mechanics, Automotive System Engineering Group, Egyetem st. 10.; 8200 Veszprém,
Hungary, e-mail: fodor@almos.uni-pannon.hu

Manuscript received October 20, 2010; revised November 7, 2010.

Abstract: The aim of this paper is to present partly the Measurement Automation
System (MAS) of an Electrolytic Capacitor Development Laboratory at EPCOS
Hungary. The main role of the MAS is to facilitate the electrolyte and capacitor research
and development, through the automation of the related measurement tasks, and to
provide a powerful database system background for data retrieval and research decision
support. The paper introduces only a few applications of the entire system. More than
27 different electrolyte and capacitor measurements were automated. All the
measurements have been implemented in a similar manner. During the process the user
initializes the measurement, sets the measurement environmental parameters, and
launches the execution. The program runs on its own, sending automatically the results
of the measurement to a database system, from where the data can be retrieved in a
predefined or a non-predefined way. For the realization of the above requirements the
National Instruments measurement, data acquisition and LabVIEW software
development tools were chosen as implementation and development platform. After
validation of the system, there are many advantages as making the measurements more
precise, more reliable, fault tolerant, parallel running, which all contribute to speed up
the research and development of new components and devices.
The developed measurement system controls and harmonizes the different devices
and supervises their work. The developers do not have to encroach. The user can simply
check the measurement phase by a glance on the screen. The programs estimate and
display the running time of the experiments, allowing for the researchers working on the
laboratory to manage the instrumental resources in time and to schedule in advance (for
hours, weeks and months) the new measurements. Another big advantage is the
database system behind, which stores the result of each measurement in an easy
searchable way. Different measurement reports and statistical diagrams can be made
automatically and the results can be reused in later research. The effectiveness of the
 D. Fodor 41

system was tested also via the inner gas pressure measurement of electrolytic capacitors
to estimate the life-span of the capacitors. According to the experiments the introduction
of the MAS system in the Lab the research and development time for new electrolytes
and capacitors has been decreased considerably.

Keywords: Measurement automation, test automation, aluminium electrolytic

capacitors, data acquisition, data mining.

1. Introduction
Capacitors play a very important role in our world [1], [2]. They can be
found in every electronic device around us, they are widespread all over the
world used as energy storage elements, filters and decouplers.
The main features of capacitors are: capacity (1pF-1F), operational voltage
(from 1,5 V up to some kV), operational temperature (from -55 °C to 125 °C),
loss factor, size and shape.
The most frequently used capacitor types in the industry nowadays are:
ceramic, foil, aluminium and tantalum capacitors. The four most important
application fields for capacitor technologies are radio techniques, electrical
power processing, energy storage and power electronics. Except for the first
application field electrolyte-capacitors can be used, so this type of capacitor is
prevalent.
The main advantage of the electrolytic capacitors is the high capacity and
voltage value, which can be attributed to the dielectric layer with very small
thickness, but with very large surface. Their disadvantage is the over voltage
sensitivity.
The main characteristics of the electrolytic capacitors are determined by the
electrolyte, the anode foil and the paper separator.
The electrolyte generally consists of the following components:
• solvent: e.g.: ethylene glycol,
• acids and bases: usually organic,
• different additives.
The electrolytes are characterized by two major features: conductivity and
breakdown potential, both of them dependent on the temperature. The change of
conductivity as a function of temperature decisively affects the electric
parameters of the capacitor. The chemical reactions, which take place inside the
electrolyte, are in direct relationship with the conductivity value at different
temperatures and the quantification of this relationship is important.
The conductivity and breakdown potential of the electrolyte influences the
maximum operating condition of the capacitors. Electrolytes with high
42 Al. Electrolytic Cap. Research and Dev. Time Optim. Based on a Meas. Automation System

conductivity are used in the low voltage capacitors, while the electrolytes with
low conductivity are used in the high voltage capacitors.
The paper is organized as follows. The short descriptions of the
measurement types, which must be automated, are given in Section 2. The
architecture of the proposed measurement system is presented in Section 3. In
Section 4 a characteristic measurement “Conductivity (T)” is presented in detail
in order to demonstrate the program structure and some implementation issues.
Some characteristic measurement results are given in Section 5, and the
conclusions are presented in Section 6.

2. Measurement types
There are two groups of measurements used in electrolyte-capacitor research
and development. The first main measurement group is related to electrolytes,
while the second main measurement group is related to capacitors. The
electrolyte measurement consists of six measurement programs as follows:
• “Conductivity (T)”: measurement on the temperature dependence of
conductivity. This is one of the most important measurements, and is
presented in details in the next section.
• “Ph (T)”: measurement of the pH value as a function of temperature.
The structure of the program is completely similar with the above-
mentioned one, with a difference that pH meter is used instead of
conductivity meter. The measurement is important because the pH
value of the used electrolyte in the electrolytic-capacitor must be in a
specified range.
• “Mixing (pH with single temperature)”: measurement of the pH value
as a function of the concentration of an electrolyte composition at a
specified temperature. As a matter of fact we use this measurement in
order to set up the pH value of the electrolyte.
• “Mixing (conductivity with single temperature)”: measurement of the
conductivity as a function of the concentration of an electrolyte
composition at a specified temperature.
• “Mixing (conductivity with multi temperature)”: measurement of the
conductivity as a function of the concentration of an electrolyte
composition at several temperatures. This and the former measurement
are used to set up the conductivity value of the electrolyte.
• “Spark detector”: measurement of the breakdown potential of the
electrolyte.

The main capacitor measurements are as follows:

 D. Fodor 43

• “Spark detector”: this measurement program is similar to the structure

of the program used for electrolyte measurement. The only difference is
that the object of the measurement is the winding impregnated with the
electrolyte. In that case we want to know the breakdown potential of the
paper impregnated with different electrolytes.
• “ESR (Equivalent Serial Resistance)”: This measurement is mostly used
in order to determine the resistance of the capacitor at different
frequencies and temperatures.
• “Gas pressure”: measurement of the internal gas pressure of the
capacitor in various operating conditions.

Figure 1: System architecture of the automated measurement system.

3. Architecture of the measurement system

The core of the automated system is the NI PXI-1042 chassis [3], with five
modules such as NI PXI-8185 controller [3] (embedded PC), NI PXI-6723 D/A
card [3] for controlling the power supplies, NI PXI-8420 and NI PXI-GPIB for
communicating with the measurement instruments via RS-232, RS-485 and
GPIB [3] (Fig.1). This compact form system gives equivalent performance as a
PC-based data acquisition and measurement control system but has some added
functions, such as trigger buses, increased bus speed, rugged and modular
44 Al. Electrolytic Cap. Research and Dev. Time Optim. Based on a Meas. Automation System

packaging. The different measuring instruments, which can communicate

through RS-232, RS-485 and GPIB have been integrated easily in the system,
with the help of the above mentioned communication cards (NI PXI-GPIB, NI
PXI-8420). In this way also a common development tool, the LabVIEW (for
details see [5]), is available to develop the related communication routines
(drivers) and the measurement programs too. The measurement results are
migrated into a database from where data can be retrieved in non-predicted and
predicted way for evaluation and decision support. (For more information about
PXI see [4])
There are five electrolytic measurements out of six whose architecture show
the same structure (see: Fig. 2). The object of the measurement, the electrolyte,
is located inside a double-jacketed vessel. In the external part of the vessel the
water is circulated by the thermostat, while the electrolyte is inside the vessel.
The most important parameters are provided by the pH meter and the
conductivity meter. Each instrument has got an electrode that can be used to
measure the temperature too. Only one of them is used during an experiment.
The experimental setup requires the simultaneous operation of four individual
instruments such as Thermostat [6], pH meter [7], [8], Conductivity meter [7],
[9], Burette. Controlling and following up the temperature precisely is essential
during the measurements so the Thermostat device is connected to the system
every time. The Burette is only used when the variation of the pH or
conductivity value as a function of a composition must be known. The mixing
type measurements give the relation of the conductivity or pH to the
concentration of the examined component.

Figure 2: Diagram of the electrolytic measurements.

 D. Fodor 45

4. “Conductivity(T)” measurement
The base of the whole software system is a framework which was originally
designed to provide a common user interface for the different measurement
programs. In spite of the fact that all the measurements have an individual
character, they were integrated into the above mentioned framework, in order to
manage the communication ports and instruments, and to provide the parallel
run of the programs.
The measurement system includes at least 27 different measurements, whose
presentation can not be done here. All the measurements have been
implemented in a similar manner. Firstly the user initializes the measurement,
sets the measurement parameters, launches the execution and leaves the
program to run on its own, sending the results of the measurement to a database
system. This process is presented afterwards through the “Conductivity (T)”
measurement.
Two instruments are involved in this measurement: the thermostat and the
conductivity meter.
Before launching the program the user initializes the measurement. During
the measurement the user can choose between two main tab controls in
Fig.3. The “Set parameters” tab contains the parameters set of the
measurement, while the “Measurement” tab shows the state flow of the
measurement, graphs and displays. The program executes the same steps
cyclically. The state diagram shows the current state of the measurement and
the remaining time before the next phase. Firstly the program sets up the
temperature. On the temperature graphs the temperature of the measuring
probe and of the thermostat can be followed up. During the stabilization time
the temperature of the electrolyte becomes the same as the thermostat’s.
Through measuring & saving phase the important conductivity values are
measured and stored locally. The remaining time before the next measurement
is indicated with a progress bar . After the measuring & saving phase the
program calculates the mean value from the stored conductivity data and only
the result is migrated into the database. On the and graphs the
conductivity and the temperature as a function of the number of measurements
can be seen. These are the most important graphs, because the electrolyte
forming can be directly correlated with the temperature. The measurement is
ended after the conductivity is measured at all of the settled temperatures. The
remaining time before the end of the experiment is shown during the
measurements . The program can be stopped with the STOP button .
46 Al. Electrolytic Cap. Research and Dev. Time Optim. Based on a Meas. Automation System

Figure 3: Graphic User Interface (GUI) of the “Conductivity(T)”

measurement program.

5. Results
In Fig. 4 it is demonstrated how decisions are supported by the automated
measurement system. Each dot represents a separate “Conductivity (T)” and
“Sparkling voltage” measurement at 85 degree Celsius- result obtained by the
automated system. By graphical evaluation electrolytes with specific
conductivity or sparkling voltage can be selected for further research.
With the help of the currently automated system the forming of an
electrolyte as a function of the temperature can be followed up. In Fig. 5
measurement results of Electrolyte 1 and Electrolyte 2, which has been obtained
using the “Conductivity (T)” measurement program are demonstrated. The
conductivity of the electrolyte is measured more than once (usually 10 times) at
the specified temperatures. A mathematical mean is calculated from the
 D. Fodor 47

measurement values at one temperature, which is indicated by dots in Fig. 5. As

mentioned, the conductivity of the electrolyte significantly influences the
electric parameters of the capacitor. Conductivity between 900 and 3000 µS/cm
at 30 degree Celsius is considered low, and conductivity values exceeding
10000µS/cm in similar thermal conditions is considered high. As shown in Fig.
5, the conductivity of Electrolyte 1 measured at 30 degree Celsius is 1500
µS/cm and the Electrolyte 2 has 2300 µs/cm at the same temperature which
indicates low conductivity in both cases, resulting in high breakdown potential.
From this reason both electrolytes can be used in high voltage capacitors.
In Fig. 6 the results of a capacitor type measurement – named ”ESR”,
mentioned shortly in the second chapter can be seen. The resistance values of a
capacitor as a function of frequency at different temperatures are given.
During capacitor development we aspire to obtain as low serial resistance as
possible. To achieve this goal a series of experiments have to be done using
different types of electrolyte and paper constructions.
As can be seen in the Fig. 6 the capacitor resistance below zero degrees
Celsius is linearly dependent on frequencies up to 1 KHz. For positive
temperatures it can be observed that the resistance values have linear variation
on frequencies above 1 KHz.
The application areas of the capacitors are based on the frequency
dependence of the resistance value in different temperature ranges.
The “Mixing (conductivity with multi temperature)” measurement is an
extended version of the “Conductivity (T)” measurement, which is completed
with one measurement device: a burette. With the “Mixing” measurement the
conductivity value as a function of a key component concentration of the used
electrolyte can be observed on Fig. 7. First of all the conductivity of the
electrolyte is measured at 25 C°, 40 C°, 60 C° and 85 C° many times (usually
10), without dosing the conductive salt solution. Each dot on Fig. 7 represents a
mathematical mean which is calculated from the measurement values at each
temperature. After that a predefined dosage from the conductive salt solution is
added to the electrolyte and the measurements of the conductivity at different
temperatures are repeated. The measurements are repeated 16 times adding each
time a new predefined conductive salt solution dosage to the electrolyte. Finally
the “Mixing” measurement results in 17 dots at each temperature.
According to the readings the increase of the conductivity of the electrolyte
is in direct relationship to the quantity of the conductive salt solution.
The results can be applied in order to set up the conductivity value of an
electrolyte.
48 Al. Electrolytic Cap. Research and Dev. Time Optim. Based on a Meas. Automation System

Figure 4: Conductivity vs. Sparkling Voltage of various electrolyte at 85 degrees

Celsius.

Figure 5: Result of the measurement of conductivity as a function of temperature of

Electrolyte 1 and Electrolyte 2.
 D. Fodor 49

Figure 6: Result of the capacitor resistance (ESR) as a function of frequency at different

temperatures.

Figure 7: Variation of conductivity of Electrolyte 1 as a function of the conductive salt

solution concentration.

6. Conclusions
More than 27 different electrolyte and capacitor measurements, have been
automated. All the measurements have been implemented in a similar manner.
Firstly the user initializes the measurement, sets the measurement parameters,
launches the execution and leaves the program to run on its own, sending the
results of the measurement to a database system, from where the data can be
retrieved in a predefined or a non-predefined way. After validation of the MAS,
there are many advantages as making the measurements more precise and more
reliable, fault tolerant (i.e.: monitoring functions are implemented like detection
50 Al. Electrolytic Cap. Research and Dev. Time Optim. Based on a Meas. Automation System

of missing of line voltage, open gate etc.), running multiple measurements in

parallel, which all contribute to speed up the research and development of new
components and devices.
The developers do not have to encroach. The developed measurement
system controls and harmonizes the different devices and supervises their work.
The user can simply check the measurement phase by a glance at the screen.
The program estimates and displays the running time of the experiment,
allowing for the researchers working on the Lab to manage the instrumental
resources in time and to schedule in advance (for hours, weeks and months) the
new measurements. Another big advantage is the database system, which stores
the result of each measurement in an easy searchable way. Measurement reports
and diagrams can be made automatically and the results can be reused in later
research.

Acknowledgements
The author fully acknowledges the support of the National Office for
Research and Technology via the Agency for Research Fund Management and
Research Exploitation (KPI), under the research grants GVOP-3.2.2.-2004-07-
0022/3.0 (KKK), GVOP-3.1.1.-2004-05-0029/3.0 (AKF) and more recently for
TAMOP-4.2.1/B-09/1/KONV-2010-0003: Mobility and Environment: Re-
searches in the fields of motor vehicle industry, energetics and environment in
the Middle- and West-Transdanubian Regions of Hungary. The Project is
supported by the European Union and co-financed by the European Regional
Development Fund.

References
[1] Theisbürger, K. H., “Der Elektrolyt-Kondensator”, FRAKO Kondensatoren- und Apparat-
bauen G.m.b.H Teningen, third edition, EPCOS internal document.
[2] Per-Olof Fägerholt, “Passive components” (internal document), 1999.
[3] National Instruments, “Product guide”, 2004.
[4] ***, National Instruments Hompage,
http://zone.ni.com/devzone/conceptd.nsf/webmain/5D1A4BAB15C82CC986256D3A0058
C66C?OpenDocument&node =1525_us.
[5] ***, National Instruments Hompage, http://www.ni.com/labview/whatis/.
[6] ***, “Haake DC30 Circulator- Technical Manual”, Thermo Electron Corporation, 2002.
[7] ***, Metrohm AG Hompage, www.metrohm.com.
[8] ***, “780 pH Meter Instruction Manual”, Metrohm Ltd., Switzerland, 2002.
[9] ***, “712 Conductometer Instruction Manual”, Metrohm Ltd., Switzerland, 2002.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 51-62

Framework for Modeling, Verification and

Implementation of Real-Time Applications
Liviu HAŢEGAN 1, Piroska HALLER 2
1
Technical University of Cluj-Napoca, Cluj-Napoca, România,
e-mail: liviu.hategan@yahoo.com,
2
“Petru Maior” University of Tîrgu Mureş, Tîrgu Mureş, România,
e-mail: phaller@upm.ro

Manuscript received October 01, 2010; revised October 30, 2010.

Abstract: This paper proposes a framework for modeling, simulation and formal
verification of embedded real-time applications running over a real-time multitasking
kernel. We extend a simple real time kernel (RTOS) with synchronous and
asynchronous message passing interface to communicate between tasks and drivers. In
the same time some embedded system’s specific drivers have been added, allowing
unified resource access through these interfaces. The process engineer defines the
control system as a set of tasks interacting with events occurring irregularly in time
(alarms, user commands, communication) and regularly in time (sampled sensor data
and actuator control signals). Taking into consideration both non-preemptive and
preemptive scheduling, we propose two models consisting of networks of timed
automata. Using a model-checker tool (UPPAAL), one can verify the timing and logical
properties of an application, changing the time constrains and priorities. In a priority-
based scheduling scheme, tasks interact both through the scheduler and through the
mutual exclusion mechanism, but there are hidden from the engineer by the framework.
The framework also offers a solution for generating the source code skeleton of the
modeled application. This reduces the risk of errors do to error-prone human coding and
most importantly ensures that the task will have the same behavior as described in the
model.

Keywords: Formal verification, timed automata, real-time applications.

1. Introduction
Real-time embedded systems have become widely used in a large number of
fields, especially in the industrial environment, playing an increasing role in
modern society and are rapidly evolving, growing in complexity. Moreover they
are often used not only by themselves, but in clusters and networks. An
52 L. Hategan, P. Haller

embedded real-time system is in close relationship with the physical

environment it interacts with, it is involved in monitoring and control of
complex physical processes. The applications running on such systems face a
set of constrains like memory, processing power, energy consumption, but
mainly timing constraints. Consequently, a real-time system must exhibit a
predictable behavior, under a given set of conditions the designer must be able
to know if the system will meet its requirements. Developing and running
embedded applications with predictable and controllable behavior requires a
real-time operating system (RTOS). This allows for an application to be
constructed as a collection of tasks managed by the RTOS according to the
scheduling policy, the timing requirements being mapped as task deadlines.
But the engineers that write embedded software are rarely computer
scientists or experts in operating systems. It should be necessary to create an
integrated framework for modeling, simulation, verification and code
generation. On the other hand timeliness, concurrency, bounded response time,
and heterogeneity need to be an integral part of the programming abstractions.
The timed automata formalism is widely used and well-proven in the
description and verification of real-time systems. In [1] timed automata are
proposed for the description of task arrival patterns. The authors present a
unified model for finite control structures, concurrency, synchronization, and
tasks with combinations of timing, precedence and resource constraints.
Another work [2] approaches the problem of formal modeling based on
timed automata of a multitasking application running under a RTOS. The
described model considers an operating system, the application tasks and the
behavior of the controlled environment. In this approach the authors also model
the internal structure, allowing for the verification of not only task
schedulability, but other complex properties like safety and bounded liveness.
In [3] the authors present a framework for modeling and verification of real-
time embedded applications running under a multitasking RTOS kernel. The
authors propose a model of a minimal operating system defined as a network of
timed-automata developed in order to use it as a framework for simulation and
verification of mini real-time applications. The authors chose and analyzed the
FreeRTOS [4] mini real-time kernel either in a non-preemptive (cooperative) or
a preemptive configuration. Access to resources is done using a unified resource
access interface.
In this paper we propose a framework for modeling, simulation and formal
verification of embedded real-time applications and a solution for automatic
source code generation of the modeled application. As in [3], the simulation and
verification can be done using the UPPAAL [5] model checker.
We extended the model described in [3] taking into consideration in more
detail the internal structure of the tasks [6], adding synchronous and
 Framework for Modeling, Verification and Implementation of Real-Time Applications 53

asynchronous message passing interface to communicate between tasks and

drivers, allowing the model to be more expressive. Access to resources is done
using a unified resource access interface, fact that simplifies the task structure
and facilitates automatic source code generation based on the model. The
existing drivers were extended with these interfaces and new drivers were added
for the SAM7-EX256 development board specific devices. For the
implementation we chose the SAM7-EX256 development board. We improved
the unified resource access interface, now consisting of three functions:
Request(), Read() and Write(). This allows for a simplified structure of the
modeled tasks and is necessary for the purpose of translating the tasks from the
model into the corresponding source code.
This paper is organized as follows: section 2 will describe the general
properties and behavior for the task templates; in section 3 we describe the
properties of the resources and the unified resource access interface; section 4
presents the cooperative model (tasks, cooperative scheduler, resources); the
model constructed for the preemptive version of the FreeRTOS kernel is
detailed in section 5; section 6 addresses the issue of simulation and formal
verification of the applications; in section 7 we describe the process of
automatically generating the source code of the application from the model and
finally, in section 8 we state our conclusions.

2. The application tasks

Each task instance is modeled by a timed automaton that is synchronized
using channels. In general, embedded tasks present the following behavior
[1], [3]:

INFINITE_LOOP
- Request access to a resource (blocking call)
- Perform a read/write operation
- Perform a computation
- Request access to another resource (blocking call)
- Perform a read/write operation
……
END_INFINITE_LOOP

Resources are accessed by tasks through blocking request calls. The desired
resource is explicitly specified through its RID (resource ID) [7]. After making
a request call, the task enters in blocking state, where it waits for the resource to
become available.
The FreeRTOS tasks are prioritized.
The computations performed by the tasks are characterized by a worst case
execution time (WCET) and a best case execution time (BCET). This is a con-
54 L. Hategan, P. Haller

sequence of branching instructions in the computations. Also, the running task

will be delayed by the occurrence of interrupts generated by the resources. The
ISR execution time will be added to the current task's WCET and BCET [2].

3. The resources
In constructing the general resource model, we considered that the
following: resources are reusable and can be shared (but only one task can have
access to a resource at any given time), a task can request a single resource at
one time and in the request call the resource is explicitly specified through its
resource ID. Every resource has a minimal inter-arrival time (the MIAT). A
resource can unblock a waiting task and provide data at any time after its MIAT
expires.

3.1 The unified resource access interface

In our implementation on the SAM7-EX256 development board, resources
provide interrupts that are managed by drivers. Depending on the peripheral,
data is read from registers and stored in queues when the interrupt occurs for an
input peripheral or data is sent when an output peripheral is ready to accept it. A
task that is waiting on the resource's queue will immediately be unblocked.
As we mentioned in the previous sections, access to a particular resource is
performed trough the unified access interface. Each resource has a state variable
associated. The read and write operations are performed using the state
variables (or any other user-defined data structure that corresponds). The
interface is composed of the following functions:
- Request ( RID )
- Read ( RID, var_rid, nr )
- Write ( RID, var_rid, nr )

When requesting a resource, it must be explicitly specified through its RID.

If the resource is not available (is being held by another task) the calling task
will block until the requested resource becomes available. The Read() and
Write() are nonblocking calls, they will be performed after the access to the
requested resource is granted and provide the means for read/write operations.
Besides these resource RIDs, the function calls must also include the state
variables associated with the resource (var_rid) and the number of bytes to be
read or written. The functions implemented in the unified access module are
also present in the model and permit a simple description of concurrent access
to the desired resources and read/write operations.
 Framework for Modeling, Verification and Implementation of Real-Time Applications 55

4. The cooperative model

4.1 The cooperative application tasks

Considering the general task form described in the previous section, we

present a simple task model that requests access to a resource (RID), performs a
read operation and executes a computation. The task in pseudocode is:

Task1{
Loop
{ Request(RID);
Read(RID, var_rid, nr);
computation();
}}

The task automaton (Fig. 1) is synchronized with the models of the

scheduler and resources via channels.
The RUN locations are characterized by a best-case and worst-case
execution time (WCET, BCET). This is modeled by the location's invariant
(y<=W), and the guard on the outgoing transition (y>=B). REQ_BCET[RID],
REQ_WCET[RID], READ_BCET[RID] and READ_WCET[RID] are constants
defined in the model and they are the BCET and WCET for requesting and
reading data in the case of the resource represented by RID. While being in a
running state, a task can be interrupted by a resource's interrupt service routine
(ISR). The execution time of the ISR is added to the W and B variables.
BLOCKED: the state is entered when the task requests a particular resource
through the request[pid][RID]! channel. The state is left when the resource
becomes available (the event_or_timer[RID]? channel is activated).
READY1
schedule[pid]?
W+=REQ_WCET[RID],
B+=REQ_BCET[RID],
y=0,running=1,t=0
RUN3
y>=B
y<=W
RUN1 y<=W
computation(),
W+=REQ_WCET[RID], y>=B
B+=REQ_BCET[RID], request[pid][RID]!
y=0 ready[pid]=0,running=0,
y>=B
W=0,B=0
Read (RID, var_rid, nr),
W=WCET3, B=BCET3, BLOCKED
y=0 event_or_timer[RID]?
ready[pid]=1,t=0
schedule[pid]?
RUN2
READY2
y<=W
W+=READ_WCET[RID],
B+=READ_BCET[RID],
running=1,y=0

Figure 1: Simple cooperative task model.

56 L. Hategan, P. Haller

If there is no task running or ready, the FreeRTOS kernel schedules the Idle
Task (Fig. 2), which is always available for scheduling. Following the general
task form, the Idle Task periodically requests the NULL resource, yielding
processor control.

y >= B
request[pid][RESOURCE_NULL]!
W=0,
B=0

schedule[pid]?
READY RUN
y = 0,
y <= W
W+=WCET,
B+=BCET

Figure 2: The Idle Task model.

4.2 The cooperative scheduler

In the cooperative behavior the running task has full control of the processor,
regardless of its priority, until it makes a blocking call. The task can explicitly
invoke the scheduler by calling the taskYIELD() macro or by requesting access
to a resource (the Request(RID) function).
The cooperative scheduler is described by a timed automaton that presents
three states: INIT, SELECT and IDLE (Fig. 3).

pid : id_task
pid == GetNextTask()
schedule[pid]!

pid : id_task,
f : id_all Init(),
request[pid][f]? x=0
x = 0,
IDLE res_all[f] = 1, SELECT INIT
Rotate(pid) x<=5

Figure 3: The cooperative scheduler model.

INIT: the necessary hardware settings and initialization of task priorities and
data structures take place. Because the state is committed, the scheduler will
leave this state immediately at startup.
SELECT: the ready task with the highest priority is chosen for scheduling
(the GetNextTask() function). The invariant x<=5 specifies the time needed to
select the next task; the value can be changed to match the actual physical time,
which is hardware-dependant.
 Framework for Modeling, Verification and Implementation of Real-Time Applications 57

IDLE: the previously chosen task is running. In order to have an equal

chance at scheduling for the tasks with the same priority, the Rotate() function
is called when the scheduler exits from the IDLE state.

4.3 The system resources

Fig. 4 illustrates the general resource model.

x>all_period[rid]
event_or_timer[rid]!
x=0,
RESOURCE W+=RES_WCET[rid],
B+=RES_BCET[rid]

Figure 4: Resource model.

The MIAT values for all of the system's resources are stored in the array
all_period[NR_RESOURCES]. The MIAT is modeled by the guard
x>all_period[rid]. The waiting task is unblocked via the event_or_timer[rid]!
channel. The RES_WCET[pid] and RES_BCET[pid] constants are used to delay
the task interrupted by the resource ISR.
In case a task must execute an action periodically, at strict interval, it can
utilize the timer resource (Fig. 5). The timer unblocks a waiting task when the x
clock has the same value as all_period[tid]. The constant tid represents the
timer's RID.

x >= all_period[tid]

res_all[tid] == 0
TIMER
x = 0, missed_timer[tid]++,
x <= all_period[tid]
W+=RES_WCET[tid], SYNC
B+=RES_BCET[tid]
res_all[tid] == 1
event_or_timer[tid]!
x=0, W+=RES_WCET[tid],
res_all[tid]=0, B+=RES_BCET[tid]

Figure 5: Timer model.

The initial state TIMER is left when the predefined period expires. The
SYNC state is committed so it is left immediately, the automaton unblocking
any waiting task. In order to avoid system deadlock, the timer is allowed to
expire even if none of the tasks are blocked waiting for it.
58 L. Hategan, P. Haller

The resource model is identical for both preemptive and cooperative

systems.

5. The preemptive model

5.1 The preemptive application tasks

In addition to the READY, RUN and BLOCKED states presented by the

cooperative tasks, the preemptive versions also have a SUSPENDED state. This
state is entered when the task is preempted (via the suspend[pid]? channel). The
invariant y'==0 will stop the y clock while the automaton is in a suspended
location. Upon rescheduling, the clock is restarted (the y'==1 expression). Fig.
6 presents the preemptive version of the simple read-request-computation task
model described in section 4.1.

schedule[pid]?
READY1
W[pid]+=REQ_WCET[RID],
B[pid]+=REQ_BCET[RID],
y=0,running=1,t=0

(y<=W[pid])&&(y'==1) suspend[pid]?
(y<=W[pid])&&(y'==1)
RUN3 running=0 SUSPENDED1
SUSPENDED3 suspend[pid]? RUN1
y>=B[pid]
y'==0 running=0 y'==0
computation(),y=0,
schedule[pid]?
W[pid]=REQ_WCET[RID], running=1
schedule[pid]? B[pid]=REQ_BCET[RID]
running=1
y>=B[pid]
request[pid][RID]!
y>=B[pid]
ready[pid]=0,
Read(RID, var_rid, nr), running=0, W[pid]=0, B[pid]=0
W[pid]=WCET,
B[pid]=BCET, y=0
BLOCKED
event_or_timer[RID]?
SUSPENDED2 suspend[pid]? ready[pid]=1,t=0
y'==0 running=0 schedule[pid]?
READY2
y=0,running=1,
schedule[pid]? RUN2 W[pid]+=READ_WCET[RID],
running=1 (y<=W[pid])&&(y'==1) B[pid]+=READ_BCET[RID]

Figure 6: Simple preemptive task model.

The Idle Task is the same as in the case of the cooperative model, except for
the fact that it doesn't suspend itself by requesting the NULL resource, but it is
preempted by the scheduler.
 Framework for Modeling, Verification and Implementation of Real-Time Applications 59

5.2 The preemptive scheduler

The preemptive scheduler model is illustrated in Fig. 7.

pid : id_task
pid == GetNextTask()
schedule[pid]!
current_pid = pid

pid : id_task,
f : id_all SELECT
y<=2 x >= iTickRate
request[pid][f]?
TICK
y = 0, tick!
IDLE res_all[f] = 1,
Init(), x=0
y=0 INIT x <= iTickRate
Rotate(pid),
tick? current_pid=GetNextTask()
y=0
suspend[current_pid]!
y = 0,
SUSPEND Rotate(current_pid),
y<=2 current_pid=GetNextTask()

(a) (b)

Figure 7: The preemptive scheduler model (a) and the tick interrupt model (b).

The preemptive scheduler can periodically perform a context switch,

temporarily suspending the running task in favor of an equal or higher priority
one. It does so by using the tick interrupt. Each time the interrupt occurs, the
kernel determines if a context switch must take place. This action is performed
in the SUSPEND state.

6. Simulation and formal verification

Simulation and formal verification can be performed using the UPPAAL [8]
integrated simulation and verification tools. For formal verification, the
properties required for an application to function according to requirements
must be expressed in UPPAAL's CTL subset [9]. The verifiable properties are:
reachability, safety, liveness, bounded liveness and deadlock-freeness.
Reachability properties are checked to see if it is possible to reach a state
where a formula p is satisfied, for example:
- E<> Task1.RUN1 - checks if Task1 can ever reach the RUN1 state;
- E<> Task1.SUSPENDED1 - verifies if there is a possibility that
Task1 will ever be suspended, a context switch taking place.
Safety properties require that a formula p is satisfied in all reachable states
or, that there is a path in which p is always true:
- A[] not (Task1.running and Task2.running) - two different tasks can
not be running at the same time;
60 L. Hategan, P. Haller

- E[] Task1.READY2 imply Task1.t<=500 - there is a path where

Task1 will not spend more than 500 time units in the READY2 state.

Liveness properties require that, in all cases, the system will eventually reach
a state where a formula p is true. Another form is that if a formula p is true,
another formula q will become true eventually:
- A<> Task1.RUN1 - Task1 will inevitably be in the RUN1 state at
some point;
- Timer(1).SYNC --> Task5.RUN2 - considering a blocked task
waiting for a timer, if Timer(1) expires then Task5 will inevitably be
scheduled.

Bounded liveness properties can be formulated “whenever p becomes true, q

becomes true within the time limit t”:
- Timer(1).SYNC --> (Task5.RUN2 and Task5.t<=200) - when
Timer(1) expires Task5 will be scheduled within 200 time units.

Deadlock-freeness [7]: A[] not deadlock.

To verify the time constrains, the execution time of the elements regarding
our implementation for the SAM7-EX256 development board (task scheduling
time, resource ISR execution time, resource access and read/write operations
time, etc.) were measured using the system's physical timers and were
introduced in the models.
The interrupt service routine execution time:
- Analog/digital converter: RES_BCET=RES_WCET=4,8 μs;
- Key pressed: RES_BCET=5,2 μs; RES_WCET=10,2 μs;
- Joystick: RES_BCET=5,5 μs; RES_WCET=10,4 μs;
- Timer0 ÷ Timer2: RES_BCET=RES_WCET=2.5 μs.

Driver unified resource access:

- Analog/digital converter: RES_BCET=RES_WCET=26,5 μs;
- Display: RES_BCET=RES_WCET=30 μs;
- External storage: RES_BCET=RES_WCET=30 μs;
- Keys: RES_BCET=RES_WCET=15,4 μs;
- Joystick: RES_BCET=RES_WCET=15,4 μs.

Write operation:
- Display: RES_BCET=390 μs (1 char.); RES_WCET≈7 ms (20 char.);
- External storage: RES_BCET=54 μs (1 char.) ;RES_WCET ≈100 μs(128 char.).
Read operation:
- Analog/digital converter: RES_BCET=RES_WCET=26,5 μs;
- Keys: RES_BCET=27,2 μs; RES_WCET=28 μs;
- Joystick: RES_BCET=27,2 μs; RES_WCET=28 μs.
 Framework for Modeling, Verification and Implementation of Real-Time Applications 61

7. Source code generation

We have developed a code generator application that, based on the model
presented as input (XML file), recognize the states and functions calls of each
task and output the corresponding source code skeleton. In the case of the
simple request-read-computation task presented in the previous sections the
following code is generated:

void Task1( void *pvParameters )

{
for( ;; )
{
Request(RID);
Read(RID, var_rid, nr);
//!!computation();
}
}

The generator creates a header file containing the declarations for all the
tasks and a C source code file with their implementation. These resulting files
can be compiled in a project along with the FreeRTOS source code. Before
compilation, all that remains is to add the computational blocks containing the
algorithms that manipulate the data. The task code is identical in both cases,
cooperative and preemptive.
Alongside the models and source code generator, we constructed a source
code project that contains the FreeRTOS source, drivers including interrupt
mechanisms for the system's resources and the module for unified peripheral
access.
The source code project also includes a special task that can be used to
directly measure the time necessary for a sequence of code to execute on this
physical system (for use in the model). Also, we included functions to facilitate
the conversion of data from the type specific to a particular resource to another
resource's type. For example, to convert and copy the data from the state
variable associated with the system's analog-to-digital converter to the state
variable of the LCD display, one can use the function ADCtoLCD(res_adc,
res_lcd ). The available functions and they execution time is:
- INTtoLCD( Integer, res_lcd ); RES_BCET=7,8 μs; RES_WCET=44,1 μs;
- INTtoSD( Integer, res_sd ); RES_BCET=66,5 μs; RES_WCET=305 μs;
- ADCtoSD( res_adc, res_sd ); RES_BCET=45,3 μs; RES_WCET=333 μs;
- ADCgetINT(res_adc); RES_BCET=RES_WCET=31,4 μs;
- ADCgetSTR( res_adc); RES_BCET=17,5 μs; RES_WCET=350 μs;
- LCDgetSTR( res_lcd); RES_BCET=RES_WCET=18 μs;
- SDgetSTR( res_sd); RES_BCET=RES_WCET=47,2 μs.
62 L. Hategan, P. Haller

These functions are also present in the model, allowing it to be more

expressive and further simplifying the code of application tasks.

8. Conclusions
This paper presents a framework that can be used to model, verify and
implement real-time multitasking applications. The operating system, resources
and application tasks are modeled by timed automata. This approach allows for
the system's simulation and verification before the actual implementation,
permitting the early detections of any undesirable behavior. The unified
resource access interface and the code generator make possible the automatic
generation of the modeled (and verified) application's source code, avoiding
most of the error-prone human coding. Because the method is susceptible to
state space explosion, the model must be abstract as much as possible, making a
compromise between model complexity and its state space size.

References
[1] Fersman, E., “A generic approach to schedulability analysis of real-time systems”, Ph.D.
Thesis, Faculty of Science and Technology, Uppsala University, November 2003.
[2] Waszniowski, L., and Hanzalek, Z., “Formal verification of multitasking applications based
on timed automata model”, Real-Time Systems, vol. 38, no. 1, Springer-Verlag, pp. 39-65,
2008.
[3] Zaharia, T., and Haller, P., “Formal verification and implementation of real time operating
system based applications”, in Proc. of the 4th IEEE International Conference on Intelligent
Computer Communication and Processing, Cluj-Napoca, Romania, pp. 299-302, 2008.
[4] FreeRTOS – portable, open source, mini Real Time Kernel; http://www.freertos.org
[5] UPPAAL – tool box for modeling and verification of real-time systems modeled as
networks of timed automata; http://www.uppaal.com
[6] Liu, J.W., “Real-time systems”, Prentice-Hall, Inc., Upper Saddle River, New Jersey 2000.
[7] Li, P., Ravindran, B., Suhaib, S., and Feizabadi, S., “A formally verified application-level
framework for real-time scheduling on POSIX real-time operating systems”, IEEE Trans.
Software Eng. vol. 9, no. 30, pp. 613-629, 2004.
[8] Hessel, A., Larsen, K. G., Mikucionis, M., Nielsen, B., Pettersson, P., and Skou, A.,
“Testing real-time systems using UPPAAL”, Formal Methods and Testing, Springer-
Verlag, pp. 77-117, 2008.
[9] Behrmann, G., David, A., and Larsen, K. G., “A tutorial on UPPAAL”, In Proceedings of
the 4th International School on Formal Methods for the Design of Computer, Communi-
cation, and Software Systems (SFM-RT'04). LNCS 3185, Springer-Verlag, 2004.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 63-72

Application Development in Database-Driven

Information Systems
Marius MUJI
Department of Electrical Engineering, Faculty of Engineering, Petru Maior University
of Tîrgu Mures, Tg. Mureş, e-mail: marius_muji@yahoo.com

Manuscript received October 1, 2010; revised October 18, 2010.

Abstract: The relational model provides extensive support for data integrity
constraints (i.e. business rules) specification, as an integral part of the data model.
Current Relational Database Management Systems (RDBMS), however, cover just
partially the various categories of data integrity constraints, mostly those directly related
with the database structure (e.g. entity integrity, referential integrity). The rest of them
are delegated to the application languages. Consequently, they are usually defined in a
function-oriented approach (e.g. the object-oriented technology), loosing their direct
link with the data model – with all the negative consequences in terms of system
scalability and logical data independence. The present paper proposes a data-oriented
approach for the development of the external level of database systems. Under the
proposed model, the external data is structured only by means of ordered sets of tuples
(i.e. arrays of tuples), and the corresponding business rules (i.e. the presentation rules)
are treated as external schema integrity constraints. Consequently, the application
developer is able to define the user views of the system in a declarative fashion, similar
to the relational database design. The immediate advantage is that he or she gains a data
designer perspective, rather than one of a programmer. The essentiality (i.e. the unique
data constructor) of the model facilitates a seamless integration with the relational
model, an entity-relationship graphical representation, and the complete automation of
the user interface development.

Keywords: Logical design– data models, schema and subschema.

64 M. Muji

1. Introduction
Database-driven information systems are developed around an integrated
and shared source of data. The integration is important when somebody needs a
general view of the system: for example, a manager who wants to track an item
from the supplier to the end-client, spanning the procurement, production and
sales activities of a company. This is why, regardless how many individual
views we have about an organization’s data, there is always needed an
integrated, general view of the entire database. On the other hand, it is also
important for initial system development and for long-term data management
purposes to work with data representations which are not dependent on the
physical storage equipment.
These requirements led to the ANSI/SPARC three levels architecture [2, 3]
(see Fig. 1), which makes a clear distinction between the physical and the
conceptual (i.e. logical) representation of the system, and between the general,
integrated community view and the individual views of the system, respectively.
The physical-logical separation provide physical data independence, which
basically means that the applications would not be affected by changes at the
physical data representations (for hardware upgrade purposes, for example); the
community-individual views separation provides logical data independence,
which means that the system could grow (through some new user views or
modification of the existent ones) without affecting the applications
corresponding to the user views that remain unchanged.

EXTERNAL
LEVEL User view 1 User view 2 User view N
...

CONCEPTUAL
Community view
LEVEL

INTERNAL LEVEL Physical view

Figure 1: The three levels architecture.

The relational model provides the theoretical support for the development of
information systems in accordance with the three levels architecture. Thus,
Relational Database Management Systems are currently the technology of
choice for the development of the physical and conceptual level, sharing with
the application languages the development of the external level.
 Application Development in Database-Driven Information Systems 65

In this context, the database professionals are traditionally responsible for:

• the data structures at the conceptual and physical level;
• some of the integrity constraints at the conceptual level (i.e. the database
rules), like: type constraints, entity integrity constraints, referential
integrity constraints;
• some of the data structures of the external level (e.g. relational views,
parameterized relation-valued operators [7]/stored procedures).
The application professionals are, in turn, responsible for:
• the remaining part of the integrity constraints for the
conceptual/community data (i.e. the application rules);
• all the data structures of the external views – even when the DBMS
provides a layer of data at the external level (e.g. relational views), the
application languages need to redefine the entire external view using their
own data constructs;
• the presentation rules [6] implementation, i.e. the end-user interface,
including CRUD (create, retrieve, update, and delete) operations, and
display customization (e.g. field labels, field alignment, background and
foreground colors, etc.).
The current trend in application development is determined by a significant
pressure coming from the programming community, which promotes an object
oriented approach for the entire architecture of the information system.
Consequently, the data structures and the business rules are usually defined in a
function-oriented approach [12], loosing their direct link with the data model
[6] – with all the negative consequences in terms of system flexibility and
logical data independence [7].
By contrast, we propose a data-oriented approach for the development of the
information systems, including the external level of the ANSI/SPARC
architecture. Thus, we defined a presentation model, which preserves the
essentiality of the relational model [4], i.e. the existence of a unique data
constructor (in our case, the array of tuples), and prescribes a declarative
solution for the presentation rules specification, perceived as external view
integrity constraints.
The model introduces a clear separation between the display-related
presentation rules (e.g. field labels, field alignment, background and foreground
colors, etc.), and data-related presentation rules (e.g. data filtering, master-detail
navigation, data ordering). CRUD operations are accomplished through the
standard behavior of the array constructor. For any CRUD operation initiated by
the end user, the system initiates automatically the invocation of some operators
from the underlying levels, which actually realize the mapping between the
presentation level and the lower levels of the system (i.e. the lower external sub-
levels and/or the conceptual level – see Fig. 2).
66 M. Muji

Section 2 provides a discussion about the external level of a database-driven

system. Section 3 presents our presentation level modeling approach, followed
by an example in Section 4. We conclude with the advantages of the proposed
approach, and some possible applications.

2. The external level - a closer look

While the external level of the three levels architecture is usually split in
multiple sub-levels [3] (see Fig.2), the presentation level of the system is
actually the outermost sublevel, which contains the external data as seen by the
end user (i.e. the external views of the system).
PRESENTATION LEVEL

Graphical
User GUI 1 GUI 2 GUI N
...
Interface

External User view 1 User view 2 User view N

...
View

External External External

sub-level 1.m1 sub-level 2.m2 ... sub-level n.mn
EXTERNAL
LEVEL

External External External

sub-level 1.1 sub-level 2.1 ... sub-level n.1

CONCEPTUAL
Community view
LEVEL

INTERNAL LEVEL Physical view

Figure 2: The external sub-levels of the system.

Some of the external sub-levels, i.e. those closer to the conceptual level, are
usually implemented under the relational model, through relational views and/or
relational operators (e.g. stored procedures). The external sub-levels closer to
the end user are built under the theoretical model employed by the application
languages (in most cases, object-oriented). The well-known impedance
mismatch issue is in fact a measure for the lack of compatibility between the
two theoretical models. The major difference is determined by the switch of
focus from data to function: the data constructs defined by the database
 Application Development in Database-Driven Information Systems 67

designers are spread among multiple function-oriented software constructs by

the application developers [13].
The majority of the mapping solutions employed today to overcome the
impedance mismatch have the aim to provide the application developer with the
means for accessing the lower (relational) levels of the system transparently,
using only the concepts and tools specific to the application languages. Even
when some specific concepts of the conceptual level models are introduced (e.g.
data entities and relationships) [1], the main purpose is to ‘push’ the mapping
layer as ‘low’ as possible.
We follow the opposite approach, which considers that the relational model
is better suited not only to design the persistent data structures of the conceptual
level, but also to build and manipulate the data structures of the external level.
However, the end user’s perception of data often implies the existence of a
current element and a certain inspection order for a given set of data. It follows
that, at least for the presentation level, there is a need for some non-relational
features. At the same time, we consider that the essentiality of the relational
model (i.e. the existence of a unique, essential, data constructor [4]) would
provide, also, at the presentation level important advantages related to
impedance mismatch and interface automation. Consequently, our model
considers the array of tuples as its unique data constructor. It could have been a
list, or any other collection type, as well – to cite from reference [7], any
preference is just “a purely psychological decision – there is no logical reason
for preferring (say) an array over a list”.

3. The user view from a data oriented perspective

From the end user’s point of view, the general behavior of a typical
application consists on a limited set of actions related to data entities. In fact,
there are just four basic actions, or data-function interfaces [12], classically
known as CRUD operations: create, retrieve, update, and delete.
Since it requires a more complex analysis, we’ll discuss first the issues
related to data retrieval. In this regard, the end user can take the following
typical actions:
1. to identify one element in a set, by means of some unique property or
set of properties which distinguishes that particular element from all the
rest in the set;
2. to determine a subset of a set, based on some filtering criteria, namely
some common properties of the subset elements;
3. given one element in a set A, and an existing relationship defined from
A to B, to identify all the related elements in B, under the rule that
defines the relationship (e.g. master-detail navigation);
68 M. Muji

4. to display the elements of a set in a particular order.

Let us consider that all the data ‘seen’ by the end user through one particular
user view, is composed at any given time by ordered sets of tuples (i.e. arrays of
tuples). The user may also be aware about some existing relationships between
two sets, under the definition provided on the reference [5]: “Let A and B be
sets, not necessarily distinct. Then the relationship from A to B is a rule pairing
elements of A with elements of B.” Note that we discuss about directed
relationships, so a relationship defined from A to B will be different from
another relationship defined from B to A.
If we consider that any filtering value, which is to be applied to the set X, is
seen as an element of another set Y, when a relationship was defined from Y to
X, then the rule which defines the relationship is the filter itself. Similarly, a
change of the display order of a set A may be also accomplished by changing
the current element of another set B (which contains the ordering sequences of
choice), when a relationship is defined from B to A. Based on the relationship
definition, and on the current element of B, the system will reorder the set A
accordingly (more accurate: the ordered collection representing A is (re)created
based on the relationship definition).
Under this approach, we are able to design the entire presentation level only
by means of (ordered) sets and relationships between sets. The processing of all
the data requests at the presentation level is hidden inside the defining rules for
set relationships. The only functionality kept at the presentation level is related
to the automatic enforcement of the relationship rules, i.e. the automatic recall
of the defining operator attached to the dependent array, when the current
element of the parent array changed its value.
Considering the update operations (i.e. insert, update, and delete), our
presentation model does not require special features, other than the existing data
access solutions employed by the application languages. However, in order to
preserve the uniformity of the model, and to increase the level of logical data
independence, the recommended solution implies the existence of a level of
update operators (e.g. stored procedures, or any other application procedures), at
the interface with the underlying levels of the system. At the presentation level,
we’ll have to declare the procedure’s name, and the name and type of its
parameters.

4. An example
The following example is inspired from the chapter about presentation rules
in reference [6]. Some details were added to enable a better presentation of our
approach (see Fig. 3).
 Application Development in Database-Driven Information Systems 69

Suppose that we have a user view that exposes to the end user data about
customers, orders, and order details. Suppose that the user will have to be able
to see at any time all the customers which simultaneously satisfy the following
conditions:
• they have a credit limit less than a certain value;
• they are located in a specific region;
• they can be ordered by name, by credit limit, or by the total value of
their orders;
• customers whose accounts are overdue must be displayed in red.
Likewise, the user should be able to see, also, at any time, the orders which
simultaneously satisfy the following conditions:
• they belong to the current customer;
• their issuing date is in a certain period, say after a start_date and before
an end_date, specified by the user;
• they can be ordered by date, value-ascending, or value-descending;
• rush orders must be displayed before regular orders.
When the user inspects a specific order, the system should provide all the
order_details that belong to that particular order. Those details should be
displayed in their part number order.

Customer
Credit limit Region
sequence

Order
Customer Time frame
sequence

Order

Order details

Figure 3: A user view example.

In Figure 3, all data structures are arrays. Some of them represent

application data (i.e. filtering and/or ordering conditions), like credit limit,
customer sequence, time frame, order sequence. They are not dependent on
other data, so their defining functions don’t have parameters.
70 M. Muji

The array named region takes its values from the conceptual level (possibly
through a relational view), but its content doesn’t depend on any other data
structure from the user view.
The customer data contained by the customer array depends on the current
region chosen by the user from the region array, on the current customer
sequence chosen by the user in the customer sequence array, and also on the
value provided by the user in the credit limit array (the credit limit array will be
a special case of an array with one tuple and one attribute, but still an array and
not a simple scalar variable, in order to preserve the essentiality of the external
view model). This is why the defining operator of the customer array should
have three parameters, which will automatically take their values at run time
from the current tuples in the region, customer sequence, and credit limit arrays,
respectively, at any refresh of the customer data.
The list of customer orders exposed to the user at a given moment, contained
by the order array, depends on the current elements of the customer array, the
time frame array, and the order sequence array. Consequently, the defining
operator of the order array should have at least three parameters, one for every
parent array. In fact, for the present example, we may consider four parameters:
one for the link with the customer array (e.g. customer_id), two for the link with
the time frame array (e.g. start date, and end date), and one for the link with the
order sequence array (e.g. order_sequence_no).
As required, the order details array will contain at any moment all the details
of the current order from the order array. The rule that the details should always
be ordered by their part number is specified inside the defining function of the
order details array, and will remain transparent at the user view design level.
We should also be able to provide solutions for the presentation rules that are
not related with relationship definitions:
• “customers whose accounts are overdue must be displayed in red” – for
this rule, we need to introduce an attribute in the customer array, which
would allow the distinction of the ‘red’ customers, so that, at the display
level, while defining the graphical object (e.g. the grid, or the list)
which displays the customers data, we’ll be able to incorporate this
presentation rule in a straightforward manner (i.e. declaratively, if
possible);
• “rush orders must be displayed before regular orders” – this rule is
implemented inside the defining function of the order array (which is
completely transparent for our model) .
So, under the proposed model, the developer is able to design the
presentation level declaratively, just specifying:
• the declaration of all the array structures: array name, attribute names,
data types;
 Application Development in Database-Driven Information Systems 71

• the defining operator of every array;

• the link between every parameter of any defining operator and its
corresponding attribute from the parent array;
• the update procedures, their triggering events, and the links of their
parameters with the corresponding attributes.
Our user view’s dependency graph was represented graphically in Fig. 3
using arrows, oriented from parent to child, but it could have been used any
other entity relationship graphical notation (e.g. crow foot, IDEF1X, IE, etc.).
Thus, the user views will have the same (E-R like) graphical representation as
the conceptual level. The only difference is that instead of foreign key
relationships, we have pairs of defining operator parameters and attributes of
the parent array(s).

5. Conclusions
There is a clear need for a data-oriented approach in application engineering.
The software engineering field is now dominated by the new trend introduced
by the OMG’s Model Driven Architecture [14], which has a strong object
oriented bias. The position sustained by this paper is that the application
development should be not only model-driven, but data-model-driven [10, 11].
The paper introduces a data-oriented model for the development of the external
level of database systems, which considers the presentation level as the only
required data layer above the relational data model. Moreover, this should be a
thin layer, with the unique purpose of data presentation, which doesn’t need to
address any business logic other than the presentation rules [6].
The standard behavior and the essentiality of our model enable the
automation of the presentation level development. At the same time, the
mapping operators (defined at the lower levels and called at the presentation
level to promote the CRUD operations to the conceptual level) are the key for
the provision of logical data independence at the presentation level. This
constitutes the major step forward from the previous attempts to automate the
interface, which failed to provide an appropriate degree of logical data
independence at the external level of the system. Trying to generate the
interface based on various entity-relationship patterns existent at the conceptual
level, and assuming that the user views are just sub-schemas of the conceptual
level [15, 16, 18], they become useless as soon the external level has multiple
sublevels, i.e. the presentation data is obtained from the conceptual data through
a series of complex operations – which is always the case for large, integrated
information systems.
The foreseen applications of the presentation model are related primarily to
the application development for database-centric systems (e.g. enterprise
72 M. Muji

resource planning systems, e-commerce systems, etc.). CASE tools which

support entity-relationship diagrams represent, also, an important area for our
model implementation.
Future work will concentrate primarily on the development of interface
automation tools, designed in an object-oriented approach, and implemented
with general purpose third-generation languages (e.g. Java, C#). In a long term
vision, the presented model could be used in data-model driven methodologies
for declarative development of database-centric applications.

References
[1] Adya, A., Blakeley, J. A., Melnik, S., and Muralidhar, S., “Anatomy of the ADO.NET
entity framework”, ACM SIGMOD International Conference On Management Of Data.
Beijing, China, 2007, pp. 877-888.
[2] ANSI/X3/SPARC Study Group on Data Base Management Systems. “Interim Report”,
ACM SIGMOD Bulletin, no. 2, 1975.
[3] Date, C. J., “An Introduction to Database Systems (8th edition)”, Addison-Wesley, 2003.
[4] Date, C. J., “Date on Database: Writings 2000-2006”, Apress, 2006.
[5] Date, C. J., “Logic and Databases: The Roots of Relational Theory”, Trafford Publishing,
2007.
[6] Date, C. J., “What Not How: The Business Rules Approach to Application Development”,
Addison-Wesley, 2000.
[7] Date, C. J., and Darwen, H., “Foundation for Future Database Systems: The Third
Manifesto (2nd Edition)”, Addison-Wesley, 2000.
[8] Halle, B., “Business Rules Applied: Building Better Systems Using the Business Rules
Approach”, Wiley, 2001.
[9] Hay, D. C., “Data Model Views”, The Data Administration Newsletter - TDAN.com, Apr.
2000.
[10] Lewis, B., “Data Lineage: The Next Generation”, The Data Administration Newsletter -
TDAN.com, Aug. 2008.
[11] Lewis, B., “Data-Oriented Application Engineering: An Idea Whose Time Has Returned”,
The Data Administration Newsletter - TDAN.com, Jan. 2007.
[12] Lewis, W. J., “Data Warehousing and E-Commerce”, Prentice Hall PTR, 2001.
[13] Lewis, W. J., “E-Commerce Vs. Data Management”, The Data Administration Newsletter –
TDAN.com, Jan. 2002.
[14] Model Driven Architecture. http://www.omg.org/mda/
[15] Pizano, A., Yukari, S., and Atsushi, I., “Automatic generation of graphical user interfaces
for interactive database applications”, Conference on Information and Knowledge
Management, Washington, D.C., 1993, pp. 344-355.
[16] Rollinson, S. R., and Roberts, S. A., “A mechanism for automating database interface
design, based on extended E-R modelling”, Advances in Databases. s.l. : Springer Berlin /
Heidelberg, 1997, pp. 133-134.
[17] Ross, R. G., “Principles of the Business Rule Approach”, Addison-Wesley Professional,
2003.
[18] Rowe, L. A., and Shoens, K. A., “A form application development system”, ACM SIGMOD
International Conference On Management Of Data., Orlando, Florida, 1982, pp. 28-38.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 73-86

Exambrev - Integrated System for Patent Application

Attila ASZALOS, József DOMOKOS, Tamás VAJDA,
Sándor Tihamér BRASSAI, László DÁVID
Department of Electrical Engineering, Faculty of Technical and
Human Sciences, Sapientia University, Tîrgu Mureş,
e-mail: aszi.atti@hotmail.com; domi@ms.sapientia.ro;
vajdat@ms.sapientia.ro; tiha@ms.sapientia.ro; ldavid@ms.sapientia.ro

Manuscript received October 1, 2010; revised October 30, 2010.

Abstract: In this paper we present the design and development of a patent

application, examination and evaluation system based on the JEE platform.
In the Introduction part of the paper we present the necessity of a patent application
system, and various organizations dealing with patent data and classification.
The second section of the paper presents the architecture of our system and its
functionalities, which include online patent request registration, reduction of the time
required for the application examination by automatic and semiautomatic verification of
the formal aspect of a patent application. The system helps the inventors in the
International Patent Code assignment procedure and provides the possibility for the
domain experts to search for similar technical solutions in the Romanian State Office
for Inventions and Trademarks (OSIM), the European Patent Office (EPO) and the
World Intellectual Property Organization (WIPO) databases and using the Google
Patent Search engine. This considerably speeds up the patent examination process.
The final part of the paper presents the results of the IPC suggestion algorithm, both
execution time and qualitative evaluation of it. It also includes the resulting execution
times of the search in the above presented patent databases.
In this article we have made an overall description of the system and we have
focused on the description and results of the IPC suggestion algorithm and on the search
process in the above mentioned patent databases.

Keywords: Patent application, patent search, evaluation and examination system,

IPC suggestion.

1. Introduction
Worldwide patent applications are growing at an average rate of 4.7% per
year, according to the 2007 edition of the World Intellectual Property
Organization (WIPO)'s Patent Report [1]. The patent examination procedure has
74 A. Aszalos, J. Domokos, T. Vajda, S. T. Brassai, L. Dávid

two stages: formal verification which follows all the formal procedural steps
and verifies if applications are patentable and the evaluation stage which checks
the grade of novelty and innovation of the patents [2], [3]. To reduce the patent
examination time and increase the quality of the evaluation, despite that the
number of the patent applications are growing, there are two possibilities: to
increase the number of employments of the State Office for Invention and
Trademarks (OSIM) or to reduce the amount of work required for registration,
formal verification and evaluation by using an online integrated system. The
following paragraphs of this section present similar existing systems.
OSIM [4] is a specialized government body that has exclusive authority in
Romania in the field of protection of industrial property. Taking into
consideration the special economic importance of the industrial property and the
need of a competitive management of information in the field of industrial
property, the OSIM has developed a system of services by which offers to the
large public useful information concerning industrial property, processed by
highly competent specialists such as to facilitate correct economic decisions to
be taken. It pays special attention to the promotion of the industrial property.
From 2006, OSIM offers the possibility to register on-line to the epoline®
system, for the following types of patents:
• patents filed according to the European Patent Convention (CBE/EPC),
through OSIM as the national office;
• patents filed according to the Patent Cooperation Treaty (PCT), through
OSIM as reception office;
In the present it is not possible to register online the Romanian national
patent. On the OSIM web page you can find important information about on-
line registration for the above mentioned patent application such as: important
announcements, details about the services, information about how to register
on-line, software for registration of the patent request at OSIM,
recommendations, assistance for clients who want to register on-line an
invention and some details about this page services.
EPO [5] provides a uniform, coherent application procedure for individual
inventors and companies from 38 European countries. It is the executive body
of the European Patent Organization and is supervised by the Administrative
Council. The main role of the EPO is to grant European patents.
The EPO carries out researches and substantive examinations on a
continuously growing number of European patent applications and international
applications filed according to the Patent Cooperation Treaty. In the case of
European patent applications, the Office gives the option of an accelerated
procedure. The Office examines also oppositions against already granted
European patents.
 Exambrev - Integrated System for Patent Application 75

Publication of the invention is very important to the European patent system.

The public can obtain copies of the patent documents from the European
Publication Server. The European Patent Register provides details of the status
of patent procedures at the EPO. All the EPO's patent documents are available
to the public through the free Esp@cenet service on the Internet. The EPO also
provides a wide range of other products for searching patent databases.
The epoline® is an EPO package of software with on-line services that
allows users to create and apply electronically for patents at the Intellectual
Property Office and other national and international offices, including the EPO
and WIPO. The epoline® is a high security system based on smart cards.
To join the system, one should make two steps: first, to get an EPO smart
card, then to fill in on-line an IPO (Intellectual Property Organization)
enrolment form and submit.
This service has a series of advantages. It is a user-friendly application that
helps inventors to build their applications and forms with a validation option
that helps users to make applications and forms right even when doing this for
the first time. Security of sending the documents is ensured, there are no postal
delivery delays or postage costs. The sender receives an immediate filing receipt
after sending the forms.
WIPO [1] is a specialized agency of the United Nations. It is dedicated to
develop a balanced and accessible international intellectual property system,
which rewards creativity, stimulates innovation and contributes to economic
development in regard to the public interest.
WIPO was established by the WIPO Convention in 1967 for the protection
of intellectual property worldwide by collaboration with other international
organizations and cooperation among states. Its headquarters are in Geneva,
Switzerland. WIPO considers that intellectual property is essential to the
economic, social and cultural growth of all countries. Thus its objective is to
promote the effective use and protection of intellectual property (IP) worldwide.
WIPO provides services for the owners and users of intellectual property,
such as international registration services, thus a single application has to be
filed that is valid in multiple countries. IP classification systems of WIPO are
used for registering IP and making it easy to search in IP databases and
registries. WIPO's Arbitration and Mediation Centre offers resolution services
for private parties involved in international intellectual property disputes [1].
PATENTSCOPE® Search Service is a service that makes possible for users
to search in all international patent applications published, starting from the first
one that was published in 1978 to nowadays, and has a special part for the latest
information and documents available online.
In this article we present an Integrated Patent Examination Expert System
(EXAMBREV) developed as a web application considering the Java EE
76 A. Aszalos, J. Domokos, T. Vajda, S. T. Brassai, L. Dávid

platform, which helps the applicants to accomplish the entire registration

procedure of a patent request and verifies the formal correctness of the patent
application form. Our system’s other functionalities include the management of
civil servants and expert users, the presentation of the application for evaluation
through a web interface. Our translator unit is a helping hand for experts for
searching similar technical solutions in a wide range of different language
patent databases and the Romanian Patent Database. Our system will also help
the management of the OSIM to see in which field to employ new domain
experts in the future.
The outline of the paper is as follows. Section 2 presents our system
architecture and describes the tasks and functionalities of each subsystem.
Section 3 presents the results of the execution time of various algorithms used
by the system, Section 4 contains the conclusions. The final part of the paper
presents the acknowledgements and references.

2. Technical information

A. System overview
The architecture of our system is presented in Fig. 1. Our system has two
main modules divided in multiple subsystems. The first module is called
Interfaces and data preparation module which manages the patent requests,
common users (UCOM), expert users (UEX), civil servants (UFUNC),
applicants (UAPP), administrators (UADM), civil servant managers
(UFUNCM) and expert managers (UEXPM) and also prepares some initial data
for the Expert system module (SIEXP). The second module is the Expert system
module (SIEXP) which gives the world wide novelty of a technical solution
proposed by an inventor and contains the legal and procedural database. In this
paper the Interfaces and data preparation module and especially the search
methods for similar technical solutions in the online patent databases are
presented.
The deployment diagram shown in Fig. 2 illustrates the connections between
the different subsystems of the Interfaces and data preparation module and
their deployment on the used servers. As we can see, all the subsystems
communicate with the system database through JPA (Java Persistence
Application Programming Interface) which communicates with the database
through the JDBC API (Java Database Connectivity API).
 Exambrev - Integrated System for Patent Application 77

Figure 1: EXAMBREV system architecture.

Figure 2: The deployment diagram of the EXAMBREV system.

78 A. Aszalos, J. Domokos, T. Vajda, S. T. Brassai, L. Dávid

B. The SISTORIC subsystem

The SISTORIC (SS-1) subsystem has two main functionalities [6]:
• the management and storage of the information regarding a technical
solution proposed by an inventor;
• the automatic formal verification of the patent application.
According to Romanian State office for Inventions and Trademarks we had
to deal with three types of patents:
• EPC (European Patent Convention);
• PCT (Patent Cooperation Treaty);
• Romanian national patent.
We have implemented for now the management system for the Romanian
national patent. The application is accessible from the Internet through a web
browser and makes possible to submit online patent requests. The data of a
patent application is stored in a relational database management system, a
MySQL database.
The software subsystem provides the following features [7]:
• Applicant user registration;
• Civil servant user registration;
• Expert user registration;
• Account activation for the above users;
• Online patent application;
• Editing patent application information;
• Editing account information;
• Patent application list;
• Semiautomatic IPC code assignment.
In the registration process the user starts the registration and fills in all the
required personal information and specific user information. In the case of the
expert users the specific information is the list of the IPC categories in which he
has knowledge. After the data validation process the system sends an activation
email to the registered person, containing a link to the activation page. The
account of the user will be accessible after clicking the link in the received
email and activating his account. In the case of the civil servant and expert users
after the activation procedure their account must be confirmed. This can be
done by the civil servant manager or the expert manager. These manager type
users are promoted from the civil servant users by the administrator. If a civil
servant is promoted to manager his account gets confirmed automatically. The
user’s personal information is saved in the database.
The login process is different for almost every user type, although there is a
common part too. In the common part of the login process the system checks
 Exambrev - Integrated System for Patent Application 79

the username, password and the activation status of the account. If the username
and password are matching and the account is activated the applicant and expert
users can login. If the expert user’s account isn’t confirmed he has limited
accessibility in the system and can only change the list of the categories he is
expert at. The civil servant users can log in only if their account is confirmed.
The patent application process consist in filling of the online application
form which is the same used at OSIM in present. This process is divided into 4
steps, because this way the amount of required data on one page isn’t too large
and if there are validation errors, the user can correct them more easily.

C. The SICLAS subsystem

The SICLAS (SS-2) subsystem contains the IPC suggestion algorithm which
can be used by the applicants to find the appropriate IPC categories for their
invention [8], [9]. After the patent application is submitted by the applicant he is
asked to select the IPC categories in which the invention should be categorized.
At this step the applicant can describe his invention by keywords in Romanian
or English and ask for a list containing the suggested IPC categories. If the
keywords are given in Romanian, the subsystem’s translator unit automatically
translates them into English. The API used by the translator unit is the Google
Translate API.
Before we describe the logic of the IPC suggestion algorithm, we present the
database diagram extract containing the tables used for the storage of the IPC
categories. Fig. 3 shows the tables representing the 5 IPC levels: sections,
classes, subclasses, main groups and subgroups. These tables contain the IPC
codes and their description of every IPC level. The one-to-many relationships
between the tables illustrate the hierarchical organization of the IPC levels.

Figure 3: Database diagram extract of the IPC category tables.

We can define the goal of the IPC suggestion algorithm as determining a list
of appropriate IPC categories for an invention described by keywords. The
algorithm can be split into 3 sections: initialization, search and result
80 A. Aszalos, J. Domokos, T. Vajda, S. T. Brassai, L. Dávid

propagation. The list of coefficients initialized in the first section and their
initialization values are shown in Table 1.
Table 1: The IPC suggestion algorithm coefficients and their initialization values.
Section Class Subclass Main group Subgroup
(ps) (pc) (psc) (pmg) (psg)
30 25 20 15 10
The search section of the algorithm contains 6 steps. In the following these
will be presented in detail. For a better understanding of the algorithm we
introduce two result sets, A and B, which will contain the temporary and final
search results. In the first step we search on all IPC levels for IPC categories of
which description contains at least one of the given keywords. These categories
are inserted into result set A. This step is shown in Fig. 4.

Figure 4: Keyword search mechanism on all IPC levels and building of the result set A.
In the 2nd step we take the subgroups from result set A, insert them into
result set B and calculate a suggestion value for them. The general form of the
calculation formula is given in Eq. 1:
SgVal = NoOfKwdsInIPC _ Desc ⋅ IPCLevCoef + OldSgVal (1)
In this step the old suggestion value is 0, the IPC level coefficient is 10, and
the number of keywords in the IPC category description is calculated for every
record. Fig. 5 shows the graphical representation of this step.
In the 3rd step we advance upwards in the IPC’s hierarchical organization to
the level of the main groups. We take those main groups from A, which does
not have subgroups in result set A, insert them into result set B, and calculate
their suggestion values with the Ec. 1. This is illustrated on Fig. 6 as the “i”
substep. The second substep “ii” consists of updating the suggestion values of
those subgroups in result set B, which belong to the main groups in the result
set A. The updated value is calculated using the formula given in the Ec. 1.
 Exambrev - Integrated System for Patent Application 81

Figure 5: 2nd step of the search in the IPC suggestion algorithm.

Figure 6: 3rd step of the search in the IPC suggestion algorithm.

The following steps from 4 to 6 are similar to the steps already presented.
We continue to advance upwards in the IPC’s hierarchical levels, insert the IPC
categories without subcategories in result set B along with their suggestion
value, and finally update the suggestion value of those categories in result set B
which had parent categories in result set A. All of the steps and the algorithmic
language of the suggestion algorithm were presented in detail in [9]. One more
step, the 4th, of the algorithm is presented in Fig. 7.

Figure 7: 4th step of the search in the IPC suggestion algorithm.

From the first version of the algorithm to the final version there were made
some changes in the technical realization of it, in order to achieve better
execution times. The necessity of this optimization was presented in [10], but
82 A. Aszalos, J. Domokos, T. Vajda, S. T. Brassai, L. Dávid

there weren’t presented the results of it. Section 3 presents a comparison

between the execution times of the non-optimized and optimized version of the
suggestion algorithm.

D. The SICOST subsystem

The SICOST (SS-3) subsystem is responsible for Expert user (UEXP) data
management [6], [8]. This subsystem also takes the IPC code given to a patent
application by SS-2, and outputs it to the civil servant manager helping him to
choose 3 expert users for this domain.
This subsystem also provides a web interface for the expert users to search
the following online patent databases:
• WIPO database;
• EPO database;
• Romanian Patent Database.
All of the above databases can be searched online from the esp@canet
webpage. In our system we give the option to the expert to search these
databases by keywords or by IPC code.
When the expert searches by keywords we build a URL with the required
request parameters for the esp@canet webpage and execute it. The response
given by esp@canet has a fixed structure containing an HTML table with the
results of the search. This HTML table is parsed by us using a HTML parser
and the results are organized in a list containing the following information: the
title of the invention, the link to the page presenting the invention in detail, the
inventors, the applicants and the IPC code of the invention.
If the expert chooses to search in the patent databases by IPC code the
process is similar to the search by keywords, the difference is in the
construction of the URL. It takes different request parameters with the value of
the IPC code given by the expert.
Our system also provides the possibility for the experts to search with the
Google Patent Search Engine.
The results of these searches are presented to the expert immediately after
the expert submits the search and help him in finding similar technical
solutions.
The detailed description of these search mechanisms was presented in [10],
where wasn’t presented the fact, that these mechanisms are implemented in two
versions: a single- and multi-threaded version. The multi-threaded version of
the search mechanism was implemented as an optimization of the execution
time of the search mechanisms. In section 3 we present the results and
comparison of the execution time of the two versions of the search mechanisms.
 Exambrev - Integrated System for Patent Application 83

E. IFS-SIEXP subsystem
The IFS-SIEXP (SS-4) subsystem is the special interface for the SIEXP
module [6], [8], [11]. It makes data transfer between the Interfaces and data
preparation module and Expert system module. It also communicates with
UEXP, UCOM and UINV via a Web interface. This is the login point to the
Web application for registered users.

3. Results
In the first part of this section there is presented a comparison between the
execution times of the optimized and the non-optimized IPC suggestion
algorithms. This is followed by the qualitative results of the previously
mentioned algorithm. The execution time of the similar invention search
mechanism and the comparison of the two versions (single and multi-threaded)
of the search are presented in the second part of this section.

A. Results of the IPC Suggestion Algorithm

Table 2 shows the execution times of the IPC suggestion algorithm. The
execution times were measured with the SQLyog software, which measures the
execution time of every executed query.

Table 2: Execution times of the IPC Suggestion Algorithm.

Non-
Optimized
Keywords optimized
(sec)
(sec)
METHOD SYSTEM COMPUTER CONTROLLED BICYCLE GEAR
19,312 6,813
SHIFTING
BICYCLE GEAR SHIFTING METHOD APPARATUS 20,984 5,203
MOUSE RODENT TRAP 2,813 2,860
ELECTRIC MOUSE RODENT TRAP 6,968 2,718
HAIR CUTTING DEVICE 26,563 6,562
EDGE DETECTION IMAGE PROCESSING 2,812 2,047
Average: 13,2420 4,3672

The second and third columns of the table contain the non-optimized and
optimized algorithm’s execution times. If we have a look at the difference
between the average execution times of the two versions of the algorithm, it is
evident that there is a 67.02% decrease in the execution time of the algorithm so
the optimized algorithm performs more than 3 times faster.
84 A. Aszalos, J. Domokos, T. Vajda, S. T. Brassai, L. Dávid

Table 3: Qualitative results of the IPC Suggestion Algorithm.

Suggestion
No. Section Class Subclass Main group Subgroup
value
1 G G06 G06T G06T0009000000 G06T0009200000 75
2 A A22 A22C A22C0029000000 A22C0029020000 70
3 A A22 A22C A22C0029000000 A22C0029040000 70
4 G G06 G06T G06T0001000000 70
5 G G06 G06T G06T0003000000 G06T0003200000 65
6 G G06 G06T G06T0003000000 G06T0003400000 65
7 G G06 G06T G06T0003000000 G06T0003600000 65
8 G G06 G06T G06T0005000000 G06T0005500000 65
9 G G06 G06T G06T0007000000 G06T0007600000 65
10 G G06 G06T G06T0011000000 G06T0011800000 65

Table 3. shows the suggested IPC categories and suggestion values for an
existing invention, an algorithm suitable for edge detection in image processing.
The following keywords were given as an input for the suggestion algorithm:
“edge detection image processing algorithm”. The IPC main group in which the
existing invention is categorized is shown in the gray cell of the table. As for
the evaluation of the algorithm’s quality, we can state, that having a look at the
suggestion values, the correct IPC category is located on the 3rd place. If we
have a look at the hierarchical structure of the IPC, we can see that the
algorithm determined correctly the section, class and subclass level of the
invention even in the first result.
B. Results of the Similar Invention Search Mechanism
Table 4 and Table 5 contain the execution times of the single- and multi-
threaded version of the similar invention search mechanisms and the number of
results.
It is important to mention that the table contains only the execution time of
the search mechanism, measured with the functions provided by the Java
language for execution time measurement and does not include the search setup
time. Table 3 shows us that the single-threaded version of the search mechanism
was approximately 2 times slower in average, comparing to the multi-threaded
version.
The difference between the test cases shown in Table 3 and Table 4 is in the
used search providers and search languages. In Table 3 there were used three
search providers provided by Esp@cenet plus the Google Patent Search with
English keywords. In Table 4 the tests were conducted on the Esp@cenet search
providers in English and Romanian languages. Having a look at the average
execution times in Table 4 we can conclude that the multi-threaded version of
 Exambrev - Integrated System for Patent Application 85

the search mechanism is approximately 2 times faster than the single-threaded

version.
Table 4: Execution times of the two versions of the Similar Invention Search
Mechanisms with Esp@cenet and Google Patent Search with English keywords.

Single Multi No. of

Keywords
(sec) (sec) results
6,233 2,926
METHOD SYSTEM COMPUTER CONTROLLED BICYCLE GEAR
6,154 2,443 96
SHIFTING
5,925 2,326
7,127 4,612
BICYCLE GEAR SHIFTING METHOD APPARATUS 7,293 3,088 126
7,785 3,010
14,471 8,346
HAIR CUTTING DEVICE 12,908 6,950 276
13,835 7,396
Average: 9,08 4,57

Table 5: Execution times of the two versions of the Similar Invention Search
Mechanisms with Esp@cenet using English and Romanian keywords.

Single Multi No. of

Keywords
(sec) (sec) results
5,545 3,115
MOUSE RODENT TRAP 5,520 2,365 6
5,549 2,393
6,728 3,971
EDGE DETECTION IMAGE PROCESSING 6,960 3,322 36
6,643 3,191
6,691 2,963
HOUGH TRANSFORMATION 6,356 3,175 30
6,296 3,200
Average: 6,25 3,08

4. Conclusion
We designed and developed a JEE based integrated system for patent
examination. The system will help applicants to make online patent application
registration for all three patent types discussed (EPC, PCT and Romanian
national patent type). The system also helps OSIM patent evaluator experts
management, employers management and patent management.
86 A. Aszalos, J. Domokos, T. Vajda, S. T. Brassai, L. Dávid

The main results obtained are the UCOM, UEXP, UINV, UFUNC and
patent application registration interfaces. The interfaces were developed
considering Java Server Faces technology and PrimeFaces 2.0 technology.
We have developed an algorithm for semiautomatic IPC code assignment for
helping the applicants and also the evaluator experts and a patent database
search mechanism which speeds up the similar technical solutions search.
The focus in this paper was on the presentation of the results of the
optimized IPC suggestion algorithm and the multi threaded similar invention
search mechanisms.

Acknowledgements
This project is developed under Partnership in Anterior Domains Program
of National Authority for Scientific Research in Romania, project code:
11-076/2007.

References
[1] WIPO webpage: http://www.wipo.int/
[2] Implementing regulations to the patent law no. 64/1991, as republished in Official Gazette
of Romania, Part I, No. 456/18 June 2008.
[3] Patent law No. 64/1991, as republished in Official Gazette of Romania, Part I, no. 541/8
August 2007.
[4] OSIM webpage: http://www.osim.ro/
[5] EPO webpage: http://www.epo.org/
[6] Radu, M., “Elaborarea strategiei de cercetare privind examinarea cererilor de brevet de
invenţie şi studiu critic asupra procedurilor de examinare aflate în uz”, Technical report
EXAMBREV, stage I, PNII – Parteneriate, no. 11-076/2007, Centrul de Cercetări pentru
Materiale Macromoleculare şi Membrane, Bucureşti, 2007.
[7] Domokos, J., Vajda, T., Brassai, S. T., Dávid, L., “Realizarea, implementarea în faza de
laborator şi testarea sistemului informatic de examinare a cererilor de brevet de invenţie”,
Technical report for stage III, EXAMBREV, PNII – Parteneriate, no. 11-076/2007,
Sapientia University, Tîrgu Mureş, 2009.
[8] Brassai, S. T., Dávid, L., Domokos, J., Vajda, T., “Technical report for stage II”, PNII –
Parteneriate, no. 11-076/2007, Sapientia University, Tîrgu Mureş, 2008.
[9] Vajda, T., Domokos, J., Brassai, S. T., Dávid, L., Aszalos, A., “Developement of EXAM-
BREV Integrated System for Patent Application”, in Proceedings of the 4th edition of The
INTER-ENG International Conference, Tîrgu Mures, România, 12-13 November, 2009, pp.
309-314.
[10] Aszalos, A., Domokos, J., Vajda, T., Brassai, S. T., Dávid, L., “EXAMBREV Integrated
System for Patent Application”, in Proceedings of the 2nd International Conference on
Mechatronics, Automation, Computer Science and Robotics (MACRo 2010), Tîrgu Mureş,
Romania, 14-15 May 2010, pp. 55-62.
[11] Domokos, J., Vajda, T., Brassai, S., T., Dávid, L., “Integrated System for Patent Appli-
cation Examination (EXAMBREV)”, in Proceedings of 17th International Conference on
Control Systems and Computer Science (CSCS17), Bucureşti, Romania, 26 - 29 May 2009,
pp. 135-139.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 87-98

Inter-Domain Traffic Engineering for Balanced

Network Load
Levente HUSZÁR, Csaba SIMON, Markosz MALIOSZ
Department of Telecommunications and Media Informatics, Budapest University of
Technology and Economics, Magyar tudósok krt. 2., 1117 Budapest, Hungary,
e-mail: {huszar | simon | maliosz}@tmit.bme.hu

Manuscript received October 01, 2010; revised October 30, 2010.

Abstract: Inter-domain cooperation at control plane level offers the possibility to

balance the traffic in a much more flexible way compared to the single domain Traffic
Engineering (TE) methods. This solution is especially effective if the offered traffic and
the network load permits to avoid congestion without recomputing the routes. Current
networking technologies and the emerging trends in aggregation-access-core network
architectures provide means to realize such a cooperative control plane. We consider a
dual opto-electronic network model where traffic grooming is also possible. In the
particular case of the layer 2 switched aggregation network, which feeds a core network
using a GMPLS control plane, this solution means redistributing the traffic between the
spanning trees of the aggregation domain. We propose an inter-domain control plane
cooperation and investigate it by means of simulations on several core network
topologies deploying IP over WDM technology. In the simulations we analyze the
effects of the traffic redistribution rate on the throughput, on the number of lightpaths
and on the number of opto-electronic conversions. We show that if congestion occurs in
the core, we can eliminate the congestion just with a proper coordination between the
control planes of the aggregation and core domains, redistributing the traffic prior
entering the core.
Keywords: Traffic engineering, knowledge plane, GMPLS, MSTP, CSPF.

1. Introduction
Emergent technologies like GMPLS, WDM and carrier-grade Ethernet will
replace legacy ones in future Internet domains. Combining of these different
data plane technologies and different services at different layers into an efficient
interworking environment is a challenging task. The resulting system should
offer a trade-off for service providers to operate their networks.
88 L. Huszár, Cs. Simon, M. Maliosz

A common trend in communication networks is the widespread of fiber

technologies. The optical networks are favored by the operators because they
offer higher network capacities. At the same time they also come with more
advanced control and management features. However, these, combined with the
new services demanded by the market, increase the complexity of the networks
and advanced network optimization mechanisms become a must. The most
common optimization mechanisms deal with the traffic, several Traffic
Engineering (TE) proposals are known in the literature. Typically such TE
solutions deal with a single domain, only [1], [2], [3]. Alternatively a joint
optimization of traffic over a cascade of core networks has been proposed [4],
[5], but they consider the same TE algorithm all over the domains.
Nevertheless, the traffic originated by the end users reaching the core networks
should cross the aggregation and access domains, which use different TE
mechanisms. This paper investigates the effect of cooperation at control plane
level between the different TE mechanisms of the core and access or
aggregation domains. The advantage of this cooperation is that it allows much
more flexibility in maintaining balanced core network usage. Additionally,
other optimization goals can be satisfied, if we can redistribute the traffic
among the edge nodes connecting the access and the core. In our paper we
investigate the impact of network load optimization on the efficiency of a dual
opto-electronic network model [6], [7].
In the followings, we will investigate the networking technologies covered
by the paper and summarize the trends in aggregation-access-core network
architectures. Then we will present our solution on the joint access-core
network optimization and we investigate it by means of simulations. Finally we
conclude our paper.

2. Networking technologies

A. Data and control plane technologies

Most of the applications deployed over the info-communication networks are
based on IP, while the access networks are predominantly Ethernet-based. From
the access network the Ethernet traffic is concentrated at the edge nodes and is
forwarded to packet-based transport in the core domain. The trends evolve
towards the wide deployment of WDM (Wavelength Division Multiplexing)
network devices in the core, which enables the transmission over the established
connection-oriented lightpaths in the optical domain.
These lightpaths form a virtual topology over the physical topology that can
be reconfigured dynamically in response to traffic changes and/or network
 Inter-Domain Traffic Engineering for Balanced Network Load 89

planning. The combination of IP directly with WDM results in an efficient

assignment of optical network resources to forward IP traffic [8], [9].
The versatile Generalized MultiProtocol Label Switching (GMPLS) protocol
enables the integrated control of both IP and WDM. Integrated routing and
wavelength assignment based on GMPLS is currently the most promising
technology, which also delivers effective TE capabilities. The typical routing
protocol in such networks is the Constrained Shortest Path First (CSPF)
[10], [11].

B. Dual opto-electronic model

In the above model the lower layer is an ’optical’ one, the upper layer is
typically an ’electronic’ one, capable of performing joint time and space
switching. Paths of the lower layer correspond to a single link in the upper
layer. Lightpaths are special routes: they arise and terminate in the electronic
layer. The upper, electrical layer can perform multiplexing of different traffic
streams into a single wavelength path (λ-path) or lightpath via simultaneous
time and space switching. Similarly it can demultiplex different traffic streams
of a single lightpath. Furthermore, it can perform re-multiplexing as well: some
of the de-multiplexed traffic will be again multiplexed into some other
wavelength paths and handled together along this other path. This is often
referred to as traffic grooming and we will refer to it as grooming [7].
The question is how these layers can be operated together. Both IETF and
ITU-T propose models and solutions how to operate these two (or more) layers
together [21], [22]. We consider the case when both layers are handled via a
distributed control plane to ensure full and joint on-line adaptivity of both of the
layers. By using dynamic optical layer, it is possible to create an adaptive set of
lightpaths that satisfies emerging traffic demands. Those two physical nodes
that are connected by a lightpath are seen as adjacent by the upper layer.
Multiplexing and demultiplexing the traffic of a lightpath is impossible by
applying only optical devices. In these cases lightpaths have to be torn down,
their traffic has to be taken up to the electronic layer that increases the number
of lightpaths. In addition, the number of applicable opto-electronic converters
per node is limited.

C. Traffic Engineering mechanisms

The traffic in the network domains is engineered in the control plane. The
major goal of such a Traffic Engineering (TE) mechanism is the enhancement
of the performance of an operational network, at both the traffic and resource
levels. Traffic engineering optimizes the use of network resources to achieve
specific goals, such as to avoid congestion, to minimize delay, to differentiate
90 L. Huszár, Cs. Simon, M. Maliosz

services, etc. Most of these methods are valid on intra-domain level, because
internal network information is typically limited outside of administrative
domains [17]. Consider the domain as a closed entity, with a given traffic
matrix. In such solutions the TE affects only the output, but it does not take into
consideration the possibility to influence the input. [2], [3], [18], [19]. Even if it
does, it considers that all domains – the one that provides the traffic, and the one
that conveys the traffic – use the same TE method (e.g. Path Computation
Element based TE - SPF) [4], [5].

3. Network Models

A. Reference network model

Our proposal is based on the presumption that both the aggregation and
access domains are, or in the near future are expected to be, based on switched
layer 2 (L2) technologies [12], which offer lower bit costs [13]. L2 switched
networks deploy Spanning Tree Protocol variants (STP, MSTP, etc.) [14] to
convey the traffic towards the core.
Fig. 1 gives a picture of the reference network model developed within the
CELTIC TIGER2 project [15]. The reference network [16] reflects the view of
major service providers and vendors on the evolution of networking
infrastructure and the way it will assimilate the new technologies. As seen in
Fig. 1, the networks are divided in three segments: the Access, the Metro and
the Core. Depending of the country and/or local geographic specificities as well
as the Internet Service Provider (ISP) choices, part of the sub-segments depicted
in Fig. 1 may be missing, but based on the current practices and medium-term
forecasts, this generic model describes all networks.

Figure 1: TIGER2 generic network reference model.

The Access network is a local area network, and is widely studied in the
literature. It connects the end users to the first Central Office (CO). Typically
they have a tree-based topology, which aggregates the traffic to the COs. Core
 Inter-Domain Traffic Engineering for Balanced Network Load 91

networks are also well studied and in this model we define it as the national or
wider area domain. Typically they have a meshed topology. As seen in Fig. 1,
the metro network, which links the access to the core, is split into three sub-
segments. In legacy infrastructures, these sub-segments together form a
hierarchy. Access areas may be connected to any metro sub-segment by COs,
and each metro is connected to the core through a Point-of-Presence (PoP).
The roles of the sub-segments should be specified in the context of the
deployed technologies. In this paper we assume that carrier-grade Ethernet-
based L2 technologies become dominant not only in the aggregation, but also in
the access [12]. Based on the above assumptions we obtain the reference
network used in this paper, derived from the generic network model [16]. In this
specific model the metro sub-segments use L2 switching in the access and edge,
while the core deploys L2/L3 TE mechanisms. Thus, the first two sub-segments
of the metro represent successive aggregation levels of the user traffic. In the
metro-access, the first aggregation level, the traffic from multiple COs is
aggregated in Concentration Nodes (CN). In the metro-edge, the second level of
aggregation, traffic from different CNs is processed by a L3/L2 metro node, and
the PoPs at L2/L3 boundary are handling several tens of thousands users. As a
summary, we consider that the metro-access and metro-edge networks form an
aggregation domain, while the metro-core segment is a meshed distribution.

B. Core network models

A specific core network model has been proposed [16], starting from current
ring topologies, widely deployed in optical networks. The model is a Double
Rings with Dual Attachments (DRDA) and it can be used in core networks. In
such topologies two rings, (the inner and the outer metropolitan rings) are
interconnected in such a way, that every node in the outer ring is directly
connected with its associated node in the inner ring, via double links (dual
attachment). These provide high connectivity and multiple back-up paths for
restoration purposes while reusing current network fiber deployments.

C. Investigated network topologies

Based on the reference networks presented in the previous section we
designed a network that was used for our simulation based investigations, and
its topology is presented in Fig. 2 (left). This network is divided in two main
parts, an aggregation network using Multiple Spanning Tree Protocol (MSTP)
[14] and a core part with Constraint based Shortest Path First (CSPF) TE [10].
The traffic sources are depicted on the left-most part of the figure, the
aggregation network conveys the packets to the core network. At the boundary
we have only three edges. In real-life networks the number of edges is kept as
92 L. Huszár, Cs. Simon, M. Maliosz

low as possible for reasons of costs. The network has six destination nodes
(sinks) represented by the exit points of the core network on the right side. The
main function of the aggregation domain is to channel end-user traffic towards
the core, thus its nodes are connected to two neighboring devices, at most.
The core domain has a meshed topology, with a 3 hop shortest distance
between the ingress and the egress. The nodes of the core have a degree of
connectivity of 3 or 4. This is a trade-off between cost effectiveness and the
assurance of alternative paths. The aggregation domain uses Ethernet-switched
technology, and the core uses WDM extended with an electronic control layer.
Apart from investigating the efficient network capacity usage and balanced
load of the core domain, we also investigated the possibility to minimize the
operations in the electronic layer and the usage of longer optical paths. These
last two parameters are characteristics of the dual opto-electronic models.

Figure 2: Topologies of aggregation and meshed core (left) and dual ring core (right).
Our proposal supposes that the domains have a control plane that apart of
running TE and other control functions are capable of communicating/
cooperating with the control planes of the neighbouring domains. Such a control
plane model is the Knowledge Plane [23] that can use MSTP in the aggregation
and CSPF in the core domains.
We also investigated the behavior of the core if it deploys a dual ring
topology (see Fig. 2 - right). We have kept the edge nodes and the output nodes
from the previous topology in order to use the same aggregation network and to
be able to compare the two results. In the following we refer to the first
topology as meshed core, while to the latter one as dual ring core.

4. Inter-domain TE cooperation
Our proposal is to use shared intelligence between control planes, where the
core intra-network functions are unchanged and only the inter-network control
planes co-operate which enhances the performance.
 Inter-Domain Traffic Engineering for Balanced Network Load 93

In Fig. 2 the traffic reaches the core network through the aggregation
domain. In case of any event (congestion on a link, link failure, etc.) the
classical TE works with the assumption that the traffic matrix remains
unchanged and it has to re-distribute the traffic volume relying on load
redistribution inside the core. Our proposal is to use the Knowledge Plane and
re-arrange the input traffic distribution outside the core edge routers. This
means that – from the point of view of the core – we change the traffic matrix,
since the load on the edges will be different.
Let us take the topology presented in the left-side of Fig. 2. Now in the
situation when the aggregation domain directs all the traffic to the e1_edge (the
“northern” one), while e2_edge (the “middle” edge) and e3_edge (the
“southern” edge) do not feed any traffic to the core. This is the worst case
situation to overload the core and corresponds to the situation when only the
tree rooted in e1_edge is used to collect the traffic in the aggregation domain.
Now, if we take the opposite situation, when we use each of the trees in the
aggregation domain to forward the same amount of traffic, then the aggregation
domain distributes the traffic evenly among the three ingresses. In this case all
regions of the core will be evenly loaded.
It is the task of the Knowledge Plane to map the traffic sources among the
trees. In our simulations we used small individual flow throughputs. Each tree is
collecting such individual demands and the sum of these represents the traffic
load at the edges. Practically the granularity of the traffic is small enough to
allow us to finely balance the load. In what follows we will use the term load
balancing as the operation of load redistribution in the aggregation domain as
described above. The goal of load balancing will be to decongest a certain area
of the core network with a minimal redistribution of the original load.

5. Simulation results

A. Traffic model
During the simulations the traffic flows originated from the sources have the
same bandwidth. We considered that we know the traffic matrix and the paths in
the core are computed by a PCE using CSPF protocol. Additionally we
generated background traffic, as well, which enter the core at the edge nodes
and sink on the most right-hand side destination nodes. The links of the core
networks had 200 Mbps capacity, which defines the load region where the core
network is congested, but not overloaded of 400 Mbps to 800 Mbps.
In our investigations we used the e2_edge node where we directed all the
traffic and tried to serve it using CSPF. The resulting paths were called the main
branch. If the demand is high enough, the traffic demand cannot be served. If
94 L. Huszár, Cs. Simon, M. Maliosz

we apply our solution to this situation that means that some part of the traffic
will be shifted to the other two edges, e1_edge and e3_edge. The paths that
follow the flows entering on these two edges are called secondary branches.
We used the background traffic to “fill” the network up to the point where
congestion might start to develop. We sent 200 Mbps background traffic on the
main branch. Then we started to add new traffic demands until we reached the
total one, which was set differently from case to case: all our simulations were
run with the 500Mbps, 600Mbps, 700Mbps and 800Mbps total traffic demand.
These are the situations when we can test the usefulness of our proposal and
evaluate its impact on the efficiency of the opto-electronic core transport.
We used a flow level simulator, already used for the research of opto-
electronic networks [24]. We generated individual flows, and the sum of these
demands resulted in the overall traffic demand. Each link was divided into
lightpaths of 10 Mbps capacity. This results in 20 lightpaths within each link
that offers enough flexibility for multiplexing the flows within the core. Based
on earlier work with the simulator we opted for 12 individual flows per
lightpath, resulting in a flow capacity of 0.83 Mbps.
Within each scenario – that is for different overall traffic demands – we have
simulated several sub-cases, where the load of the main branch was gradually
re-distributed among the secondary branches. At first we started with the
situation where 30% of the traffic was entering at edges e1_edge and e3_edge
(15% on each of them). From there on, we stepwise directed more and more
traffic towards to the secondary branches while the network was able to carry
the traffic without loss. In order to be sure on that, we also simulated the next
step following this point. The individual flow demands were scheduled
randomly. For each situation we run ten simulations and averaged the results.

B. Spanning Trees in the aggregation domain

In order to get the input traffic at the ingresses of the core domain, we had to
build the trees that bring the traffic to the edges of the core. For this, we had to
build the set of trees that form the basis of the MSTP operation. We used a
combination of the TOTEM [25] and BridgeSim [26] tools to simulate these
trees.
With the combination of these two tools we could use the topologies created
in TOTEM and apply the STP formation protocol implemented in BridgeSim.
We generated all the spanning trees that can potentially be used in our
scenarios, that is, all the trees that are rooted in one of the three edges and the
traffic sources are their leaves. Fig. 3 presents the tree generated for the
northern edge, the other two trees have a similar shape. The actual traffic
distribution among the three trees is decided according to our solution and
should be enforced by the Knowledge Plane, as mentioned earlier.
 Inter-Domain Traffic Engineering for Balanced Network Load 95

e1 e1_b e1_edge

e2 e2_b e2_edge

e3 e3_b e3_edge

Figure 3: The STP rooted in node e1_edge.

We did not investigate the behavior (delay, blocking, packet loss, etc.) of the
aggregation domain, only used it to generate the MSTPs and determine the
input traffic for the core domain. In what follows we will be interested only
about the traffic distribution among the three ingress nodes.

C. Balancing the load in the core

100
Rate of served demands [%]

90

80 500 Mbps
600 Mbps
70 700 Mbps
800 Mbps
60
30 40 50 60 70 80
Redistributed traffic [%]

Figure 4: The successfully served flow demands as the function of traffic re-distribution
for the meshed core (left) and the dual ring core (right).

The left-hand side of Fig. 4 presents the ratio of successfully served traffic
demands in the meshed core. It can be seen that 500 Mbps total traffic will be
served if we redirect 35% of the traffic on the secondary branches. The traffic
volumes that must be redirected to achieve a loss-free ratio for the 600Mbps,
700Mbps and 800Mbps traffic scenarios are 50%, 60% and 70%, respectively.
These results confirm that if we redistribute the traffic before it hits the core
edges, we can balance the core load, thus it is a viable mechanism to actively
increase the efficiency of the core traffic engineering process.
We achieved similar results for the dual ring topology, as well (right-hand
side of Fig. 4). The congestion-free core is achieved for the redistribution of the
35% of traffic for the 500Mbps case and 70% for the 800Mbps (worst) case.

D. Dual opto-electronic model

In the following, we present our simulation results on the effect of our
proposal on the efficiency of the opto-electrical dual transport core network.
96 L. Huszár, Cs. Simon, M. Maliosz

First we explain the results using the meshed core. The first parameter is the
number of lightpaths. We can see in Fig. 5 (on the left) that as the rate of
successfully served traffic demands is rising, but is still below 100%, the
number of lightpaths is increasing. This is due to the fact that more and more
individual flows are in the network and these are following new (alternative)
routes. Thus, the increase of this parameter is not a consequence of the
decreasing efficiency but of the growth of the core utilization.
This trend is reversing if we keep redistributing the traffic even after all the
traffic reaches its destination. This corresponds to the situation depicted in
Fig. 4 by the dots on the 100% line. As we already mentioned, for each overall
traffic load scenario we simulated two cases when all the demands were served
by the core: the “break-even” point and one following step where we increased
the traffic redistribution by 10%. The results obtained for these cases are
encircled in Fig. 5 and as we can see the number of paths is decreasing. The
more the core is loaded, the more these trends are accentuated, therefore it can
be seen the best result on the curve corresponding to the highest loads.
350 700
Opto-electronic conversions [#]

300 600

250
Lightpaths [#]

500

200
500 Mbps 400 Meshed core
600 Mbps
150
700 Mbps
Dual Ring Core
300
800 Mbps
100
30 40 50 60 70 80
30 40 50 60 70 80
Redistributed traffic [%]
Redistributed traffic [%]

Figure 5: The efficiency of the lightpath management. Number of lightpaths for meshed
core (left) and number of opto-electronic conversions (right).

If the primary goal is the minimization of the operations in the electrical

layer, the best option is to distribute the traffic. The drawback of this solution is
that we have to use the alternative branches. In regular operation the alternative
paths are associated with higher costs. Therefore it is the decision of the
operator depending on its business model to find the balance between the path
costs (that usually are translated into financial costs) and the efficiency of the
opto-electronic layer (higher-layer delays are translated into worse QoS).
The right-hand side of Fig. 5 presents the number of opto-electronic
conversions done in the core (values that yielded 100% success rate are
highlighted with a circle). We have plotted in the same graph the results for
both the meshed core and dual ring core. These conversions can happen only in
 Inter-Domain Traffic Engineering for Balanced Network Load 97

the nodes that make a grooming operation and multiple conversions may
happen in such a node. Based on our simulations there are several hundreds of
such conversions per node. The trend observed for the number of the lightpaths
is valid also here, for both core topologies.

6. Conclusion
This paper proposed a traffic management solution that improves the
performance of the core network. The aggregation network is supposed to
deploy L2 switched technologies, while the core network will use WDM in
combination with GMPLS. We have prepared a meshed and a dual ring
topology following the principles of the TIGER2 project’s reference network
and investigated our proposal by means of simulations.
We have shown that if congestion occurs in the core, we can eliminate the
congestion just with a proper coordination between the control planes of the
aggregation and core domains, redistributing the traffic prior entering the core.
We deployed MSTP protocol in the aggregation domain and the traffic
redistribution was done using these spanning trees. This solution increases the
ratio of successfully served traffic demands, increasing the utilization of the
core. The traffic redistribution at the aggregation has positive effects even if
there is no congestion in the network, because in such cases it increases the
efficiency of the opto-electronic transport layer.
As a conclusion we can say that the cooperation of the control layer of the
aggregation and core domains has multiple advantages. In the future we plan to
investigate the trade-off between the cost of load balancing and opto-electronic
efficiency.

Acknowledgements
This work has been partially funded in the framework of the CELTIC
TIGER2 project (CP5-024) as part of the EUREKA cluster program.

References
[1] Osbourne, E., Simha, A., “Traffic Engineering with MPLS”, Cisco Press, Indianapolis,
ISBN 978-1-58705-031-2, 2003.
[2] Fortz, B., Rexford, J., Thorup, M., “Traffic engineering with traditional IP routing
protocols”, IEEE Comm. Magazine, vol. 40, no. 10, pp. 118-124, 2002.
[3] Dasgupta, S., de Oliveira, J. C., Vasseur, J.-P., “Dynamic traffic engineering for mixed
traffic on international networks: Simulation and analysis on real network and traffic
scenarios”, Computer Networks, vol. 52, no. 11, pp. 2237-2258, 2008.
98 L. Huszár, Cs. Simon, M. Maliosz

[4] Casellas, R., Martinez, R., Munoz, R., Gunreben, S.,“Enhanced Backwards Recursive Path
Computation for Multi-area Wavelength Switched Optical Networks under Wavelength
Continuity Constraint”, Journal of Optical Communications and Networking (JOCN), vol.
1, no. 2, pp. A180-A193, 2009.
[5] Ho, K-H. et al, “Inter-autonomous system provisioning for end-to-end bandwidth
guarantees”, Comp. Commun., vol.30, no. 18, pp. 3757-3777, Dec. 2007.
[6] Sabella, R., Zhang, H., eds.: “Traffic Engineering in Optical Networks”, IEEE Network,
vol.17, no. 2, pp. 6-7, 2003.
[7] Cinkler, T., “Traffic- and λ-Grooming”, IEEE Network, vol. 17, no. 2., pp. 16-21, 2003.
[8] Liu, K. H., “IP Over WDM”, John Wiley & Sons Inc., ISBN: 978-0-470-84417-5, 2002.
[9] Mukherjee, B., “Optical WDM Networks”, Optical Networks Series, Springer, ISBN: 978-
0-387-29055-3, 2006.
[10] Ziegelmann, M., “Constrained Shortest Path and Related Problems: Constrained Network
Optimization”, VDM Verlag Dr. Müller, ISBN 978-3-8364-4633-4, 2007.
[11] Lee, Y., Mukherjee, B., “Traffic engineering in next-generation optical networks”, IEEE
Comm. Surveys and Tutorials, vol. 6, no. 1-4, pp. 16-33, 2004.
[12] Fang., L., Bitar, N., Zhang, R., Taylor, M., “The Evolution of Carrier Ethernet Services:
Requirements and Deployment Case Studies”, IEEE Comm. Mag., vol. 46, no. 3, pp. 69–
76, 2008.
[13] Occam Networks whitepaper, “Switching Versus Routing in Access Networks”,
http://www.occamnetworks.com/pdf/SWITCH_VS_ROUT_WP_FINAL.pdf, May, 2007.
[14] Caro, L. F., Papadimitriou, D., Marzo, J. L., “A performance analysis of carrier Ethernet
schemes based on Multiple Spanning Trees”, VIII Workshop in G/MPLS networks, Girona,
Spain, Jun. 2009.
[15] CELTIC TIGER2 project homepage, http://projects.celtic-initiative.org/tiger2/
[16] Dorgeulle, F., “Rationales and scenarios for investigations on next generation of access,
backhauling and aggregation networks”, CELTIC TIGER2 public report D20, Nov. 2009.
[17] Awduche, D. et al., “Overview and Principles of Internet Traffic Engineering”, RFC3272,
May 2002.
[18] Feamster, N., Borkenhagen, J., Rexford, J., “Guidelines for interdomain traffic enginee-
ring”, SIGCOMM Comput. Comm., Rev. 33, pp. 19-30, 2003.
[19] Vigoureux, M., et al., “Multi-layer traffic engineering for GMPLS-enabled networks”,
IEEE Comm. Mag., 0163-6804 vol. 43 (7), pp. 44–50, 2005.
[20] Mukherjee, B., “Optical Communication Networks”, McGraw-Hill, ISBN 978-0-070-
44435-5, 1997.
[21] Chiu, A., et al., “Unique Features and Requirements for The Optical Layer Control Plane”,
IETF internet draft, work in progress.
[22] International Telecommunication Union, “OTN – ITU-T Recomm. on ASTN/ASON
Control Plane”, http://www.itu.int/ITU-T/2001-2004/com15/otn/astn-control.html.
[23] Clark, D., Partridge, C., Ramming, J. Ch., Wroclawski, J. T., “A knowledge plane for the
internet”, ACM SIGCOMM 2003, Karlsruhe, Germany, pp. 3-10, Aug. 2003.
[24] Hegyi, P., Cinkler, T., Sengezer, N., Karasan, E., “Traffic Engineering in Case of Inter-
connected and Integrated Layers”, IEEE Networks, Budapest, Hungary, pp. 1-8, Sept. 2008.
[25] Homepage of TOTEM simulator, http://totem.run.montefiore.ulg.ac.be/features.html.
[26] Homepage of BridgeSim simulator, http://www.cs.cmu.edu/~acm/bridgesim/index.html.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 99-113

Energy-Efficient Networking: An Overview

László SZILÁGYI, Tibor CINKLER, Zoltán CSERNÁTONY

Budapest University of Technology and Economics,

Dept. of Telecommunications and Media Informatics,
e-mail: szilagyi@tmit.bme.hu; cinkler@tmit.bme.hu; csernatony@tmit.bme.hu

Manuscript received June 30, 2010; revised October 30, 2010.

Abstract: Recently – for economical and ecological reasons –, energy-efficiency

has been receiving an emerging attention from both industrial and fundamental
researchers. Given the large-scale growth and resource overprovision of communication
networks, the issues related to energy consumption get more significant than ever. In
this paper, we provide an overview of some of the latest contributions and most
important trends on energy-efficient networking. Since today’s networks are
heterogeneous to a large extent, different concepts and aspects of networking are
discussed in separate sections.
Following the introduction of the fundamental discipline of energy-proportional
computing (and also its relationship with resource over-provisioning), we give an
overview on the power-saving opportunities of core networks with focusing on two of
the most-widely applied techniques for energy management. A comparison of circuit
and packet switched networking approaches is also made, along with some
considerations taken for the cases in which electrical and optical switching are
employed. In case of access networks, in addition to considering the commonly used
landline connections, we give a brief description about the popular handoff mechanisms
for wireless networks. Finally, we consider energy-saving opportunities on data centers
and take a short overlook on cloud computing. Despite all the evident differences, being
between distinct segments of networking, some of the methods applied at distinct
networking technologies show considerable similarities with each other due to the fact
that some of the occurring problems exhibit similar patterns in terms of modeling and
problem formulation.

Keywords: energy-efficiency, energy consumption, green networking.

100 L. Szilágyi, T. Cinkler, Z. Csernátony

1. Introduction
With the explosive growth of communication networks, energy consumption
has risen to a major economical (operational expenditures) and ecological (CO2
emission) concern in the past few years. About 2 percent of the total CO2
emission is produced by the Information Technology and Communication
sector (ICT) which is more than the contribution of the whole aviation industry
[1]. A recent study puts more emphasis on this issue by showing that the rise of
energy consumption of large communication systems corresponds to Moore’s
law [2]. Therefore, power consumption has become a critical factor of
communication networks, IT facilities, data centers and high performance
network elements. Energy-efficient designing helps cutting the Operating
Expenses (OPEX) as well [3], [4], and in addition to that, it also might result in
more reliable network elements (due to the decrease of heat dissipation).
In order to save energy in communication networks, first, we have to reveal
the reasons of energy wastage in existing systems. Energy inefficiency might
come from architectural (SW related) and physical design (HW related). From
the energy-efficiency point of view, the most important feature of networking is
underutilization. While networks are generally designed to handle peak-time
traffic, most of the time, their capacity remains (heavily) unexploited. This is
called over-provisioning. According to [1], the magnitude of network utilization
is 33% for switched voice, 15% for internet backbones, 3~5% for private line
networks and 1% for LANs, while the energy consumption of network
equipments remains substantial even when the network is idle. A rather physical
related issue is that the energy consumption of network elements is not
proportional to their utilization, i.e. energy cost is a function of capacity, not
throughput. These facts result in high energy wastage.
Today’s networks are mostly configured statically, running at full
performance all the time which is not necessary. In order to achieve higher
energy-efficiency, network management methods need to be able to
dynamically adopt network characteristics to the actual traffic demands during
operation. Switching off underutilized (or idle) parts of the network and
dynamically adapting transmission rates (with satisfying certain QoS
constraints) are ultimately important approaches of designing a greener
network. In order to make energy-aware management possible, network
elements also should support these features. On-demand frequency-scaling of
CPUs and data storage modules and network interfaces with adjustable
transmission rates (rate-adaptation support) are all mandatory for attaining
greener network elements.
Finding efficient ways for cooling network equipments is also a big
challenge. In case of data centers, roughly 50% of electricity is consumed by the
 Energy-Efficient Networking: An Overview 101

cooling infrastructure, the other 50% used for computing [5], [6]. Increasing
cooling efficiency and using alternative cooling methods have a huge
contribution to the electric bill of equipment rooms. Employing alternative
energy sources (e.g. solar, wind) for supplying network nodes (base stations) is
also a matter of interest nowadays.
In this paper, we provide an overview of the latest results concerning energy-
efficient networking, discussing the different functional parts of the ICT
infrastructure separately. In Fig. 1, the estimated share in energy consumption
by different areas of ITC can be seen [7]. Energy-efficiency is examined from
an operator’s point of view with focusing on networking infrastructure and data
centers, but leaving PCs, monitors, printers, and other user equipments out of
consideration.

Figure 1: The estimated energy consumption shares of different functional

parts of ITC [7].
102 L. Szilágyi, T. Cinkler, Z. Csernátony

The paper is organized as follows. Section 2 explains the importance of

power consumption dynamic range. In the remaining sections, we provide a
closer look at the issues with different areas of communication networks:
Section 3 discusses some of the most important features of core networks, while
Section 4 and 5 do the same for access networks and data centers, respectively.
Finally, in Section 6, we conclude our experiences and reveal a few important
directions for future solutions.

2. The dynamic range of power consumption: Rate adaptation versus

switching off components
Rate-adaptation is highly important in energy-aware data transmission and
data processing for the following reason. For transmission power-rate function
is a relationship that gives the required amount of transmission power for a
certain rate. Keeping the bit-error probability fixed, the required power is a
convex function of the rate for most encoding schemes. This results from
Jensen’s inequality which states that transmitting data at lower rates and over
longer time intervals has less energy cost compared to faster rate transmissions
[8]. For data processing variable adaptive rate control means adjusting CPU
frequency for the required performance in order to save power.
The dynamic range of power consumption is one of the most important
attributes of network elements: it is the relative difference of energy
consumption of an element between 0% and 100% of relative utilization. For
example, study [6] shows that the utilization–power usage characteristics of a
system (consisting of servers) shown in Fig. 2 has a smaller (50%) dynamic
range compared to the one on Fig. 3 (90%). As suggested by the results, given
the typical utilization region (10-50%), a 50% save can be achieved in terms of
energy consumption when having a dynamic range of 90%. Consequently,
network component vendors should increase the dynamic range of their
equipments as much as possible. The dynamic range of 100% addresses the case
of energy-proportional computing.
However – even when applying rate-adaptation –, network components with
low dynamic range still tend to consume a considerable amount of energy when
being not utilized at all (if the offset of power usage is too high). Even so, in
many cases (apart from rate-adaptation), power consumption can be further
reduced by switching off certain underutilized (or idle) components. The price
to pay here is the potential occurrence of additional delay, however – due to the
typically high power on consumption – transition between active and passive
operation mode cannot be done arbitrarily often.
 Energy-Efficient Networking: An Overview 103

A real challenge for these solutions is finding a method to decide whether a

network element should operate or not, and if so, then determining the
operational rate in specific moments.

Figure 2: Relative power usage and energy-efficiency for a typical data center
scenario.

Figure 3: Relative power usage and energy-efficiency for a data center scenario
with larger dynamic range [6].
104 L. Szilágyi, T. Cinkler, Z. Csernátony

3. Core networks, packet versus circuit-switching

The backbone infrastructure of a communication network transferring
aggregated data flows is called a core network. Given that the typical rate of
utilization shows a considerable variation within a single day, and that (as stated
previously) networks are provisioned to handle peak traffic, networks tend to be
underutilized for most of the time.
An opportunity for energy conservation method is to shut down idle network
elements. Since the transition between active and sleeping states requires
additional energy and time, inactive periods along with the routing has to be
scheduled properly in order to preserve power without exceeding a given value
of delay. In [9], a frame-based periodic scheduling is introduced. The network
elements work either at a full or at a zero rate. The lengths of active periods are
proportional to the traffic loads of network elements while periods cannot be
arbitrarily small since transitions consume extra power, however too long
periods result in increased delays. In [2], the network is controlled on the basis
of traffic statistics: the day is sliced into one-hour long “network configuration
periods”, and the state of the system is formalized as a “linear programming
problem” with the daily operation scheme being determined to satisfy the
specific QoS constraints.
An important feature of core networks is that multiple paths might exist
between distinct nodes of them. This fact allows us to distinguish between
nodes to determine the ones that can be switched off during low backbone
utilization periods. Many high capacity links within a core network actually are
compositions of smaller capacity ones. In [10], a method is demonstrated for
shutting off cables in bundled links for lower traffic periods. The problem of
determining the sublinks to shut down is formulated as an Integer Linear
Program (ILP) that is proven to be NP-complete. Given this fact, for tackling
this intractability, efficient heuristics have to be proposed. In [10], a fast greedy
heuristic solution suggested for a dedicated multicore server is proven to be
feasible. Optimization cycles becoming 8 seconds to 50 minutes long
(depending on the actual topology) mean a dynamic power-saving solution.
Underutilization means that a link or a network element does not need to
operate at its full capacity to satisfy the corresponding QoS constraints. Upon
this fact, rate adaptive routers, switches and links can be employed in order to
save energy [2], [11], [12]. A suitable hybrid application of element switching-
off and rate-adaptation is a promising solution [11].When multiple transmission
lines form a single large-capacity link, rate-adaptation can be the combination
of operations: switching off transmission lines, and tuning the transmission rate
of individual lines (width control) [13].
In [14], core networks are examined for their power-efficiency from a
different perspective. Today's backbone network infrastructures are composed
 Energy-Efficient Networking: An Overview 105

by optical links between large distance nodes with the processing and switching
of traffic being performed mostly in the electrical domain. While the energy
consumption of electrical components is getting lower by every year, large
reduction could be achieved in terms of power consumption if the whole
processing and switching is executed in the optical domain. Although, optical
switching has become a reality (MEMS, CMOS), IP, nowadays’ dominant
packet switched transmission technology, requires random access memory for
buffering which is yet to be implemented purely optically. Large fiber delay
lines are not practical as they require power-consuming signal regenerators, not
to mention the impractical size of them.
From an energy-efficiency point of view, it is important to note that –
despite the connectionless nature of IP – above 90% of the traffic within
backbone networks is transported via the connection oriented Transmission
Control Protocol (TCP) [14] (consider applications such as IP television, voice
over IP, video conferencing or online gaming services, all those by which a very
high quality of service is required).
As being shown in Fig. 4, within a node of an IP network, more than half of
the energy is consumed by the traffic processing and forwarding engine
(TP/FE).

Figure 4: An estimation of power consumption by subsystems

(the efficiency of power conversion is not considered) [14].
106 L. Szilágyi, T. Cinkler, Z. Csernátony

On the contrary, the implementation of all-optical, circuit-switched nodes

does not require complex traffic processing and large memory units (within the
electrical domain).
In contrast, all-optical circuit-switched nodes whose implementation doesn't
require complex traffic processing, large memories in the electrical domain and
optical to electrical to optical conversions consume less than 10% of the power
consumed by ordinary optical packet-switched architectures using electronic
traffic processing. Similarly, electronic circuit-switched designs consume 43%
less power than packet-switched one. Taking all these into account, future
energy-efficient core networks nodes might use circuit switching in the core,
along with packet switching edge nodes in order to reduce complexity, thus
power consumption.

4. Access networks
Since most of the physical elements are located in the segment of access
networks, the energy saving by each type of elements is multiplied by a large
factor. This makes an important contribution to the reduction of total
consumption [15].
Nowadays, the most widely deployed technology for broadband landline
connections still employs copper lines for bearers. With the continuous increase
of bandwidth requirements, broadband penetration and bandwidth demands,
more energy is required than ever. Although new transmission technologies,
such as VDSL2, allow higher speeds, they induce increased complexity and
power consumption [16]. By today’s networking technologies, serving high-
bandwidth demands together with sustainable energy consumption can only be
achieved through progressive optical fiber deployment in FTTCab, FTTB (and
also in the longer-term FTTH) architectures which are expected to shorten the
copper access network and to boost the overall performance of xDSL systems.
The deployment of such systems however requires certain capital expenses to
be involved which makes this technological shift a gradual one. Dynamic
Spectrum Management [17] and energy-aware solutions of such kind can make
copper technologies more sustainable, but they only yield the industry little
additional time for making the required technological shift towards optical
networks.
Mobile operators with radio access networks (2G, 3G etc.) have to provide
services for very large physical areas and for several subscribers as well [2].
Given the necessity for several base stations, operating them requires a large
amount of energy. As mobile broadband is expanding rapidly, the energy-
efficiency of radio access networks is expected to receive even more significant
attendance in the future.
 Energy-Efficient Networking: An Overview 107

As presented in Fig. 5, base stations take most of the energy requirements of

cellular networks [18]. For that reason, in the remaining parts of this section, we
focus on the issues of radio base stations.

Figure 5: Power consumption in cellular networks [18].

Since radio access networks are also dimensioned for peak traffic loads, a
large amount of energy is being wasted due to underutilization. The reduction of
traffic level in some portions of a cellular network comes from the combination
of two effects: first, the characteristics of the typical day-night behavior of
users, and second, the daily swarming of users carrying their mobile terminals
around different kinds of areas (residential, office districts, etc.). On the one
hand, this induces the need for a large capacity in all areas at peak usage times,
on the other hand, it reduces resource requirements during lightly occupied time
intervals (day for residential and night for office districts) [15].
Energy can be saved by switching off certain underutilized cells. When some
cells are switched off, we assume that radio coverage and service provisioning
can be taken care of by the cells which remain active, possibly with a smaller
increase in the emitted power, in order to guarantee the availability of service
over the whole area. This energy-saving approach assumes that the original
network dimensioning is essentially driven by traffic demands, as it normally
happens in metropolitan areas, comprising a large number of small cells [15]. In
regular cellular topologies, around 25-30% of the energy is possible to save by
dynamically powering off active cells during under-utilized (low-traffic)
periods [15], [19]. Fig. 6 illustrates the method itself.
108 L. Szilágyi, T. Cinkler, Z. Csernátony

Figure 6: Dynamically powering off cells in cellular networks.

Rate-adaptation is an applicable procedure for wireless links too. Modern
wireless devices are equipped with rate-adaptive capabilities which allow the
transmitter to adjust its transmission rate over time. This can be achieved in
various ways that include adjusting the power level, symbol rate, coding
scheme, constellation size and various combinations of these approaches. In
order to satisfy QoS requirements, existing systems can change transmission
power/coding scheme dynamically depending on the noise floor. If large
bandwidth is not required, “low power” transmission over longer time periods
might also fit for the actual application [8].
In many areas, multiple access technologies are available. Matching
bandwidth requirements with available access technologies, while taking
energy-efficiency into consideration can save energy. There are two types of
handoffs considered: vertical and horizontal handoffs. Vertical handoff is an
interworking technique between different networking technologies whose aim
originally is to assure the best connectivity to applications for mobile terminals
by providing a transparent and seamless roaming between systems of different
technologies. Horizontal handoff means the switching within the same
technology (or layer). Initially, handoff decisions were aimed to be made on the
basis of a lot of different factors such as QoS, cost-of-service, etc. [20]. Lately,
there has been revealed that by suitably involving energy consumption into the
objective (function) of the handoff, energy-efficiency aspects can also be taken
into consideration [21], [22], [23], [24]. In Fig. 7, an illustration of horizontal
and vertical handoff of two base station cells and a WLAN cell can be seen.
 Energy-Efficient Networking: An Overview 109

Figure 7: An illustration of vertical and “horizontal” handoff mechanisms between

cellular networks and WLAN.
Base stations, exploiting alternative energy resources, already exist [25].
Employing alternative energy resources such as sun and wind for supplying
base stations make mobile communication “greener”, and in the same time
these more sustainable sources ease the deployment of BSs far from the electric
grid.

5. Data centers and cloud computing

The majority of networked services are being hosted in data centers. With
the rapid growing of the demand for such data-intensive services (e.g. cloud
computing, video broadcasting and online social networking), the employment
of large data centers is of emerging significance. By now, they have evolved to
networks consisting of tens of thousands of high data-rate servers. Such systems
require special design and management approaches to provide the demanded
robustness, performance and reliability.
As indicated in Fig. 8, according to [6], servers in a typical large system are
being under-utilized in most of the time. As presented in Fig. 5, the power
consumption is still significant in case of data centers. The problem is with the
power consumption dynamic range, as stated above. CPUs are easily tunable,
but they are not the major consumers. HDDs and even memory modules are
much harder to manage from the aspect of power consumption. Switching off
110 L. Szilágyi, T. Cinkler, Z. Csernátony

underutilized servers is not feasible in a large scale for servers, as applications

and data are usually distributed over numerous machines. Switching off
components has too large a penalty, for HDD-s latency becomes in order of
1000 times bigger, and the spin up consumption consumes energy which can be
saved in minutes of being off. In order to increase the efficiency of servers, their
utilization should be maximized, while machines should be created with as wide
dynamic power range as possible (energy proportional computing). Provided
the dynamic range indicated in Fig. 2 could be achieved, 50% of the energy
could be saved compared to servers with characteristics shown in Fig. 1.

Figure 8: CPU utilization in a typical large-scale system [20].

The architecture of a data center influences the power consumption.
Different architectures for a given number of servers mean different links and
switching fabrics, different ways of connecting them to each other. The applied
links and switching fabrics determine both the throughput and energy
consumption. However, more sophisticated structures might provide bigger
throughput, smaller delays, higher access speeds, but also require more power.
Architects have to make a compromise between throughput and power
consumption [26]. Today's popular symmetric architectures (BCube, DCell, Fat-
tree, Balanced tree) are quite rigid and make it hard to apply them for the given
constraints (throughput, power). Many systems are unnecessarily over-
provisioned due to this factor wasting much power. Future architectures should
employ more asymmetricity to provide more fine-grained datacenter designs.
It is important to note that the concept of cloud computing itself has got a
large potential to make the whole IT industry greener [27]. Concepts like
Software as a Service (SaaS) or Infrastructure as a Service (IaaS) helps
 Energy-Efficient Networking: An Overview 111

companies and governments not only to reduce cost with optimized resource
usage, but to keep their business greener. Cloud computing service providers
have large scale data centers and server farms and provide computing services
over the Internet making companies having neither to invest in their own server
parks nor to worry about over-provisioning their systems. The capital
expenditure of a startup company can be restricted to investing in thin clients to
access the demanded services. Although such infrastructures might be more
energy-efficient compared to the legacy approach (i.e. ordinary companies
providing a single PC for every employer while operating a more or less over-
provisioned server park), large scale data centers still tend to be more energy-
efficient.

6. Summary
In this paper, we have revealed a number of possibilities for saving power in
different parts of the network. We have started the discussion with presenting
the general concept that the over-provisioning of the resources results in
suboptimal energy-efficiency by making the energy consumption of the network
disproportional to its utilization. In the following, we have given a review on
the various power-saving opportunities of underutilized network elements for
core networks via sleeping and rate adaptation. Moreover, it has been shown
that by deploying circuit-switched all-optical networks, it can be capitalized on
that the vast majority of Internet traffic is transported via the connection-
oriented TCP. Afterwards, access networking has been discussed: among
landline connection technologies, optical-based FTTx solutions seem to be
promising, while for wireless networks, horizontal and vertical handoff
mechanisms have been presented for adjusting range of radio communication.
Finally, energy-efficient ways for operating data centers and the concept of
cloud computing have been briefly overviewed.
Based on the state-of-the-art, we suspect that the biggest challenge will be
the effective control and cooperation of network components of varying energy-
awareness. Today’s large-scale communication networks are of great
heterogeneity in terms of the employed networking technologies. Consequently,
network management software has to handle different networking equipment
types and generations at the same time. For both complexity and heterogeneity
reasons, there is a fundamental need for a shift to be made from centralized
solutions towards a distributed approach. Designing future’s energy-efficient
network components and architecture needs strong inter-technological
cooperation to match HW capabilities with management techniques.
112 L. Szilágyi, T. Cinkler, Z. Csernátony

References
[1] Nordman, B., “Energy Use and Savings in Communications”, in Proc. of IEEE ICC, 2009,
Keynote.
[2] Bolla, R. et al., “Energy-Aware Performance Optimization for Next-Generation Green
Network Equipment”, in Proc. of ACM SIGCOMM, Workshop on Programmable routers
for extensible services of tomorrow, 2009, pp. 49-54.
[3] Qureshi, A., Weber, R., Balakrishnan, H., Guttag, J. and Maggs, B., “Cutting the Electric
Bill for Internet-Scale Systems”, in Proc. of ACM SIGCOMM, Conference on Data
communication, 2009, pp. 123-134.
[4] Odlyzko, A. M., “Data networks are lightly utilized, and will stay that way”, Review of
Network Economics, vol. 2, no. 3, pp. 210-237, 2003.
[5] Fan, X., Weber, W. D., Barroso, L.A., “Power Provisioning for a Warehouse-Sized
Computer”, in Proc. of ACM International Symposium on Computer Architecture, 2007,
pp. 13-23.
[6] Barroso, L. A. and Hölzle, U., “The Case for Energy-Proportional Computing”, in Proc. of
IEEE Computer, 2007, pp. 33-37.
[7] Kumar, R. and Mieritz, L., “Conceptualizing ‘Green IT’ and data centre power and cooling
issues”, Gartner Research Paper, No. G00150322, 2007.
[8] Zafer, M. A., Modiano, E., “A calculus approach to energy-efficient data transmission with
quality-of-service constraints”, in Proc. of IEEE/ACM, Transactions on Networking, vol.
17, issue 3, pp. 898-911, 2009.
[9] Andrews, M., Anta, A. F., Zhang, L., and Zhao, W., “Routing and scheduling for energy
and delay minimization in the powerdown model”, in Proc. of IEEE INFOCOM, 2010, pp.
21-25.
[10] Fisher, W., Suchara, M., and Rexford, J., “Greening backbone networks: reducing energy
consumption by shutting off cables in bundled links”, in Proc. of ACM SIGCOMM, Green
Networking, 2010, pp. 29-34.
[11] Nedevschi, S., Popa, L., Iannaccone, G., Ratnasamy, S., and Wetherall, D., “Reducing
Network Energy Consumption via Sleeping and Rate-Adaptation”, in Proc. of 5th USENIX
Symposium on Networked Systems Design and Implementation, 2008, pp. 323-336.
[12] Bolla, R., Bruschi, R., Davoli, F., Ranieri, A., “Performance Constrained Power Consump-
tion Optimization in Distributed Network Equipment”, in Proc. of IEEE ICC, Workshop on
Green Communications, 2009, pp. 1-6.
[13] Kant, K., “Power Control of High Speed Network Interconnects in Data Centers”, in Proc.
of IEEE ICC, 2009, pp. 145-150.
[14] Aleksic, S., “Analysis of Power Consumption in Future High-Capacity Network Nodes”, in
Proc. of IEEE/OSA JOCN, vol. 1, issue 3, 2009, pp. 245-258.
[15] Marsan, M. A., Chiaraviglio, L., Ciullo, D. and Meo, M., “Optimal Energy Savings in
Cellular Access Networks”, in Proc. of IEEE ICC, Workshop on Green Communications,
2009, pp. 1-5.
[16] Bianco, C., Cucchietti, F., Griffa, G., “Energy consumption trends in the Next Generation
Access Network - a Telco perspective”, in Proc. of INTELEC, 2007, pp. 737-742.
[17] Cioffi, J. M., Zou, H., Chowdhery, A., Lee, W., and Jagannathan, S., “Greener Copper with
Dynamic Spectrum Management”, in Proc. of IEEE GLOBECOMM, 2008, pp. 1-5.
[18] Zuckerman, D., “Green Communications – Management Included”, in Proc. of IEEE ICC,
2009, Keynote.
[19] Marsan, M. A., Meo, M., “Energy Efficient Management of two Cellular Access
Networks”, in Proc. of ACM SIGMETRICS, Performance Evaluation Review archive, vol.
37, issue 4, 2010, pp. 69-73.
 Energy-Efficient Networking: An Overview 113

[20] Hasswa, A., Nasser, N., Hassanein, H., “Generic Vertical Handoff Decision Function for
Heterogeneous Wireless Networks”, in Proc. of IFIP Conference on Wireless and Optical
Communications, 2005, pp. 239–243.
[21] Choi, Y. and Choi, S., “Service Charge and Energy-Aware Vertical Handoff in Integrated
IEEE 802.16e/802.11 Networks”, in Proc. of IEEE INFOCOM, 2007, pp. 589-597.
[22] Yang, W.-H., Wang, Y.-C., Tseng, Y.-C., and Lin, B.-S. P., “An Energy-Efficient
Handover Scheme with Geographic Mobility Awareness in WiMAX-WiFi Integrated
Networks”, in Proc. of IEEE WCNC, 2009, pp. 2720--2725.
[23] Seo, S. and Song, J., “Energy-Efficient Vertical Handover Mechanism”, in Proc. of IEICE
Transactions on Communications, vol. E92-B, no. 9, 2009, pp. 2964-2966.
[24] Petander, H., “Energy-aware network selection using traffic estimation”, in Proc. of ACM
MICNET, 2009, pp. 55-60.
[25] Barth, U., Wong, P., Bourse, D., “Key Challenges for Green Networking”, in Proc. of
Ercim News 79, 2009, pp. 13.
[26] Gyarmati, L., Trinh, T. A., “How Can Architecture Help to Reduce Energy Consumption in
Data Center Networking?”, in Proc. of ACM SIGCOMM, 2010, pp. 183-186.
[27] Chang, V., Bacigalupo, D., Wills, G., and De Roure, D., “A Categorization of Cloud
Computing Business Models” in Proc. of IEEE/ACM CCGRID, 2010, pp. 509-512.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 114-122

Localization of the Mobile Calls Based on SS7

Information and Using Web Mapping Service
Virgil CAZACU 1, Laura COBÂRZAN 2, Dan ROBU 3,
Florin SANDU 4
1
BitDefender, Bucharest, Romania, e-mail: virgil.cazacu@gmail.com
2
Softvision, Cluj-Napoca, Romania, e-mail: laura.cobarzan@gmail.com
3
Siemens Program and System Engineering, Brasov, Romania,
e-mail: dan.robu@siemens.com
4
Faculty of Electrical Engineering & Computer Science, “Transilvania” University,
Brasov, Romania, e-mail: sandu@unitbv.ro

Manuscript received October 01, 2010; revised October 20, 2010.

Abstract: Localization of the calls is a topic that has been coming up even from the
early time of the telephony. Calls made from mobile phones were even more interested
to be localized due to their mobility. This paper presents a localization solution that uses
information from the mobile network, being a technical solution that ensures the
acquisition of the localization information of the calls from the terminals in the mobile
network and which is delivering this data to a localization server. The localization
solution that is presented has three major features: receiving calls’ information from
mobile networks and obtaining the localization information from the Signaling System
#7(SS7); data processing from signaling frame and IP-transmitting of this information
to a localization server; visualization of the call location on the map. Due to client-
server architecture, users of the system can access calls locations using digital maps.

Keywords: Localization, mobile networks, service, client-server, integration.

1. Introduction
Localization of the calls is useful not only from the legal point of view but
also in case of emergencies as is for example the usage of the short number 112
or 911. In this case, the localization of the person who is in possible danger is
vital.
Using SS7 localization approach has drawbacks which are treated in the
presented solution: each mobile phone service provider supplies the localization
information within the Initial Address Message (IAM) field of the ISDN User
Part (ISUP) protocol from SS7 frame in its’ own specific format [1]. Thus, the
 V. Cazacu, L. Cobârzan, D. Robu, F. Sandu 115

solution offers the possibility to configure the necessary parameters, depending

on the place of deployment.
From the design point of view, the solution ensures the service of
acquirement of the localization information for the terminals in the mobile
networks and it is delivering this information to a localization server for calls
that are selected to be localized. The call processing, localization information
extraction and delivering these on the interface to the localization server is
performed in almost-real time, delays appearing if the load of the system is
high. This solution is adaptable with minimal costs for future changes of the
architecture. These changes might include resizing the necessary input/output
traffic and the modification of the field from the SS7 signaling frame in which
the localization information is transmitted by using parameterized components.
SS7 localization data is sent using “Cell ID” type from ISUP protocol, in the
IAM message, “Location number” field and/or “Called number address”
field [1].
The next table presents an example of localization data format that is
specific for each mobile service provider.

Table 1: Localization data format.

Network code
(e.g.72,74 or Services bit (reserved) Location area code Cell ID
4072, 4074)
2-4 digits 1 digit 5 digits 5 digits

Based on these frames, each mobile operator maintains a database with

geographic information that can offer information about the caller position
based on the positioning string. The database structure is different, according to
the telephony provider and contains the equivalent geographical coordinates for
the above data from SS7 ISUP frame as it is shown as an example in Table 2.

Table 2: Geographical coordinates database structure.

Lat Long
Cell (Grade, (Grade,
District City Street Azimuth BSC LAC
code Minutes, Minutes,
Seconds) Seconds)
Specific Int (6 Int (6 Int (3 Specific Int (4
String String String
Code digits) digits) digits) Code digits)
116 Localiz. of the Mob. Calls Based On Ss7 Inform. and Using Web Mapping Service

The location of these databases is defined by each mobile operator, which

also manages and maintains it. These databases should be interfaced with a
solution like the one presented in this paper.
The general solution of the architecture is presented in Fig.1.

Figure 1: Overview of the solution architecture.

The solution, as the picture above shows, is divided into four modules and it
is targeted to be used in operational centers that can coordinate emergency
activities:
• Extracting Localization String modules from the Localization Server
presumes the definition of rules for parsing SS7 ISUP information.
The main service of the server is to define interaction protocols as
well as Geo Information Database Server interaction.
• Geographic Information Database Server is maintaining the geo-
coordination of the radio cells.
• Client side module is handling communication between Graphical
User Interface (GUI) module and Localization Server.
• GUI module is handling specific interface functions and the digital
maps using web mapping service available on the market at the
solution’s implementation time.
 V. Cazacu, L. Cobârzan, D. Robu, F. Sandu 117

2. Description of the main components

This section presents the technical implementation of the solution modules
using as examples the case when mobile users are using the Emergency Service
112 and are customers of one of the Romanian mobile operators.

A. Localization Server
In this module, there are two major functionalities: SS7 frame parsing and
communication protocols between the client and the Geo Database Server.
The parsing module consists of two parts, one dealing with the SS7
Integrated Services Digital Network User Part (ISUP) communication, while
the other being responsible for the protocol message parsing. It is out of the
paper’s scope to detail the SS7 communication between our solution and the
mobile network. The technical approach taken is to use the JAIN ISUP API that
gives the possibility to exchange ISUP messages in the form of Java Event
Objects [2], [3].
One rule for parsing SS7 information is the fact that independently of the
mobile operator, SS7 frames are in standard format and the relevant parameters
for our solution can be found under the Initial Address heading. The relevant
parameters are presented in the Table 3.

Table 3: IAM parameters used for localization.

Parameter Name Explanation

Nature of address: either National or International.
Calling address signal: the telephone number of the
Calling party number
caller party (with a 2 digit prefix for international
calls).
Called address signals. For Emergency cases, the
Called party number
called telephone number is 112.
The localization string, that contains all the
Location number
localization information for the given provider.
The mobile operator internal ID for the radio cell
Cell ID
where the call is made from.

The phone numbers are received without the prefix digits, so in this
implementation the “Calling party number” parameter is taken into account. For
national calls a 0 digit and for international calls two 0 digits are inserted at the
beginning of the caller number. Also, to determine from which mobile operator
the call is performed and knowing that every telephone number begins for
example with 07XY, where depending on the XY digits, the solution can extract
118 Localiz. of the Mob. Calls Based On Ss7 Inform. and Using Web Mapping Service

the provider of the call based on a table correspondence and on interrogating the
portability server, if available.
In the implementation of the module, the extracted information is stored in
an object called SS7Object with fields like String callingNumber, String
calledNumber, String Nature, String LocalizationString, Date dateCreated,
String Provider. SS7Objects are sent to the Localization Server module for
further processing.
The communication protocol with the client uses sockets and when new
localization objects are received from the SS7 parsing module, the server will
send the object to the client in order to use it on the GUI. After sending the
localization object, the server is waiting for an answer from the client. If the
client does not confirm the reception of the object in the previously defined time
frame, the localization server will resend this information. The Localization
objects that are not confirmed are maintained in a waiting list. When the
localization server receives a message from GUI/Google Maps, it will delete the
corresponding object from the “waiting list”, meaning that it will not wait for
the confirmation for that object.
The communication protocol between Localization Server and the Geo
Information Database Server is done by calling the getCoordinatesBy
LocalizationString (localization_string) method which takes a string parameter,
representing the localization string and as returned value, an object which
contains the coordinates of the area from which the call was made. The
coordinates will be the latitude and the longitude, each one containing 3 fields:
degrees, minutes and seconds. The calling of the class is done using Remote
Method Invocation (RMI).

B. Geographical Information Database Server

As it was mentioned earlier in the paper, this server has to be located at each
mobile operator since it contains internal information about the place where
radio cells are deployed from geographical point of view. For completeness of
the solution description, the RMI Database Server will be presented, that has
several classes in order to extract the coordinates from the local database.
The Coordinate class is common with the RMI client, the
CoordinateInterface class which contains the remote method and the
CoordinateImplementation class, which implements the remote method as
shown below:

public Coordinate getCoordinatesByLocalizationString(String localization_string)

The Provider class has a static method String getProvider(String

localization_str), which returns the provider, based on the localization string and
 V. Cazacu, L. Cobârzan, D. Robu, F. Sandu 119

the ExtractCoordinate class has a static method that returns the coordinates,
based on the provider and the localization string, as it can be seen next capture.

public static Coordinate getCoordinatesByProvider(String provider, String

localization_str)

The ExtractFromDatabase class contains one static method for each

provider, to extract the coordinates from the local database, based on the
localization string, as shown below:

public static Coordinate ExtractVodafone(String localization_string)

A database structure example for this server is presented in Fig. 2.

Figure 2: Structure example of geo coordinates database.

C. Client side including GUI

As the communication between the client and the Localization Server is
detailed in section B, this part is focused on the usage of web mapping service
and user interface.
The graphical user interface presented in this paper has a proof of concept
oriented design. The GUI of the solution is composed of two frames: one frame
with 4 tabs: View Calls, View Archive, Options and Help. The other frame is
displaying the digital map with all its options.
120 Localiz. of the Mob. Calls Based On Ss7 Inform. and Using Web Mapping Service

Only one tab is presented in this paper, the View Calls tab which contains
the recent calls information in a list. The call information contains the exact
time of the call, the caller number, the provider and the location of the call as it
is received from the Location Server. Each call has its own checkbox, which
will specify if the call was processed or not. When a call is selected in the list,
the application is marking automatically the location of the call in the map
frame. More calls can be selected simultaneously, so the map can be marked in
more locations. Calls that are checked (processed) are deleted from the list and
the marks from the map disappear.
In order to integrate a digital map into the solution, Google Maps API was
chosen due to several considerations [5].
Google Static Maps API embeds a Google Maps image without requiring
JavaScript but the problem is that it returns the map as an image (GIF, PNG or
JPEG) in response to a HTTP request via a URL. This way, the benefits of the
zoom and navigation facilities disappear.
JXMapViewer embeds mapping abilities into Java application, but at the
solution’s development time it was not possible to use it with Google Maps or
Yahoo since there were legal restrictions.
One other strong reason why the Google Maps API was chosen for
integrating the web mapping service was the possibility to control the zoom and
navigation features from the application’s back-end.
Since Google Maps API uses JavaScript, the JWebBrowser class from the
chrriis.dj.nativeswing.components package has been used in the development of
the Java client application; this offers the possibility to have a native web
browser component in the application [4]. Because the client application has to
be operating system independent, the web browser component was developed to
use the Mozilla engine.

NSOption opt = new NSOption(JWebBrowser.useXULRunnerRuntime());

web_browser = new JWebBrowser(opt);
JWebBrowser.useXULRunnerRuntime();

The digital map from Google is loaded using the following code line [5].

web_browser.navigate(gmapfilelocation.getAbsolutePath());

The parameter gmapfilelocation points to the file containing the script which
loads the map.
In Fig. 3, the client GUI is shown together with the calls markers, each
marker descriptor containing a string defining the location to place the marker
and the visual attributes to use when displaying the mark.
V. Cazacu, L. Cobârzan, D. Robu, F. Sandu 121

Figure 3: Calls markers on the map.

Figure 4: Zoom on caller location.

122 Localiz. of the Mob. Calls Based On Ss7 Inform. and Using Web Mapping Service

Figure 4 shows a case when one call is selected from the list and
automatically the application zooms in on the location where the caller is
positioned. If another call is selected, so that two calls are on the map, the
application automatically zooms out exactly as it is necessary to display both
callers on the map.

3. Conclusion
The solution of call localization presented in this paper is still under
development since topics like high degree of availability or the ability to work
in load-balanced and failed-over conditions between locations are not
implemented. The application’s architecture has been implemented by the
authors of the paper and solution allows further improvements, in order to
enable features like accepted input traffic of a high number of simultaneous
voice calls to be implemented as easy as possible.
But the goal of the research, at least in this phase, was achieved, since the
usage of a web mapping service for calls location has been demonstrated by the
solution presented in this paper.

Acknowledgements

The authors wish to thank their colleagues who contributed with their effort
to achieve the results presented in this paper, especially the colleagues located
at the Cluj-Napoca Siemens PSE site.

References
[1] Dryburgh, L., Hewett, J., “Signaling System No. 7 (SS7/C7): protocol, architecture, and
services”, Cisco Press, 2005.
[2] Jepsen, T. C., Anjum, F., “Java in telecommunications: solutions for next generation
networks”, John Wiley & Sons, 2001.
[3] *** http://jcp.org/en/jsr/summary?id=ISUP.
[4] *** http://djproject.sourceforge.net/ns/documentation/javadoc/index.html.
[5] *** http://code.google.com/apis/maps/documentation/reference.html.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 123-135

Analysis of Neuroelectric Oscillations of the

Scalp EEG Signals
László F. MÁRTON, László SZABÓ, Margit ANTAL, Katalin GYÖRGY
Department of Electrical Engineering, Faculty of Technical and Human Sciences,
Sapientia University, Tg. Mureş, e-mail: martonlf@ms.sapientia.ro

Manuscript received November 15, 2010; revised November 30, 2010.

Abstract: Electroencephalography (EEG) or magneto-encephalography (MEG) are

usual methods adopted in clinics and physiology to extract information relative to
cortical brain activity. EEG/MEG signals are a measure of the collective neural cell
activity on restricted regions of the cortex. The brain activities ensue from the
interaction of excitatory and inhibitory populations of neurons. Their kinetics vary
depending on a particular task to be fulfilled, on the particular region involved in that
task, and in any instant during the task. In order to improve our understanding of
EEG/MEG signals, and to gain a deeper comprehension of the neuro-physiological
information contained, various mathematical models and methods have been proposed
in previous years. Based on these concepts we develop procedures for an optimal,
online and hardware based EEG signal processing. EEG recordings are analyzed in an
event-related fashion when we want to gain insights into the relation of the EEG and
experimental events. An approach is to concentrate on event-related oscillations
(EROs). The method suited to analyze the temporal and spatial characteristics of EROs,
is the time-frequency analysis namely wavelet transforms. Recently introduced wavelet-
based methods for studying dynamical interrelations between brain signals will be
discussed.

Keywords: brain-computer-interface (BCI), time frequency analysis, filters, EEG

signals, EMG signals, stationarity of signals.

1. Introduction
We propose a novel framework to analyse electroencephalogram (EEG)
biosignals from multi-trial visually-evoked potential (VEPs) signals recorded
124 L. F. Márton, L. Szabó, M. Antal, K. György

with a brain-computer interface (BCI). Electromagnetic activities of the

neuromuscular system, including electroencephalography (EEG), electro-
myography (EMG), and magneto-encephalography (MEG) signals, have been
widely used in the study of motor control in humans. VEPs signals contain
components of EMG, MEG and EEG [1], [2], [3], [4].
Oscillations are characterized by their amplitude, frequency and phase. The
amplitude of a recorded oscillation is typically between 0 and 10 µV. The
(cyclic) phase ranges are between 0 and 2π. At every scalp point at every time
moment the amplitude and phase of an oscillation can be recorded. The
frequency band of the typical recordable cortical oscillations range is usually
from 0.1 Hz to 80 Hz (highly correlated to a 256 Hz sampling rate).
Several methods exist to extract oscillations of biologically specific
frequency bands from ERO data. Among the most used are band filtering,
Fourier analysis, and wavelet analysis. Oscillating potentials derived from a
specific scalp surface, originating from the outer layer of cortex (grouped
neuron structures in the layer I cortical areas) are called visual-evoked potential
(VEP) signals. These signals are related to the brain’s response to visual
stimulation and have applications in numerous neuropsychological studies.
EROs comprise exogenous and endogenous components. Exogenous
components are obligatory responses which result on the presentation of
physical stimuli. The endogenous components (say P300 component of an ERO
signal) manifest the processing activities which depend on the stimuli’s role
within the task being performed by the subject. Usually, EEG-signals are based
on various phenomena like, for example, visual and P300 evoked potentials,
slow cortical potentials, or sensory-motor cortical rhythms. The P300 shape is
an event related potential, elicited by a generally stochastic, task related stimuli.

2. Materials and methods

We have used a BrainMaster recording system (BrainMaster AT-1 W2.5
Clinical Pro Wideband System, see Fig. 1.), battery powered, portable, two
channel 2E neurofeedback module hardware, with added cleaning gel and
conductive paste. Recordings have been made with 5 gold plated electrodes (2
recording electrodes, 2 reference electrodes and a ground electrode). The used
sampling rate was 256 Hz, sufficient for the studied frequency bands.
Recording areas were the left and right hand side of cortical surface at
Primary Motor cortex and sometime on Pre-Motor and Supplementary Motor
Cortex (Secondary Motor Cortex) on the scalp of young persons. The recorded
and sampled signals from two channels (Ch1, Ch2) are stored in text type files
and processed using MATLAB (The Math-Works Inc., Natick, MA.) platform.
The software package is elaborated by the authors based on concepts presented
 Analysis of Neuroelectric Oscillations of the Scalp EEG Signals 125

in [5]. The test recordings are usually 1 to 1.5 minutes long. The ERO intervals
alternate with relax state with a period of usually 10 seconds. The time-
frequency method used in this application is the continuous wavelet transform
based on Morlet, Paul and DOG (m = 2 and m = 6) wavelet base functions. The
Morlet and Paul bases are providing a complex continuous transform proper for
time-frequency component analysis of the recordings. For the subjects of the
experiments, during the recordings a deckchair was used to avoid extra EMG
noise created by the body stability problems.

Figure 1: The BrainMaster AT-1 System (Brain Master LTD company product image).
In the EEG–EMG experiments, subjects are trained for two different motor
tasks: a left-right or up-down movement of the closed or opened eye balls and
right or left hand movement. The scenario of the performed task is recorded in
the header of the generated file. The recording technique is a not invasive
recording method.

3. Data analysis tools

A period of baseline EEG+MEG was recorded at the beginning of each
experiment when the subjects rested. During the offline analysis, signals within
an approximately 10 s epoch were selected from the rest period and averaged to
obtain baseline EEG. Subsequently, the value of the baseline EEG+MEG was
subtracted from the entire EEG+MEG data set to acquire baseline corrected
signals. The corrected values were saved into files for further processing.
An important fact is that the magnitudes of cyclic visual-evoked potential
components are much more detectable in the 0–10 Hz frequency range. This is
important for the further analysis.
126 L. F. Márton, L. Szabó, M. Antal, K. György

From the experimental studies of VEPs, relative to the recorded oscillations, the
literature clearly depicts the delta (1–4 Hz) and theta (4–10 Hz) ranges as
containing main components of power in frequency domain for the waves. We
will consider these bands for further identification of the activity patterns [1].
Now, we are considering the important details of wavelet transform used in
these processings.
By decomposing a time series into time–frequency space, one is able to
determine both the dominant modes of variability and how those modes vary in
time. The first tested method was the Windowed Fourier Transform (WFT). The
WFT represents one of analysis tool for extracting local-frequency information
from a signal. The WFT represents a method of time–frequency localization, as
it imposes a scale or ‘response interval’ T into the analysis. An inaccuracy
arises from the aliasing of high- and low-frequency components that do not fall
within the frequency range of T window. Several window lengths must be
usually analyzed to determine the most appropriate choice of window size to be
sure to contain within the window the main, but unknown basic oscillatory
components. To avoid this difficult task, in our analysis finally we have used
wavelet transform (WT) methods.
The WT can be used to analyze time series that contain nonstationary power
at many different frequencies. The term ‘wavelet function’ is generically used to
refer to either orthogonal or nonorthogonal wavelets. The term “wavelet basis”
refers only to an orthogonal set of functions. The use of an orthogonal basis
implies the use of the discrete wavelet transform (DWT), while a nonorthogonal
wavelet function can be used with either the discrete or the continuous wavelet
transform (CWT).
A brief description of CWT is following. Assume that the recorded time
series, xn, is with equal time spacing δt (sampling period) and n = 0…N-1. Also
assume that one has a wavelet function, Ψ0(η), which depends on a non-
dimensional ‘time’ parameter η.
To be ‘admissible’ as a wavelet, this function must have zero mean and must
be localized in both time and frequency space. An example is the Morlet
wavelet, consisting of a plane wave modulated by a Gaussian function:

Ψ 0 (η ) = π −1/ 4 ⋅ ei⋅ω0 ⋅η ⋅ e−η

2
/2
(1)
This is a wavelet basis function, where w0 is the non-dimensional frequency,
here taken to be 6 to satisfy admissibility condition.
The continuous wavelet transform of a discrete sequence xn is defined as the
convolution of xn with a scaled and translated version of Ψ0(η):
N −1  (i − n) ⋅ δt 
Wn ( s ) = ∑i =0 xi ⋅Ψ * 
 s ,

(2)
 Analysis of Neuroelectric Oscillations of the Scalp EEG Signals 127

where the (*) indicates the complex conjugate. By varying the wavelet scale s
and translating along the localized time index n, one can construct a picture
showing both the amplitude of any features versus the scale and how this
amplitude varies with time. The subscript 0 on Ψ has been dropped to indicate
that this Ψ has also been normalized. It is possible to calculate the wavelet
transform using (2), and it is considerably faster to do the calculations in
Fourier space. By choosing N points, the convolution theorem allows us to do
all N convolutions simultaneously in Fourier space using discrete Fourier
transform (DFT). To ensure that the wavelet transforms at each scale s are
directly comparable to each other and to the transforms of other time series, the
wavelet function at each scale s was normalized to have unit energy.
Normalization is an important step in time-series analysis and is used at each
scale s.
Morlet wavelet function Ψ (η ) is a complex function, the wavelet transform
Wn(s) is also complex. The transform can then be divided into the real and
imaginary part, or amplitude and phase. Finally, one can define the wavelet
power spectrum as |Wn(s)|2. The expectation value for |Wn(s)|2 is equal to N times
the expectation value for the discrete Fourier transform of the time series. For a
white-noise time series, this expectation value is σ2/N, where σ2 is the variance
of the noise. Thus, for a white-noise process, the expectation value for the
wavelet transform is |Wn(s)|2 = σ2 at all n and s. Based on this knowledge, the
same logic is used to calculate the expected value of red noise. The |Wn(s)|2 / σ2
is the measure of the normalized signal value relative to white noise. As the
biological background noise is a red noise type, the normalization method is
relative to red noise as it is described in [5] (in a way as it was used in the
results of this paper).
An important concept of this study is the so called Cone of influence (COI).
The cone of influence is the region of the wavelet spectrum in which edge
effects become important because of the finite length of signal and used
window. The significance of the edge effect is defined as the e-factor (power
spectrum edge drops by a factor e−2) time of wavelet power at each scale. The
edge effects are negligible beyond the COI region. This must be considered for
an accurate analysis. In each figure, COI is represented at the edge of the
wavelet transforms (lighter area in figures).
Another important factor we have added to this analysis is the significance
level of the correlation studies. The theoretical white/red noise wavelet power
spectra are derived and compared to Monte Carlo simulation results. These
spectra are used to establish a null hypothesis for the significance of a peak in
the wavelet power spectrum (the question to be answered: is a power peak from
a wavelet figure the result of biological events or it is a result of stochastic
red/white noise effect?).
128 L. F. Márton, L. Szabó, M. Antal, K. György

The null hypothesis is defined for the wavelet power spectrum as follows. It is
assumed that the time series has a mean power spectrum, if a peak in the
wavelet power spectrum is significantly above this background spectrum, then it
can be assumed to be a true feature with a certain percent of confidence.
(‘significant at the 5% level’ is equivalent to ‘the 95% confidence interval’).
Our application is highlighting the biological events, with surrounding the
significant peaks of correlations at the confidence interval of 95%. It is
important that in biological studies the background noise can be modeled by a
red (or pink) noise. A simplest model for a red noise is the lag-1 autoregressive
[AR (1), or Markov] process. In the following figures, all significant
localization of a biological event (VEPs) in time-frequency domain is also
statistically significant. This is an important result. What is not within a
significant area, is not considered in the results. Another important result in our
analysis is the representation of the phase relationship between two recordings.
In each cross-wavelet spectrum and cross-coherence spectrum the phase
relationship is represented with arrows. A horizontal arrow to the right means
that in that time frequency domain the two biological signals are in phase (if
that domain is significant at 5% level and is not within COI domain). The
opposite arrow orientation has the meaning of opposite phase correlation. The
angle of arrows relative to horizontal line is showing the phase angle in that
time frequency domain. Our application is calculating and is representing all
these phase values. The definition of wavelet-cross-correlation and wavelet-
cross-coherence are defined in [5] and [6].

4. Results
The following figures are slim examples from our recordings and their
analysis. Fig. 2 is the amplitude/time representation of a channel signal as a
recording on the right hand side of the Motor Cortex area when the left hand has
been lifted two times during a 50 s recording session. This figure is showing
also the application menu created for this WFT type of analysis. Fig. 3 is the
WFT representation of the Fig. 2 recording in different frequency bands. The
vertical axis is for the frequency, and the horizontal one is the time axis. The
frequency bands for the different windows are 0.1 – 50 Hz for the top left, 0.1 –
8 Hz for bottom left, 24 – 30 Hz top right window and finally 30 – 36 Hz for
bottom right window. The frequency bands are representative for biological
events. A color-code represents the intensity of a time-frequency domain for
that WFT decomposition (the corresponding color-code is represented at the
right hand side of each window). The shape within each window is
characteristic for a real time motion recorded as EMG+EEG. Fig. 4 is similar
with Fig. 3 but it is represented in 3D for a better visual understanding of the
 Analysis of Neuroelectric Oscillations of the Scalp EEG Signals 129

significant domains contained in this time-series decomposition. Based on these

figures, it can be concluded that a ‘shape’ in time-frequency domain is related to
an arm motion in time domain. The automatic identification of these domains
(peaks) in time-frequency decomposition it must be related to the corresponding
arm motion what has happened in time domain. For this peak identification one
can use a pattern recognition procedure.

Figure 2: A recorded (right hand side Motor Cortex) amplitude/time representation.

Figure 3: The wavelet Fourier transform (WFT) in four different frequency bands.
As it was mentioned, the same hand lifting time event recording has been
done on both cortical sides. The left cortical side recording and decomposition
with WFT is not represented here, but it has a similar configuration. Fig. 5 is
the difference in time-frequency domain of the two side signals. It is visible that
the two side recordings are not the same as it is known from theory. This
130 L. F. Márton, L. Szabó, M. Antal, K. György

analysis is WFT, and here the COI and significance test was not used. As it was
mentioned, the WFT is very sensitive to the window length (T) used in
decomposition of the time signal [7], [8]. In the spectrum, not controllable
frequency interference is present, but the method is powerful enough to be
usable in detection and classification of not very sensitive types of motor
actions.

Figure 4: The 3D representation of figure 3 WFT decomposition.

Figure 5: The difference of the two channels recording decomposition by WFT.

A most sensitive procedure is the use of Morlet type of Wavelet transform
(WT) with the representation of COI and with significance test of biological
events based on wavelet Cross-correlation and wavelet Cross-Coherence [5],
[9]. These results are more accurate and with higher resolution in time
 Analysis of Neuroelectric Oscillations of the Scalp EEG Signals 131

frequency in comparison with the previous WFT analysis. The amount of

calculation is higher than in case of WFT but optimizing the procedures on
hardware based processor units, this method should be very powerful.

Figure 6: The two channels (Ch1, Ch2) amplitude/time representation of the recordings.
The bottom figure is showing the two superimposed signals. The green highlights are
the eye movement time sequences (sample size on vertical).
Fig. 6 is representing left-right movement of eye balls with a relax time
between them. It is visible the time sequence of left and right hand cortical side
EMG+EEG. The whole recording length is about 100 seconds. The high
amplitude signals in 25-37 seconds interval in Ch2 recording is an extra EMG, a
noise from the experiment point of view. The time sequence is containing three
eye movement events. These are between (12, 25) sec, (39, 55) sec and finally
(71, 85) seconds. These are highlighted by green line segments. In the third
(bottom) window it is visible that the Ch1 and Ch2 recordings are in opposite
phase. But this will be obvious from cross correlations calculated and
represented in Fig. 9.
The next two figures (Fig. 7 and Fig. 8) are the representation of Morlet WT
of these channel recordings. The COI and the significant areas are represented.
Domains of the signal within these closed contours are significant, outside are
not significant (should not be considered biological events). We are considering
the two parallel lines delimiting roughly the (0.75 – 1.5) Hz frequency interval.
Within these limits we can consider the events of eye movement left-right-left
in the detected time intervals. It is very obvious the presence of significant
132 L. F. Márton, L. Szabó, M. Antal, K. György

domains, and they can be easily identified. In Fig. 8, in 25 sec to 37 sec interval,
the EMG ‘noise’ is there, but the basic components are present also at much
higher frequency domains.

Figure 7: Morlet WT of Ch1 recording with localization of eye left-right movement.

Figure 8: Morlet WT of Ch2 recording with localization of eye left-right movement.

It is very important to discuss, what Fig. 9 represents. The COI is present as
the not significant domain, but also a phase relationship between the two signals
is calculated. The arrows’ orientation within the significant area is to the left
 Analysis of Neuroelectric Oscillations of the Scalp EEG Signals 133

(0.75 – 1.5Hz). This means that the correlation between the two channels, in
this frequency band is in opposition. In the recordings with half way eye
movement (left to middle, or right to middle) the phase shift is not opposite but
is around of 90/270 degrees. These phase events permit the detection of the
direction of eye movements. Fig. 9 bottom image is the normalized version of
the same cross-correlation, the so-called cross-coherence between the two
channels. This information about the interrelation of the two recordings is more
relevant to characterize the ERO contained visual evoked potentials.

Figure 9: The cross-correlation (up) and the cross-coherence of the

two channels signals.
The same analysis has been made also for the up-down eye movement
direction. The results are very similar with the left-right movement conclusions,
134 L. F. Márton, L. Szabó, M. Antal, K. György

and are not presented in this paper, but can be considered for technical
applications based on EEG+MEG recordings. The cross-coherence matrix is
processed to extract the information (signals) for further control tasks.

5. Conclusions

Every characteristic information of EROs (extracted numeric values of the

processed VEPs) is usable for further signal processing tasks. Detecting motion
related, EEG+MEG signal configuration, the possibly extracted information is
usable in external system control tasks. The sensitivity of the electrodes is not
the subject of this paper. It is obvious that with a much higher sensitivity of
electrodes, it should be possible to record much more detailed signals with
deeper correlations. The ideas used in this study are at the beginning of more
event sensitive identification and complicated command possibilities [8], [10],
[11]. We must also consider the effect of so called grid cells in the cortical area
which display regular responses to the position in a virtual, internal 2-D space.
This study should be possible using multichannel (>2) recordings, a next step in
our research.

References
[1] Nicolelis, M. A. L. and Lebedev, M. A., “Principles of neural ensemble physiology under-
lying the operation of brain–machine interfaces”, Nature Reviews Neuroscience, vol. 10, pp
530-540, July 2009.
[2] Hockensmith, G. B, Lowell, S. Y, and Fuglevand, A. J. “Common input across motor
nuclei mediation precision grip in humans”. J. Neuroscience, vol. 25, pp. 4560–4564, 2005.
[3] Caviness, J. N., Adler, C. H., Sabbagh, M. N., Connor, D. J., Hernandez, J. L., and
Lagerlund T. D., “Abnormal corticomuscular coherence is associated with the small
amplitude cortical myoclonus in Parkinson’s disease”, Mov Disorder, 2003, no.18, pp.
1157–1162, 2003.
[4] Grosse, P., Cassidy, M. J., Brown, P. “EEG–EMG, MEG–EMG and EMG–EMG frequency
analysis: physiological principles and clinical applications”, Clin Neurophysiol, no. 113, pp.
1523–1531, 2002.
[5] Torrence, C., and Compo, G. P., “A Practical Guide to Wavelet Analysis”, Bulletin of the
American Meteorological Society, vol. 79, no. 1, pp. 61-78, January 1998.
[6] Yao, B., Salenius, S., Yue, G. H., Brownc, R. W., and Liu, Z. L., “Effects of surface EMG
rectification on power and coherence analyses: An EEG and MEG study”, Journal of
Neuroscience Methods, no.159, pp. 215–223, 2007.
[7] Ermentrout, B. G, Gala´n, R. F., and Urban N. N. “Reliability, synchrony and noise”
Review: Trends in Neurosciences Cell Press, vol. 31, no. 8, pp. 428-434, 2008.
[8] Faes, L., Chon, Ki. H., and Giandomenico, N., “A Method for the Time-Varying Nonlinear
Prediction of Complex Nonstationary Biomedical Signals”, IEEE Transactions on
Biomedical Engineering, vol. 56, no. 2, pp. 205-209, February 2009.
[9] Chua, K. C., Chandran, V., Rajendra Acharya, U., and Lim C.M. “Analysis of epileptic
EEG signals using higher order spectra” Journal of Medical Engineering & Technology,
vol. 33, no. 1, 42–50, January 2009.
 Analysis of Neuroelectric Oscillations of the Scalp EEG Signals 135

[10] Guo, X., Yan, G., and He, W. “A novel method of three-dimensional localization based on
a neural network algorithm”, Journal of Medical Engineering & Technology, vol. 33, no. 3,
pp. 192–198, April 2009.
[11] Ajoudani, A., and Erfanian, A., “A Neuro-Sliding-Mode Control With Adaptive Modeling
of Uncertainty for Control of Movement in Paralyzed Limbs Using Functional Electrical
Stimulation”, IEEE Transactions on Biomedical Engineering, vol. 56, no. 7, pp. 1771-1780
July 2009.
[12] Harrison, T. C., Sigler, A., and Murphy, T. H., “Simple and cost-effective hardware and
software for functional brain mapping using intrinsic optical signal imaging”, Journal of
Neuroscience Methods, no. 182, pp. 211–218, 2009.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 136-145

Nonlinear Filtering in ECG Signal Denoising

Zoltán GERMÁN-SALLÓ
Department of Electrical Engineering, Faculty of Engineering, “Petru Maior”
University of Tg. Mureş, e-mail: zgerman@engineering.upm.ro

Manuscript received October 15, 2010; revised November 08, 2010.

Abstract: This paper presents a non-linear filtering method based on the

multiresolution analysis of the Discrete Wavelet Transform (DWT). The main idea is to
use the time-frequency localization properties of the wavelet decomposition. The
proposed algorithm is using an extra decomposition of the identified noise in order to
reduce the correlation between the electrocardiogram (ECG) signal and noise. The
linear denoising approach assumes that the noise can be found within certain scales, for
example, at the finest scales when the coarsest scales are assumed to be noise-free. The
non-linear thresholding approach involves discarding the details exceeding a certain
limit. This approach assumes that every wavelet coefficient contains noise which is
distributed over all scales. The non-linear filter thresholds the wavelet coefficients and
subtracts the correlated noise. The used threshold depends on the noise level in each of
the frequency bands associated to the wavelet decomposition. The proposed non-linear
filter acts by thresholding the detail coefficients in a particular way, in order to
eliminate the correlation between the noise and the signal. In this paper, in order to
evaluate the proposed filtering method, signals from the MIT-BIH database have been
used, and the filtering procedure was performed with added Gaussian noise. The
proposed procedure was compared with ordinary wavelet transform and wavelet packet
transform based denoising procedures, the followed parameters are the signal to noise
ratio and the denoising error.

Keywords: Wavelet decomposition, wavelet shrinkage, non-linear filtering.

 Z. Germán-Salló 137

1. Introduction
The interest in using the wavelet transform to denoise the electrocardiogram
(ECG) signals is increasing. The wavelet transform is a useful tool from time–
frequency domain, preferred for the analysis of complex signals. The
application of this transform to ECG signal processing has been found
particularly useful due to its localization in time and frequency domains. The
discrete wavelet transform- based approach produces a dyadic decomposition
structure of the signals. In this way, the wavelet packet approach is an adaptive
method using an optimization of the best tree decomposition structure
independently for each signal.

2. Methods
The continuous wavelet transform (CWT) of the signal x(t ) is defined as a
convolution of a the signal with a scaled and translated version of a base
wavelet function, [1]:
+∞ +∞
1 t −b
Wa x (b) = ∫ x(t ) ⋅ψ a,b (t ) dt = a
∫ x(t ) ⋅ψ   dt
a 
(1)
−∞ −∞

where the scale ‘a’ and translation ‘b’ parameters are nonzero real values and
the wavelet function is also real. A small value of ‘a’ gives a contracted version
of the mother wavelet function and then allows the analysis of high frequency
components. A large value of the scaling factor stretches the basic function and
provides the analysis of low-frequency components of the signal.
The discrete wavelet transform (DWT) is defined as a convolution between
the analyzed signal and discrete dilation and translation of a discrete wavelet
function. In its most common form, the DWT applies a dyadic grid (integer
power of 2 scaling with ‘s’ and ‘l’) and orthonormal wavelet basis function: .

( )
s
−
ψ ( s ,l ) ( x ) = 2 2 ψ 2 − s x − l (2)

The variables s and l are integers that scale and translate the mother function
ψ to generate wavelets (analyzing functions). The scale index s indicates the
wavelet's width, and the location index l gives its position. The mother wavelets
are rescaled, or “dilated” by powers of two, and translated by integer ‘l’ values.
In this case we have a dyadic decomposition structure. These functions define
an orthogonal basis, the so-called wavelet basis [3], [5]. The Discrete Wavelet
Transform (DWT) decomposition of the signal into different frequency bands
138 Nonlinear Filtering in ECG Signal Denoising

(according to Mallat’s algorithm [2]) can be performed by successive high-pass

and low-pass filtering (digital FIR structures) in the time domain followed by
downsampling to eliminate the redundancy, as shown in Fig.1.

Figure: Second order dyadic scale decomposition and reconstruction.

In discrete wavelet analysis of x(t) is decomposed on different scales, as
follows:
K ∞ ∞
x(t ) = ∑ ∑ d j (k )ψ j,k (t ) + ∑ aK (k )ϕ K ,k (t ), (3)
j =1 k = −∞ k = −∞

where ψ j , k (t ) are discrete analysis wavelets and ϕ K , k (t ) are discrete scaling

functions, d j (k ) are the detailed wavelet coefficients at scale 2 j and aK (k )
are the approximated scaling coefficients at scale 2 K . The discrete wavelet
transform can be implemented by the wavelet and scaling filters:

L(n ) = ϕ (t ),ϕ (2t − n ) ,

1
(4)
2

H (n ) = ψ (t ),ϕ (2t − n ) = (− 1)n L(n),

1
(5)
2
which are quadratic mirror filters [3]. The estimation of the detail signal at level
j will be obtained by convolving the approximation signal at level j-1 with the
coefficients L(n). Convolving the approximation coefficients at level j-1 with
the coefficients H(n) gives an estimate for the approximation signal at level j.
This results in a logarithmic set of bandwidths. Fig.2 shows the wavelet
decomposition tree, the time-frequency blocks and the resulted bandwidths for a
third order DWT decomposition. The first stage divides the spectrum into two
equal parts. The second stage divides the lower part in quarters and so on. This
results in a logarithmic set of bandwidths.
 Z. Germán-Salló 139

Figure 2: Time-frequency blocks and resulted relative bandwidths for third

order dyadic scale decomposition.
The wavelet packet analysis is a generalization of discrete wavelet analysis
providing a redundant decomposition procedure, where both detail and
approximation signals are split at each level into finer components. This
produces a decomposition tree as shown in Fig. 3.

Figure 3. Wavelet packet decomposition tree.

The tree contains several admissible bases one of which is the wavelet basis
itself. Having a large but finite library of bases it is possible to extract the best
basis relatively to some criterion. The best basis algorithm finds a set of wavelet
bases that provide the most desirable representation of the data relative to a
particular cost function which may be chosen to fit a particular application [7].
This basis can be any subtree of the initial entire tree. The reconstruction
procedure is similar to the inverse wavelet transform.
Figure 4 shows the time-frequency blocks for a second order wavelet packet
decomposition, L and H are the low- and highpass filters with downsamplers.
140 Nonlinear Filtering in ECG Signal Denoising

Figure 4: Time-frequency blocks for second order dyadic wavelet packet

decomposition.

3. The proposed non-linear filtering method

The goal of any denoising procedure is to extract the useful signal from the
noisy one, by eliminating the identified noise. The most simple model of the
noisy signal is the superposition of the signal and a Gaussian type of noise. The
main idea of non-linear filtering is to use the time-frequency localization
properties of the discrete wavelet decomposition. The non-linear denoising
approach assumes that every wavelet coefficient contains noise and it is
distributed over all scales. The non-linear thresholding means discarding the
detail coefficients exceeding a certain limit. There are two types of
thresholding, the soft and the hard methods. With hard thresholding the
coefficients which are lower than the threshold are set to zero. In soft
thresholding, the remaining non-zero coefficients are shrunk toward zero. In
this paper we assume that the identified noise is contained by the first order
detail coefficients. It can be observed on the first order decomposition of an
electrocardiogram signal that the first order detail coefficients shows some
correlation with some characteristic points of the signal, like local maxima as
shown in Fig. 5. The main idea in this work is to reduce this correlation by an
extra decomposition followed by a nonlinear thresholding.
 Z. Germán-Salló 141

noisy ECG
1

0.5

-0.5
0 50 100 150 200 250 300 350 400 450 500

1st order approximation

2

-1
0 50 100 150 200 250

1st order detail

0.5

-0.5

-1
0 50 100 150 200 250

Figure 5: Correlation between first order detail and approximation coefficients.

The proposed non-linear filter acts by thresholding the detail coefficients in
a particular way. The estimated noise, ne, assumed to be the first order detail
part is decomposed in a discrete wavelet structure, after that it is reconstructed
only from the second order detail coefficients and is subtracted from the initial
noise. The proposed smoothing procedure consists of:
1. third level DWT decomposition of the noisy ECG, resulting 3 detail
coefficient sets (H, HL, HLL,) and an approximation coefficient set LLL;
2. two step second level decomposition of the estimated noise (H), resulting
HHH, LHH, LH;
3. reconstruction of H’ only from HHH, LHH;
4. the detail coefficients HL, HLL are thresholded;
5. reconstruction of the filtered signal through a third order Inverse Wavelet
Transformation.
The procedure can be seen in Fig. 6. Only the detail coefficients were
thresholded, the estimated correlation between noise and signal (LH) was
removed. Thresholding the average coefficients (LLL) can lead to another type
of filtering and can be the subject of another research paper. Both of hard and
soft thresholds have been applied [4], [6].
142 Nonlinear Filtering in ECG Signal Denoising

Figure 6: The proposed non-linear filtering procedure.

4. Results
To estimate the ability of this denoising procedure the followed parameters
were the obtained signal to noise ratio and the absolute value of the error
defined as:
 PoriginalECG 
SNR1[dB] = 10 lg  (6)
 PoriginalECG − PdenoisedECG 
Error = abs (originalECG − denoisedECG ) (7)
Figure 7 presents the original signal, the wavelet transform based denoised (soft
thresholding) signal and the result of proposed nonlinear filtering. One can see
(visual analysis) that the new method preserves more accurate information
about signals characteristic points than the DWT based procedure. Figure 8
shows the obtained signal-to-noise ratios by different denoising methods. The
results show that the proposed method performs better denoising if the signal
has lower signal-to-noise ratio.
 Z. Germán-Salló 143

noisy ECG
1

0.5

-0.5
0 50 100 150 200 250 300 350 400 450 500

soft denoised ECG

1

-1
0 50 100 150 200 250 300 350 400 450 500

new method denoising ECG

1

-1
0 50 100 150 200 250 300 350 400 450 500

Figure 7: The results obtained using different denoising procedures.

Comparing SNRs

12
10
8
SNR
6
SNRwp
SNR

4
SNRnew
2
SNRdwt
0
-2 1 2 3 4 5

-4
nr of tests

Figure 8: Comparison between signal-to-noise ratios obtained by

different denoising methods.
144 Nonlinear Filtering in ECG Signal Denoising

denoising errors

0.3

0.25

0.2
ErrorWP
errors

0.15 ErrorNEW
ErrorDWT
0.1

0.05

0
SNR 10.15 5.87 2.72 2
SNR

Figure 9: Comparison between different denoising errors.

Figure 7 presents the filtering errors for different methods, the proposed
procedure seems to be slightly better than the wavelet packed based denoising
method.

5. Conclusions
The main idea was to estimate the correlation between the noise and the
signal. The discrete wavelet decomposition algorithm offers a good opportunity
to have access to different time-frequency domains in order to perform non-
linear filtering. An extra decomposition of the noise was used to reduce this
correlation. This method was compared with ordinary wavelet decomposition
and wavelet packet decomposition based filtering techniques. Wavelet and
wavelet packets based denoising methods gave different performances, due to
the different division strategies of the signal decomposition structures.

References
[1] Donoho, D. L., “De-noising by soft-thresholding”, IEEE, Transaction on Information
Theory, vol 41, no 3, pp. 613-627, 1995.
[2] Aldroubi, A., and Unser, M.: “Wavelets in Medicine and Biology”, CRC Press New York
1996.
[3] Misiti, M., Misiti, Y., Oppenheim, and G., Poggi, J-M.: “WaveletToolbox. For Use with
Matlab. User’s Guide”, Version 2, The MathWorks Inc 2000.
[4] Coifman, R. R., and Wickerhauser, M. V., “Entropy-based algorithms for best basis selec-
tion”, IEEE Transaction on Information Theory, vol. 38, no 2, pp. 713–718, 1992.
 Z. Germán-Salló 145

[5] Mallat, S. A., “A theory for multi-resolution signal decompositions: The wavelet represent-
tation”, IEEE Trans. On Pattern Analysis and Machine Intelligence, vol. 11, no. 7, pp. 674-
693, 1989.
[6] Donoho, D. L., and Johnstone, I. M. “Ideal spatial adaptation by wavelet shrinkage”, Bio-
metrika, Engineering in Medicine and Biology 27th Annual Conference, Shanghai, China,
September 1-4, vol. 81, 2005, pp. 425-455.
[7] Chang, C. S., Jin, J., Kumar, S., Su, Q., Hoshino, T., Hanai, M., and Kobayashi, N.,
“Denoising of partial discharge signals in wavelet packets domain”, IEEE Proceedings of
Science, Measurement and Technology, vol. 152, no. 3, pp. 129-140, 2005.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 146-158

Microstructural Modification of (Ti1-xAlxSiy)N Thin Film

Coatings as a Function of Nitrogen Concentration
Domokos BIRÓ 1, Sándor PAPP 2, László JAKAB-FARKAS 2
1
Department of Mechanical Engineering, Faculty of Technical and Human Sciences,
Sapientia University, Tg. Mureş, e-mail: dbiro@ms.sapientia.ro
2
Department of Electrical Engineering, Faculty of Technical and Human Sciences,
Sapientia University, Tg. Mureş, e-mail: spapp@ms.sapientia.ro, jflaci@ms.sapientia.ro

Manuscript received November 02, 2010; revised November 24, 2010.

Abstract: In the past few years a considerable research activity has been directed
towards understanding the structure forming phenomena of nanostructured materials. A
wide range of composition and structure of multiphase material systems have been
investigated in order to allow a fine tuning of their functional properties. We have
studied one of the most promising material systems composed of Al, Ti, Si, N, where a
significant reduction in grain growth was achieved through control of phase separation
process. Nanostructured (Al, Ti, Si)N thin film coatings were synthesized on Si(100)
and high speed steel substrates by DC reactive magnetron sputtering of a planar
rectangular Al:Ti:Si=50:25:25 alloyed target, performed in Ar/N2 gas mixture. For all
the samples we have started with deposition of a nitrogen-free TiAlSi seed layer. Cross-
sectional transmission electron microscopy investigation (XTEM) of as-deposited films
revealed distinct microstructure evolution for different samples. The metallic AlTiSi
film exhibited strong columnar growth with a textured crystalline structure. Addition of
a small amount of nitrogen to the Ar process gas caused grain refinement. Further
increase of nitrogen concentration resulted in fine lamellar growth morphology
consisting of very fine grains in close crystallographic orientation showing up clusters
of the chain-like pearls in a dendrite form evolution. Even higher N concentration
produced homogeneous compact coating, with an isotropic structure in which we can
observe nanocrystals with average size of ~3nm. The kinetics of structural
transformations is explained in the paper by considering the basic mechanism of
spinodal decomposition process.

Keywords: Nanostructured (Ti,Al,Si)N thin films, cross-sectional transmission

electron microscopy (XTEM) investigation, grain refinement, lamellar growth
morphology, spinodal decomposition.
 D. Biró, S. Papp, L. Jakab-Farkas 147

1. Introduction
In the last decade intense research activity was devoted to investigate
nanocomposite coating materials, consisting of a nanocrystalline transition
metal nitride and an amorphous tissue phase. These coating materials are
characterized by high hardness [1], enhanced elasticity [2] and high thermal
stability [3], which define their unusual mechanical and tribological properties.
Various studies revealed that in multiphase nanocomposite materials the
microstructure and the ratio between hardness and elastic modulus H/E are
important in the coating performance [4]. Recently the most studied material is
the quaternary (Ti, Al, Si)N nitride system revealing the most promising results.
As it has been suggested by Veprek [5], in nanocomposite materials the
structure and size of the nanocrystalline grains embedded in the amorphous
tissue phase together with the high cohesive strength of their interface, are the
main parameters which control the mechanical behavior of the coatings. The
reported results revealed that adatom mobility may control the microstructure
evolution in multi-elemental coating systems, where the substrate temperature
and the low energy ion/atom arrival ratio have significant effect on the growth
of nanocrystalline grains.
The microstructure and growth mechanism of arc plasma deposited TiAlSiN
(35 at.% Ti, 42 at.% Al, 6.5at.% Si) thin films were investigated by Parlinska et
al. [6, 7]. It was shown that compositionally graded TiAlSiN thin films with Ti-
rich zone close to the substrate exhibited crystalline structure with pronounced
columnar growth. Addition of Al+Si leads to a grain refinement of the coatings,
and a further increase of the Al+Si concentration results in the formation of
nanocomposites, consisting of equiaxial, crystalline nanograins surrounded by a
disordered, amorphous SiNx phase.
In our study (Al, Ti, Si)N single layer thin film coatings were deposited on
Si(100) and high-speed steel substrates by DC reactive magnetron sputtering.
We investigated the micro structural modification of (Ti1-xAlxSiy)N thin film
coatings as a function of nitrogen concentration by conventional transmission
electron microscopy.

2. Experimental details and characterization technique

Deposition experiments of (Al, Ti, Si)N quaternary nitride coatings were
carried out in a laboratory scale equipment by DC driven magnetron sputtering,
whose details are reported elsewhere [8]. The three independently operated
sputter sources were closely arranged side by side on the neighboring vertical
walls of a 75 l octagonal all-metal high vacuum chamber. The closely disposed
UM magnetrons arranged on an arc segment were highly interacting by their
148 Microstruct. Modif. of (Ti1-xAlxSiy)N Thin Film Coatings as a Function of Nitrogen Conc.

magnetic fields, leading to a far extended active plasma volume. In the

presented deposition experiments only the central magnetron source was active,
while the adjacent two magnetron sources contributed only in the closed
magnetic field. A high purity planar rectangular target material of alloyed
AlTiSi was used. Elemental composition of the PLANSEE GmbH. alloyed
target was 50 at.% Al, 25 at.% Ti, and 25 at.% Si, with 165×85×12 mm3 in size,
which was partially covered on the erosion zone with a high purity 99.98% Ti
sheet. Prior to deposition in the vacuum chamber a base pressure of ·10 2 -4 Pa
was established by operating a 540 l/s turbo molecular pump.
Polycrystalline high-speed steel (HSS) substrates were used for tribological
measurements, and native SiO2 covered mono-crystalline <100> Si wafers were
also used as substrates for XTEM microstructure investigation of the as-
deposited (Al, Ti, Si)N single layer thin film coatings. The target-to-substrate
distance was kept constant at 110 mm in all runs. The substrates were
positioned in static mode on a molybdenum sheet substrate holder, which
allowed application of Us = –75 V bias voltage. The Mo sheet was externally
heated to a controllable substrate temperature of Ts = 400 ºC.
Prior to the starting of the deposition process, the surface of the substrates
was plasma-etched by a DC glow discharge in argon for 10 min at 0.8 Pa, while
the bias voltage was limited up to 350 V. During the ion etching of substrates,
the target surfaces were also sputter cleaned by operating the magnetron unit at
limited discharge power (pre-sputtering power of 150 W). The substrate
surfaces were shielded during the pre-sputtering. The reactive sputtering process
was performed in a mixture of Ar and N2 atmosphere at 0.28 Pa pressure.
During the reactive sputtering process the nitrogen mass flow rate was
controlled with an Aalborg DFC 26 flow controller, which contains a solenoid
valve. The argon gas throughput (qAr = 6.0 sccm, measured by GFM 17 Aalborg
mass flow meter) was adjusted by a servo motor driven mass flow rate
controller (MFC-Granville Phillips S 216).
A constant sputtering power with a current density of 10 mA·cm-2 was
selected for about 10 min sputter cleaning of the targets. During deposition the
discharge power at the target surface was raised to 500 W, and the development
of coating started with the deposition of a 50 nm thick AlTiSi metallic seed-
layer performed in pure Ar atmosphere. In the next step of deposition an (Al, Ti,
Si)(N) interlayer with gradient composition was reactively grown, while PC
controlled N2 flow rate was increased slowly up to the pre-selected value. The
argon gas flow was kept constant at 6.0 sccm. The typical thickness of the
coatings was approximately 2 μm.
The experimental conditions for preparation of (Al, Ti, Si)N coatings are
listed in Table 1. The microstructure and growth morphology of the as-
deposited coatings was examined by use of a 100 kV operated JEOL 100U
 D. Biró, S. Papp, L. Jakab-Farkas 149

transmission electron microscope. In order to prepare cross-sectional XTEM

samples for transverse observations, the samples were subjected to ion-milling
in view of thinning up to electron beam transparency. Thin specimens for
XTEM investigations were prepared in a Technoorg-Linda Ltd. model 4IV/H/L
ion beam thinning unit. High energy ion beam thinning was completed with a
low angle and low energy (200 eV) ion beam process in order to eliminate the
amorphous by-products and etching defects induced by the high energy ions.

Table 1: Summary of deposition parameters used for preparation of (Al, Ti, Si)N
coatings: Pd- DC magnetron discharge power, qN2- nitrogen mass flow rate, Ts- substrate
temperature, Us- substrate bias voltage.

Samples Pd [W] qN2 [sccm] Ts [°C] Us [V]

TiS_01 500 - 400 -75
TiS_07 500 1.0 400 -75
TiS_08 500 1.0 400 -75

TiS_04 500 2.0 400 -75

TiS_09 500 2.0 400 -75

TiS_10 500 2.0 100 -75

TiS-06 500 3.0 400 -75
TiS-03 500 4.0 400 -75
TiS-05 500 5.0 400 -75
TiS-02 500 6.0 400 -75

Bright-field (BF) and dark-field (DF) transmission imaging techniques were

used for microstructure investigation of the as-prepared samples. The
identification of the crystallographic phases and the crystal orientation were also
performed by evaluation of selected area electron diffraction (SAED) patterns.
The SAED patterns were processed with the ΄Process-Diffraction΄ software tool
developed by Labar [9].

3. Results and discussion

In this XTEM study of our prepared (Al, Ti, Si)N coatings, we combined
direct imaging and selected area electron diffraction modes (SAED), which
150 Microstruct. Modif. of (Ti1-xAlxSiy)N Thin Film Coatings as a Function of Nitrogen Conc.

facilitate obtaining information on the microstructure morphology, grain size

and crystallographic preferred orientation.
Cross-sectional image of the metallic polycrystalline AlTiSi coating’s
columnar structured morphology is given in Fig. 1. The micrograph shows that
the crystallite in a conical shape evolution starts close to the substrate and grows
in a competitive mode. The columns with crystalline grains grow through the
entire film thickness up to the top surface of the coating. The columnar grains
are normally oriented to the substrate’s surface. The large AlTi(Si) crystallites
of approximately 80 nm in width are separated by the more electron-transparent
TiSi2 phase segregated to the grain boundaries. The SAED patterns taken from
the bulk region of the film exhibit well-defined spotted diffraction rings (not to
be seen here). The SAED pattern taken from the near substrate region of the
coating proved crystalline character of the Si doped TiAl film (inset of Fig. 1.).
The phases that can be derived from the diffraction pattern are mainly fcc-B1
NaCl-type of TiAlSi solid solution crystallites. The simulation of the diffraction
rings was performed taking into account an fcc-type structure. It can be clearly
seen the <200> preferential growth direction, indicated by significant brightness
increase due to reflections from (200) crystallite planes that are oriented in
parallel to the growing surface.
The chemical composition of the as deposited TiAlSi thin film was evaluated
from EDS spectra analysis, and found to be of 34 at.% Ti, 46 at.% Al and 20
at.% Si.

Figure 1: XTEM micrograph showing a cross-sectional view of the columnar structured

polycrystalline TiAlSi thin film coating (TiS_01 sample). The inset of SAED electron
diffraction pattern indicates an fcc-structured TiAlSi solid solution phase, showing (200)
texture evolution in the growing direction.
 D. Biró, S. Papp, L. Jakab-Farkas 151

By adding a small amount of nitrogen as reactive gas in the argon process

gas, the growth morphology of the film dramatically changed (Fig. 2).

Figure 2: XTEM micrograph and SAED electron diffraction pattern of the (AlTiSi)N
coating grown by nitrogen flow rate of qN2=2 sccm (sample TiS_09): a). Bright field
(BF) image indicates a weakly columnar structure evolution in close vicinity of
transition zone from the ternary TiAlSi sub-layer to the quaternary (AlTiSi)N
overgrown layer, b). On the enlarged micrograph slightly curved fine lamellar growth
morphology could be identified inside the individual columns.
For a nitrogen flow rate of qN2 = 2 sccm the microstructure indicates a weak
columnar evolution (Fig. 2a). Slightly curved fine lamellar growth morphology
could be identified inside of the individual columns (see on the enlarged
micrograph, Fig. 2b).
Selected area electron diffraction pattern (SAED) performed in close vicinity
of transition zone –including also the Si(100) bulk–, claims for a two-phase
mixture of fcc-TiAlN nanocrystals embedded in an amorphous tissue phase
(inset of Fig. 2b). Furthermore, (200) preferential growth in close vicinity of
transition zone from the ternary TiAlSi sub-layer to the quaternary (AlTiSi)N
overgrown layer was slightly maintained. The presence of continuous reflection
rings suggests a grain refinement of the coating with a strong tendency for
evolution from the textured polycrystalline phase to a mixture of nanocrystalline
AlTi(Si)N phase and possible formation of silicon nitride amorphous tissue
phase.
Chemical composition of the as deposited (Ti, Al, Si)N thin films was
evaluated from EDS spectra, and found to be 23 at.% Ti, 46 at.% Al, 26 at.%.
Si, and about 5 at.% N.
152 Microstruct. Modif. of (Ti1-xAlxSiy)N Thin Film Coatings as a Function of Nitrogen Conc.

With further increase of nitrogen amount in the coating deposition process

(TiS_05 sample performed by qN2 =5 sccm nitrogen flow rate, which determined
in our reactive magnetron sputtering process a pN = 0.0016 mbar nitrogen partial
pressure) the crystalline character of the coating disappeared and formed an
isotropic nanocomposite structure, possibly consisting of nanocrystalline
Ti3AlN/TiSi2 grains of 2…3 nm in size surrounded by SixNy and/or AlN
amorphous matrix phase (Fig. 3a). The SAED diffraction pattern revealed that
an increased nitrogen amount leads to a nanocomposite structure, consisting of
equiaxially distributed Ti3AlN/TiSi2 nanocrystalline grains surrounded by a
very thin amorphous Si3N4 phase.

Figure 3: Bright field XTEM micrograph of (AlTiSi)N thin film deposited with an
increased nitrogen flow rate (TiS_05 sample, qN2=5 sccm): a). The coating’s
microstructure indicates the development of a competitive columnar evolution of the
ternary TiAlSi sub-layer followed by the growth of the quaternary (AlTiSi)N overgrown
layer developed in an isotropic morphology. b). The enlarged micrograph clearly shows
a random distribution of very fine nc-(Al1-xTix)N grains, having an average size of ~ 3
nm, with disordered grain limiting boundaries.

The chemical composition of the as deposited thin film was evaluated from
EDS spectra analysis, and found to be 12 at.% Ti, 19 at.% Al, 23 at.% Si and 46
at.% N. The oxygen impurity content decreased to about 0.2 % which was
related to a prolonged outgassing process of the vacuum chamber and thermal
degassing of the substrate by heating to 600 °C prior to the deposition process.
Veprek et. al [10, 11] in their recently published review paper emphasized
that ultra-hard nanocomposite nitride phase coatings based on (Ti, Al, Si)N
elemental composition can be managed by well-controlled plasma and
deposition conditions. The development in a periodic structure of the
 D. Biró, S. Papp, L. Jakab-Farkas 153

nc-(Al1-xTix)N/a-Si3N4 nanocomposite coatings, composed from the uniformly

distributed (Al1-xTix)N nanocrystals and amorphous tissue phase of Si3N4 with
about one monolayer (ML) thickness, was explained by the spontaneous
separation in spinodal decomposition and self-organization upon phase
segregation process. The strong immiscibility of Si3N4 phase in crystalline
TiAlN phase, as well the absence of the Ostwald ripening, are appropriate
conditions for the development of nanocomposite coatings.
Favvas and Mitropoulos [12] in their paper (see also the cited paper therein)
compiled the IUPAC definition of spinodal decomposition: “A clustering
reaction in a homogeneous, supersaturated solution (solid or liquid) which is
unstable against infinitesimal fluctuations in density or composition. Therefore,
homogeneous solution separates spontaneously into two phases, starting with
small fluctuation and proceeding with decrease in the Gibbs free energy without
a nucleation barrier. ”
It is known from thermodynamic theory that by fast cooling of a
homogeneous solution the diffusion process occurs with net reduction in Gibbs
free energy of the system. The free energy of mixing, ∆G mix , defined by the
thermodynamic equation of Gibbs has a general form:

∆G mix = ∆H mix − T ⋅ ∆S mix , (1)

where the enthalpy change ∆H mix is associated with the interactions between
the components of the mixture, and the entropy change ∆S mix is associated
with the random mixing of components.
At high temperature of the system, with partially miscible components A and
B, the components give rise to a continuous solution (in a liquid or solid phase)
due to a complete solubility. At lower temperature the solution becomes
unstable and may exists compositional ranges where the co-existing solid
phases are more stable.
Therefore, during the cooling of liquid, phase separation proceeds in order to
minimize the free energy. If the super-cooling of the homogeneous solution
takes place into the coherent spinodal region, defined by the compositional
interval between points of inflexion on the free energy diagram as a function of
molar composition G = f ( X A ) , the phase separation process occurs by the
decomposition of the homogeneous solution. Decomposition is favored by
small fluctuations produced in the chemical composition or infinitesimal
perturbations in chemical potential. In accordance with Fick's first law, the
diffusion flux of a component, e.g. jA is proportional to the concentration
∂C A
gradient of the respective component A, :
∂x
154 Microstruct. Modif. of (Ti1-xAlxSiy)N Thin Film Coatings as a Function of Nitrogen Conc.

∂C A
j A = − DA ⋅ , (2)
∂x
where DA stands for diffusion coefficient in the first empirical law of Fick.
The diffusion flux jA can be driven also by the free energy gradient of
component A:
∂G
j A = −C A ⋅ µ A ⋅ A , (3)
∂x
where µ A stand for the mobility constant of component A.
From the above equations the diffusion coefficient DA can be written as a
derivative function of the Gibbs free energy in respect to the concentration:
∂G
DA = C A ⋅ µ A ⋅ A (4)
∂C A
∂G A
If the diffusion coefficient is positive, D A > 0, i.e. > 0, the chemical
∂C A
potential gradient has the same direction as the concentration gradient, therefore
the diffusion flux occurs along the concentration gradient. For diffusion
∂G A
coefficient D A < 0, i.e. < 0, the diffusion flux occurs against to the
∂C A
concentration gradient.
By correlating the phase composition diagram (i.e. a diagram of phases,
where the dependence for temperature T versus the molar fractions XA and XB of
components indicates the composition ranges for equilibrium phases) and the
diagram of the free energy change versus the molar fraction of the mixture, it
can be seen that below the spinodal (where the second derivative of the free
∂ 2 ∆G
energy of mixing versus the molar fraction XA is zero, = 0) the system is
∂X A 2
unstable (Fig. 4).
For negative values of the second derivative of the free energy of mixing
∂ 2 ∆G
versus the molar fraction XA, e.g. of component A, i.e. < 0, the
∂X A 2
homogeneous supersaturated solution is unstable and spinodal decomposition
may proceed.
The spinodal decomposition process happens in an unstable region, where
further instability is caused by the small fluctuation occurred in the local
concentration, while the diffusion process takes place from the lower to higher
concentration (i.e. “up-hill” diffusion).
 D. Biró, S. Papp, L. Jakab-Farkas 155

Figure 4: Illustration of the spinodal decomposition on the phase separation mechanism:

a). Equilibrium phase diagram with the phase boundary limits: for a temperature T of
the system points a and c stand for compositional interval of miscibility gap (with two
stable solid phases), the interval between points b and d stand for the spinodal limits; b).
Diagram of the Gibbs free energy changes in the mixing process of two partially
miscible components A and B, boundary line (1) separates a stable region (s) of the
homogeneous liquid phase and the metastable region (m) of the precipitated solid
phases. Line (2) is the limit of spinodal, below of which the thermodynamically
unstable system (u) turns from one phase (liquid) to a mixture of two phases (solid)
system with different elemental composition

Therefore, spinodal decomposition process is a spontaneous reaction, which

is the main route to develop periodic structures with uniform size.
156 Microstruct. Modif. of (Ti1-xAlxSiy)N Thin Film Coatings as a Function of Nitrogen Conc.

Our experimental results on fine lamellar growth morphology of (Ti, Al, Si)N
nitride coatings, consisting of chain-like pearls in a dendrite evolution with very
fine grains in close crystallographic orientation, may be explained in accordance
with Veprek’s theory by partial spinodal decomposition and phase segregation
during the film’s growth while percolation threshold composition is attained by
an increased nitrogen activity.
On the other hand, the increase of deposition rate induces a decrease of the
surface mobility related to the decrease of the ion-to-atom arrival rate ratio.
These particular deposition conditions explain the columnar structure of TiAlSi
solid solution crystallites, which can be clearly observed in the XTEM image of
TiS_01 sample. Addition of minor amounts of nitrogen leads to an
encapsulation of the growing TiAl(Si)N crystallites by process segregated
amorphous phase.
From the detailed observation of the SAED diffuse diffraction pattern of
sample TiS_05 obtained with an increased nitrogen flow rate, the presence of an
amorphous phase surrounding the Ti3AlN nanocrystallites can be attributed to
Si3N4 matrix phase (inset of Fig. 3a). The formation of amorphous TiSi2 and
AlN phase due to the partial segregation of Al and Si atoms should be also
considered due to the effect of enhanced ion bombardment provided by the
focused plasma beam that is characteristic to the present experimental
conditions [8].
When the atomic surface mobility in the growing film is adequate, the
segregated atoms can nucleate and develop the new phases controlled by
deposition temperature and by the energy transfer from an increased incident
ion-to-atom arrival rate ratio [13-15].
Further experiments are in progress to investigate the influence of the
deposition temperature on structure evolution of (TiAlSi)N coatings.

4. Conclusions
In the present work it was shown that:
a) Columnar structure of polycrystalline AlTiSi thin film coating evolved by
non-reactive DC magnetron sputtering applied to Al:Ti:Si = 50:25:25 alloyed
target (performed in pure Ar atmosphere, where the 500 W discharge power,
Ts = 400 ºC substrate temperature and Us = –75 V bias voltage were held
constants).
b) Addition of a small amount of nitrogen to the process gas leads to a grain
refinement of polycrystalline (Ti, Al, Si)N thin films. Increase of N concen-
tration (qN2 = 2 sccm flow rate) resulted in fine lamellar growth morphology of
coatings, showing chain-like pearls in a dendrite evolution, consisting of
clusters of very fine grains in close crystallographic orientation.
 D. Biró, S. Papp, L. Jakab-Farkas 157

c) Further increase in the nitrogen amount (qN2 = 5 sccm) leads to evolution

of a nanocomposite coating consisting of crystalline Ti3AlN nanograins in 2…3
nm size surrounded by an amorphous SixNy covalent nitride and/or AlN matrix
phase.
d) Kinetics of the structural transformations were explained by considering
the basic mechanism of spinodal decomposition process.

Acknowledgements
The authors are thankful for the financial support of this project granted by
Sapientia Foundation − Institute for Scientific Research, Sapientia University.
The EDS analyses of the investigated coatings were performed in a CM 20
Philips 200kV TEM electron microscope by Professor P. B. Barna from
RITPMS, Budapest. Professor P. B. Barna’s contribution to investigating
elemental composition and the valuable discussions are highly appreciated.

References
[1] Yoon, J. S., Lee, H. Y., Han, J., Yang, S. H., Musil, J., “The effect of Al composition on
the microstructure and mechanical properties of WC–TiAlN superhard composite
coating”, Surface and Coatings Technology, vol. 142-144, pp. 596-602, 2001.
[2] Duran-Drouhin, O., Santana, A. E., Karimi, A., “Mechanical properties and failure modes
of TiAl(Si)N single and multilayer thin films”, Surface and Coatings Technology, vol.
163-164, pp. 260-266, 2000.
[3] Musil, J. and Hruby, H., “Superhard Nanocomposite Ti1-xAlxN Films Prepared by
Magnetron Sputtering” , Thin Solid Films, vol. 365, pp. 104-109, 2000.
[4] Ribeiro, E., Malczyk, A., Carvalho, S., Rebouta, L., Fernandes, J. V., Alves, E., Miranda,
A. S., “Effect of ion bombardment on properties of d.c. sputtered superhard (Ti,Si,Al)N
nanocomposite coatings”, Surface and Coatings Technology, vol. 151-152, pp. 515-520,
2002.
[5] Veprek, S., “New development in superhard coatings: the superhard nanocrystalline-
amorphous compositics”, Thin Solid Films. vol. 317, pp. 449-454, 1998.
[6] Parlinska-Wojtan, M., Karimi, A., Cselle, T., Morstein, M., “Conventional and high
resolution TEM investigation of the microstructure of compositionally graded TiAlSiN
thin films”, Surface and Coatings Technology, vol. 177-178, pp. 376-381, 2004.
[7] Parlinska-Wojtan, M., Karimi, A., Coddet, O., Cselle, T., Morstein, M., “Characterization
of thermally treated TiAlSiN coatings by TEM and nanoindentation”, Surface and
Coatings Technology, vol. 188-189, pp. 344-350, 2004.
[8] Biro, D., Barna, P. B., Szekely, L., Geszti, O., Hattori, T., Devenyi, A., “Preparation of
multilayered nanocrystalline thin films with composition-modulated interfaces”, Nuclear
Instruments and Methods in Physics Research vol. 590, pp. 99-106, 2008.
[9] Lábár, J. L., “ProcessDiffraction: A computer program to process electron diffraction
patterns from polycrystalline or amorphous samples”, Proceedings of the XII EUREM,
Brno (L. Frank and F. Ciampor, Eds.), vol. III., pp. I 379-380, 2000.
158 Microstruct. Modif. of (Ti1-xAlxSiy)N Thin Film Coatings as a Function of Nitrogen Conc.

[10] Veprek, S., Veprek-Heijman, M. G. J., Karvankova, P., Prochazka, J., “Different approa-
ches to superhard coatings and nanocomposites”, Thin Solid Films, vol. 476, pp. 1-29,
2005.
[11] Veprek, S., Zhang, R. F., Veprek-Heijman, M. G. J., Sheng, S. H., Argon, A. S., “Super-
hard nanocomposites: Origin of hardness enhancement, properties and applications”, in
Surf. And Coat. Technol., vol. 204, pp. 1898-1096, 2009.
[12] E. P. Favvas, A. Ch. Mitropoulos: “What is spinodal decomposition?”, Journal of Engi-
neering Science and Technology Review, vol. 1, pp. 15-27, 2008.
[13] Carvalho, S., Rebouta, L., Ribeiro, E., Vaz, F., Dennnot, M. F., Pacaud, J., Riviere, J. P.,
Paumier, F., Gaboriaud, R. J., Alves, E., “Microstructure of (Ti,Si,Al)N nanocomposite
coatings”, Surface and Coatings Technology, vol. 177-178, pp. 369-375, 2004.
[14] Carvalho, S., Rebouta, L., Cavaleiro, A., Rocha, L. A., Gomes, J., Alves, E.,
“Microstructure and mechanical properties of nanocomposite (Ti,Si,Al)N coatings”, Thin
Solid Films, vol. 398-399 , pp. 391-396, 2001.
[15] Vaz, F., Rebouta, L., Goudeau, P., Pacaud, J., Garem, H., Riviere, J. P., Cavaleiro, A.,
Alves, E., “Characterization of TiSiN nanocomposite films”, Surface and Coatings
Technology, vol. 133-134 , pp. 307-313, 2000.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 159-165

On Some Pecularities of Paloid Bevel Gear

Worm-Hobs
Dénes HOLLANDA, Márton MÁTÉ
Department of Mechanical Engineering, Faculty of Technical and Human Sciences,
Sapientia University, Tg. Mureş,
e-mail: hollanda@ms.sapientia.ro, mmate@ms.sapientia.ro

Manuscript received November 11, 2010; revised November 24, 2010.

Abstract: The paper discusses the generating surfaces of the paloid bevel worm hob
used for paloid gear cutting. A pitch modification is required by this type of tool in
order to ensure the optimal contact pattern by gearing. As a consequence of this
modification the flank line of the plain gear tooth results as a paloid- a more general
shaped curve than the theoretical involute of the basic circle. Equations of the
generating surfaces result as equations of a generalized arhimedic bevel helical surface
presenting those modifications that arise from the variation of the tooth thickness.
The first subsection discusses the essential geometrical peculiarities of the paloid
worm hob. Here it is to remark that the most important characteristic of the tool is the
variation of the tooth thickness on the rolling tape generator. The tooth thickness has its
minimum value at the middle of the generator, and is maximum on the extremities. As a
consequence, the generated gear tooth presents an opposite variation of thickness. The
thickness variation is realized by moving the relieving tool on an ellipse, but this is not
the only possible trajectory to be used.
The second subsection presents the generalized mathematical model of the tooth
thickness variation. Starting from the ellipse used by classical relieving technologies,
and writing the equation of the ellipse reported to the coordinate system of the paloid
hob it results the radius function of the revolved surface of the reference helix. This
function is used in its condensed form. With this, the developed mathematical model
can be used for other forms of the relieving tool trajectory.
The next paragraphs present the matrix transformations between the coordinate
systems of the hob and the relieving tool. Finally the parametric equations of the hob
tooth flank are obtained. These equations depend on the radius modification function.
Using other relieving trajectories that differ from an ellipse, other tooth flank forms will
be obtained. Using this model, the flanks of the cut gear tooth can be easily written for
all types of trajectories.
160 D. Hollanda, M. Máté

Keywords: paloid bevel worm hob, tooth thickness modification, generating

surface.

1. General description
Paloid bevel gears are realized on Klingelnberg type teething machine-tools,
using paloid bevel gear worm-hobs [1], [6]. These tools present a straight-
shaped edge in their axial section. As a conclusion, the origin surface of the
paloid worm hob is an Arhimedic bevel worm having the half taper angle of 30°
as shown in Fig. 1. [4], [5]. The chip-collecting slots are axially driven. In order
to realize the clearance angle on all edges, a helical relieving, perpendicularly
oriented to the bevel generator is allowed.

Figure 1: The paloid worm hob characteristic dimensions.

In order to ensure the good positioning of the contact patch by paloid gears,
the pitch of the tool is variable but the tooth thickness (measured on the rolling
taper generator) must increase at the extremities of the tool. Using this principle,
the tooth thickness on the rolling tape generator increases in both directions
starting from the point N (Fig. 2).
 On Some Pecularities of Paloid Bevel Gear Worm-Hobs 161

Figure 2: The tooth thickness distribution along the cone generator.

The thickness of teeth in case of gears manufactured using the worm-hob
described above results smaller at the tooth extremities and larger in the middle
of it. This localizes the contact pattern in the middle of the tooth-flank when
gearing. This modification causes the deformation of the tooth line on the plain
gear too, transforming the theoretical involute into a more general shaped curve
named paloid.
Both the threading and the relieving operation are realized on a relieving
machine, where the axis of the worm-hob is declined with a 30° angle reported
to the axis of the chuck, in a horizontal plain containing the axis. This way the
taper generator becomes parallel with the direction of the longitudinal slider.
The curved generator of the worm-hob is realized through leading the cutter
slider by a profiled turning-template. The same template is used by the
threading and relieving operations.
Bevel worm hobs present one thread, and allow the cutting of any teeth
number by a fixed module and pressure angle. The pitch is normalized along the
rolling cone generator and can be calculated using the formula: pN = π ⋅ mN .
The axial pitch (defined on the worm hob’s axis) is determined by
p A = π ⋅ mN ⋅ cos 30 .
162 D. Hollanda, M. Máté

2. The geometry of the tooth thickness modification

Considering the bevel worm-hob as a helical bevel surface, like any bevel
thread, it is to mention that next to the axial pitch, a radial pitch can be defined,
depending on the axial pitch and the half taper angle of the rolling cone (30o). If
the generator of the worm hob were a straight line, the radial pitch value would
satisfy the expression h = p A ⋅ tg30o. However, the generator is curved and the
radial pitch is calculated taking into account the fact that the endpoint of the
characteristic radius must be on that generator. Considering the tool’s pressure
angle α , the distance between the curved and the straight generators (based on
the dimensions of the tool profile indicated in [1], [2], [6]) can be calculated
using the formula
∆1 = ( pN + 0,01 ⋅ mN − pN ) / 2 ⋅ tgα = 0,01 ⋅ mN / 2 ⋅ tgα
applicable for point A (Fig 3). The same distance is indicated related to point C
(Fig 3). In point D, the distance discussed above will increase to double as
follows from the formula ∆2 = ( p N + 0,02 ⋅ mN − p N ) / 2 ⋅ tgα = 2∆1 .
The curved generator is usually an arc of ellipse of which major axis
(superposed to the Ox) is parallel to the rolling cone generator, and it’s minor
axis (superposed to Oy) passes through point B (Fig 3), situated at a distance of
Feo + 2 p N from the top of the rolling cone.

Figure 3: The elliptically curved generator.

 On Some Pecularities of Paloid Bevel Gear Worm-Hobs 163

It is to remark that the ellipse passes through the following points:

A(− p N , a − ∆ 1) ; B (0, a ) ; C (−2. p N , ∆ 1) and D(−4. p N , a − ∆ 2 ) .
The canonic equation of an arbitrarily shaped ellipse, reported to its axes, is:
x2 y2
+ −1 = 0 (1)
a2 b2
Writing the equation of the ellipses passing through the points A, B, C and D
a system of equation will result, of which solution leads to the reference radius
endpoint function e.g. h = f ( p N / cos 30 o ) . In this paper the radial pitch is
accepted having a general form as expressed by the above formula. As a
consequence, the relations deduced in the following can be used for other forms
of the curved generator too. Using this form, the essential aspects of the
dependences will not be influenced by the very sophisticated expressions of the
radial pitch [3].
The matrix equation of the bevel worm hob’s edge, reported to the self
coordinate system, is:

 π ⋅ mN 
 Feo −  [cos 30° + tg30° sin(30° − α )] + λ sin(30° − α )]
 4 
rM = λ cos α (2)
0
1

This edge is fixed to the mobile coordinate system OM X M YM Z M

superposed at the initial moment with the fixed coordinate system OXYZ . The
parametric form of the edge expressed in the mobile system will is denoted by
rM but it keeps the form of rt given by (2).
The surface of the generator flank of the bevel worm hob can be obtained
considering the followings: the edge is fixed to the mobile coordinate system
(Fig 4) that executes a helical motion reported to the stationary system. The
origin moves in the direction of axis with an amount of p A ⋅ ϕ , simultaneously
with a rotation by angle ϕ of the worm-hob (representing the workpiece in the
above case).
164 D. Hollanda, M. Máté

Figure 4: The position of the right edge related to the used reference system.
First the edge must be reported to an auxiliary coordinate system. Between
this and the auxiliary system OM X M YM Z M there exist only translations by
amounts of p A ⋅ ϕ / 2π and h ⋅ ϕ / 2π respectively. A transfer matrix describing
the above translations is:
pA
1 0 0 ϕ
2π
h
M aM = 0 1 0 ϕ (3)
2π
0 0 1 0
0 0 0 1

After that the edge’s equation is transformed to the stationary system,

mentioning that auxiliary system executes only a rotation around its X axis,
characterized by the matrix
1 0 0 0
0 cos ϕ − sin ϕ 0
M Oa = (4)
0 sin ϕ cos ϕ 0
0 0 0 1
 On Some Pecularities of Paloid Bevel Gear Worm-Hobs 165

The pointing vector of the bevel helical surface described by the edge,
reported to the stationary reference system is obtained from the following
matrix equation
r = M Oa ⋅ M aM . ⋅ rM = M OM ⋅ rM (5)
Multiplying M Oa by M aM it results
pA
1 0 0 ϕ
2π
h
0 cos ϕ − sin ϕ
ϕ . cos ϕ
M OM = 2π (6)
h
0 sin ϕ cos ϕ ϕ .sin ϕ
2π
0 0 0 1
Finally, the matrix expression of the pointing vector is:
pA
1 0 0 ϕ
X 2π XM
h
Y 0 cos ϕ − sin ϕ ϕ . cos ϕ YM
= 2π . (7)
Z h ZM
0 sin ϕ cos ϕ ϕ .sin ϕ
1 2π 1
0 0 0 1
Expression (7) realizes the matrix form of the right flank of the bevel worm-
hob tooth, reported to the stationary system OXYZ, attached to the hob.
Analytical expression is obtained after multiplying the matrices in the
equation before. Similarly results the equations of the opposite flank, if starting
the calculus with the equations of the opposite edge.

References
[1] Krumme, W., “Klingelnberg-Spiralkegelräder Dritte neubearbeitete Auflage”, Springer
Verlag, Berlin, 1967.
[2] *** “MASINOSTROIENIE”- ENCYCLOPEDY, Vol.VII., Masghiz Moskou.
[3] Máté, M., Hollanda, D. “The Enveloping Surfaces of the Paloid Mill Cutter”, in Proc. 18th
International Conference on Mechanical Engineering, Baia Mare, 23-25 April 2010, pp.
291-294.
[4] Michalski, J., Skoczylas, L. “Modeling the tooth flanks of hobbed gears in the CAD
environment", The International Journal of Advanced Manufacturing Technology, vol. 36,
no. 7-8, 2008.
[5] Shu-han Chen, Hong –zhi Jan, Xing-zu Ming, “Analysis and modeling of error of spiral
bevel gear grinder based on multi-body system theory", Journal of Central South
University of Technology, vol. 15, no. 5, 2008.
[6] Klingelnberg, J. “Kegelräder- Grundlagen, Anwendungen”, Springer Verlag, 2008.
 Acta Universitatis Sapientiae
Electrical and Mechanical Engineering, 2 (2010) 166-176

Kinematic Analysis of a 6 DOF 3-PRRS

Parallel Manipulator
Zoltán FORGÓ
Department of Mechanical Engineering, Faculty of Technical and Human Sciences,
Sapientia University, Tg. Mureş, e-mail: zforgo@ms.sapientia.ro

Manuscript received October 14, 2010; revised November 08, 2010.

Abstract: The number of parallel mechanism applications in the industry is growing

and the interest of the academia to find new solutions and applications to implement
such mechanisms is present all over the world. In this paper, after a summarised group
theory presentation, a symmetrical six degrees of freedom mechanism (3-PRRS) will be
defined using this theory. Enumerating some possible kinematic chains for Schoenflies-
motion, one solution is kept in order to build up the proposed mechanism. The easy way
of mathematical modelling is given by the fact that the mechanism can be considered as
an extended well known planar Delta manipulator. The double driven joints in each
limb ensure the third translation and other two rotations of the moving platform
complementing the planar motion of the Delta manipulator. After the kinematical
modelling of the presented mechanism, the actuation of the links is considered. A new
parallel driven actuation system is presented in order to fulfill the rotation and
translation movements required for the PRRS limb actuation. The aspects of singular
configurations, which are similar to the planar Delta mechanism singular configurations
with some extensions, are considered also in the presented paper. The paper closes by
enumerating some major advantages of the proposed 6 degrees of freedom manipulator.

Keywords: parallel mechanism, kinematics, group theory, Lie algebra.

1. Introduction
The number of applications in the industry which use parallel mechanisms
are growing and the interest of the academia to find new solutions and
applications to implement such mechanisms is present all over the world. The
lower degree of freedom mechanisms which are suited for some specific tasks
 Z. Forgó 167

are preferred because of the architecture simplicity and therefore the easy
mathematical modeling and finally, but not at least for economical reasons.
The 6 degrees of freedom (DOF) parallel mechanism is introduced by
Steward and Gough [1] and since then many aspects of the mechanism and its
application are revealed. During the last decades more attention has been paid to
the study of 6 DOF parallel mechanisms, including synthesis and analysis on
kinematics, dynamics, singularities, error and workspace. Some milestones in
the analysis of those mechanisms are set by Earl and Rooney using a method for
synthesis of new kinematic structures [2], Hunt studied the manipulators on the
basis of screw theory [3], Tsai is using systematic methodology in [4] and
Hervé discussed the structural synthesis of parallel robots using the
mathematical group theory [5]. More recently Shen proposed a systematic type
synthesis methodology for 6 DOF kinematic structures enumerating 29 parallel
structures [6]. Hereby Shen defines the hybrid single-open chains (HSOC)
which are able to generate three translations and three rotation angles. Using
those HSOCs four 6 DOF manipulators are presented with symmetrical
arrangement of the limbs (see No.3-No.6 architectures, Table 2. from [6]).
According to Tsai [7], the symmetry implies the use of the same number of
actuators on the same positions in each limb. Moreover he says that a parallel
manipulator is symmetrical if it satisfies the condition that the number of limbs
is equal to the number of degrees of freedom of the moving platform. In the
case of double actuated limbs (with two actuated joints) the last presented
condition can be omitted. So the HSOCs defined by Shen can be replaced by
serial chains which enable three translations and three rotations also.
This paper presents some kinematic structures according to the above
mentioned criteria without the aim of full discussion about all the possible
structures. The geometrical model of one architecture is presented as well.

2. General motion generators

The enumeration of serial topology limbs which enable the spatial motion
(three translations and three rotations) is greatly simplified by using the Lie
group of rigid body displacement as introduced by Hervé [8]. If each limb of a
parallel manipulator generates a subset of possible displacements, which is a Lie
subgroup, the intersection set is also a Lie subgroup of the mobile platform.
According to this statement if the platform undergoes the spatial, general
motion, each limb must ensure the three translations and three rotations.
According to Table 1 {D} denotes the general rigid body motions for the 6 DOF
mobile platform and {Li} denotes the displacement Lie subgroup of the ith limb.
The relation between them is given by:
 168 Kinematic Analysis of a 6 DOF 3-PRRS Parallel Manipulator

{L1} ∩ {L2 } ∩ {L3} =

{D} . (1)
It is obvious that the only possibility for a true equation (1) is:
{=
L1} {=
L2 } {=
L3} {D} . (2)
To obtain simple mechanical structures, better symmetry and good
manufacturing for the three limbs the same architecture is considered. For this
reason, the analysis of the {Li} displacement Lie subgroup is carried out. The
notations for displacement Lie subgroups are recalled in Table 1 [9].
According to the group theory it can be stated:

=
{Li } {=
D} {T }{S ( N=
)}
(3)
{T (u)}{T (v )}{T ( w )}{R ( N , u)}{R ( N , v )}{R ( N , w )} ∀N .

Table 1: List of displacements Lie subgroups [4].

Lie subgroup Description of the subgroup

{E} identity
{T(u)} translations parallel to the u vector
{R(N,u)} rotations around the axis determined by N and u
{H(N,u,p)} helical motions with axis (N,u) and the pitch p
{T(Pl)} translations parallel to the Pl plane
{C(N,u)} cylindrical motions along an axis (N,u)
{T} spatial translations
{G(u)} planar gliding motions perpendicular to u
{S(N)} spherical motions about point S
{X(u)} Schoenflies motions
{D} general rigid body motions or displacements

Based on [9] a planar joint has 5 equivalencies as presented below:

{G (u)} = {R ( A, u)}{R( B, u)}{R(C , u)}; (4)
={G (u)} {R ( A, u)}{T (v )}{R(C , u)} v ⊥ u; (5)
={G (u)} {T (v )}{R ( B, u)}{R (C , u)} v ⊥ u; (6)
={G (u)} {T (v )}{R ( B, u)}{T ( w )} v,w ⊥ u; (7)
={G (u)} {T (v )}{T ( w )}{R (C , u)} v,w ⊥ u. (8)
 Z. Forgó 169

Considering equations (4) and (8) respectively equality N ≡ C the equation

(3) becomes:
{Li } {T (u)}{R ( A, u)}{R (C , u)}{R( B, u)}{R( B, v )}{R ( B, w )} ∀A, B, C ; (9)
{Li } {T (u)}{R ( A, u)}{R (C , u)}{S ( B)} ∀A, B, C. (10)
The above defined {Li} displacement Lie group variants are presented in Fig. 1.

Figure 1: The {Li} displacement Lie group variants incorporating the X-motion
generator.
The X-motion (or Shoenflies motion) generator can be easily observed, due
to equation (9) and Fig. 1a. Considering primitive Schoenflies-motion
generators [10] equivalences can be applied. Extending those generator family
members with the universal joint as seen in Fig. 1, new generators for {D}
displacement Lie group can be introduced. However, this enumeration is out of
the topic of this paper. Because of the reduced link number and simplicity, in
further investigation, the Fig. 1b variant is preferred. Using other geometrical
constraints the architecture is presented in [11] also. The schematic design of
such a limb for a 6 DOF manipulator is presented in Fig. 2b. The index i is
introduced because the same type of limbs are used for moving the manipulator
platform.

3. Kinematics of 3-PRRS mechanism

The general setup for the parallel mechanism having three translations and
three rotations for the end effector (denoted by P) is presented in Fig. 2c. For
simplicity the mechanism is presented from top view. The geometrical
parameters used in the mathematical modelling are enumerated in sketches b)
and c) from Fig. 2. Further the real number values xN, yN and zN are introduced
as the coordinates of a point N in the Cartesian space 0x0y0z0.
Using the equivalency between sketches a) and b) from Fig. 2 it can be stated:
170 Kinematic Analysis of a 6 DOF 3-PRRS Parallel Manipulator

Ci =
Bi C i Di + Di = Bi Ci Di x ⋅ i + Ci Di y ⋅ j + Di Bi ⋅ k , (11)
C i=
Bi Ci Bi x ⋅ i + Ci Bi y ⋅ j + Di Bi ⋅ k , (12)

where i, j and k are the unit vectors of the x0, y0 and z0 Cartesian axes. The setup
of the mechanism (based on the projection of the manipulator on the 0x0y0 plan
– top view from Fig. 2) suggests a planar Delta manipulator.

Figure 2: Schematic design of ith limb of the 6 DOF manipulator (a, b), and the top view
of the proposed mechanism (c). The shaded couplings are the active joints (one
prismatic and one rotation for each limb), and the white ones are passive bonds.
For this reason the mathematical modelling of the proposed mechanism is
made easily and it is like a well known planar Delta manipulator modelling [7]
with some completions. These completions must be made due to the fact, that it
 Z. Forgó 171

is possible to rotate the platform around the x0 and y0 axes too, and so the
projections of the BiBi+1 platform length are variable. Through these paragraphs
the inverse and direct kinematics of the proposed mechanism is defined, and
issues about singular configurations are presented as well.
At the beginning the closure equation is considered for the three limbs:
OAi + Ai C i + C i Bi =OP + PBi where i =1, 2,3 . (13)
In case of inverse kinematic modelling the right side of equation (13) is
given through the coordinates of the characteristic point (denoted by P) and
through the three rotation angles around the axes of the fixed 0x0y0z0 system:
= X [ xP yP z P α β γ ]T . The task is to determine the robot parameters
q = [q1 q2 q3 q4 q5 q6 ]T from the left side of the equation. Assuming that
vector a has the components axy parallel to the 0x0y0 plan and az parallel to the
z0 axis, equation (13) becomes:

OAixy + Ai C ixy + C i Bixy =OPxy + PBixy

OA + A C + C B =OP + PB where i = 1, 2,3 . (14)
 iz i iz i iz z iz

In order to determine the qi translational parameters (i=1,2,3), introduced in

Fig. 2b, the second equation from (14) is considered:

qi ⋅ k + C i Biz = z P ⋅ k + PBiz , respectively in scalar form (15)

qi =z P + PBiz + Ci Biz (16)

Hence CiBiz is a constant geometrical parameter of the manipulator, the first

two terms from the right side of equation (16) contain the general parameters
because PB= iz PBiz (α, β, γ ) .
To obtain the qi+3 rotation joints parameters (i=1,2,3) the first equation from
(14) is recalled and presented in scalar form:

OAi cos αi + Ai Ci cos( qi+3 +αi −π) + Ci Bi cos( qi+3 +αi −π−βi=) xP + PBix
 ,(17)
OAi sin αi + Ai Ci sin ( qi+3 +αi −π) + Ci Bi sin ( qi+3 +αi −π−βi=) yP + PBiy

=
where PBix PBix (α, β, γ ) and PB
=iy PBiy (α, β, γ ) respectively i=1,2,3. To
eliminate the βi parameter belonging to a passive joint, the equations are
rearranged, and summing the square of the two equations in (17) yields:
e1i ⋅sin ( qi+3 +αi −π) + e2i ⋅cos( qi+3 +αi −π) + e3i =0 , (18)
 172 Kinematic Analysis of a 6 DOF 3-PRRS Parallel Manipulator

where

(
e1i =−2⋅ Ai Ci ⋅ yP + PBiy − OAi sin αi ; )

e2i =−2⋅ Ai Ci ⋅( xP + PBix − OAi cos αi ); (19)

( )
2
e3i = ( xP + PBix − OAi cos αi )2 + yP + PBiy − OAi sin αi + Ai Ci 2 − Ci Bi 2 .

Solving equation (18) by using the substitutions:
 2ti
sin ( qi+3 +αi −π) = 2
 1+ ti q +α −π
 2
where ti = tan i+3 i , (20)
cos q +α −π =1− ti 2
 ( i+3 i ) 1+ t 2
 i

the qi+3 parameters (i=1,2,3) are given by:

−e1i ± e12i + e22i − e32i

qi +3 = π − αi + 2 tan −1 . (21)
e1i − e2i
Equations (16) and (21) define the robot parameters in case of inverse
kinematics. To obtain the general coordinates of the mechanism it is necessary
to calculate the position of the Bi(xBi,yBi,zBi) joints (i=1,2,3) in Cartesian space
and knowing the geometrical dimensions of the mobile platform the
= X [ xP yP z P α β γ ]T vector is obvious. The xBi, yBi, and zBi values are
defined through the following nine equations:
( x − x )2 + ( y − y )2 = Ci Bi2
 Ci Bi Ci Bi i =1,2,3

( ) ( ) ( )
2 2 2 2
x −
 Bi Bjx + y − y =d − z − z for i +1, if i =
1,2 (22)
z =
Bi Bj Bi Bj j =
z − Di Bi  1 , if i = 3
 Bi Ci

=
where xCi x=
Ci (qi +3 ), yCi y=
Ci (qi +3 ), zCi zCi (qi ) respectively Ci Bi and Di Bi
are constant geometrical dimensions using i = 1,2,3. It is important to mention,
that in present case the forward kinematics deals with only 8 solutions (as by
the planar Delta robot).
To complete the kinematic calculations the relation between the actuated
joints and the platform velocities is needed and obtained through:
J x ⋅ X = J q ⋅q , (23)
 Z. Forgó 173

where the matrices:

b1x b1 y b1z e1 y b1z − e1z b1 y e1z b1x − e1x b1z e1x b1 y − e1 y b1x 
=J x b2 x b2 y b2 z e2 y b2 z − e2 z b2 y e2 z b2 x − e2 x b2 z e2 x b2 y − e2 y b2 x  , (24)
 
b3 x b3 y b3 z e3 y b3 z − e3 z b3 y e3 z b3 x − e3 x b3 z e3 x b3 y − e3 y b3 x 

b1z 0 0 a1x b1 y − a1 y b1x 0 0 

J q  0 b2 z 0 0 a2 x b2 y − a2 y b2 x 0 , (25)
 
 0 0 b3 z 0 0 a3 x b3 y − a3 y b3 x 

can be written using the notations a = Ai C i , b = C i Bi and e = PBi . Equation (23)

can be considered for calculation of direct and inverse kinematics.

4. Parallel drive actuation of a manipulator limb

To assure the parallel mechanism concept for the 3-PRRS manipulator the
parallel drive of the three limbs must be realized. Therefore a toothed belt drive
H-shaped system can be applied as it can be seen in Fig. 3. At the bottom of the
mechanism the actuated pulleys (gray color fill) induce the motion in the
mechanism by the qiM and qiM+3 driving parameters. The values qi and qi+3 are
set as the output parameters.
Using the parallel drive system two rotation inputs are transformed into
translation and rotation output. The relation between them is given by the
following equation:

 r r 
−
 qi   2 2   qi  .
M
=q   r ⋅
   (26)
 i +3   − r   qiM+3 
−
 2 R 2 R 
The inverse geometry calculus can be performed using the following
equation:

 1 R
−
q   r M
r   qi  .
=
  
i
⋅
R   qi +3 
M
(27)
q  − 1i +3
−
 r r 
Using the above formulation and considering equations (16) and (21) the
inverse geometry is obtained in the following form:
174 Kinematic Analysis of a 6 DOF 3-PRRS Parallel Manipulator

 1 R 
 r 0 0 − 0 0 
r
 1 R 
 q1   0
M 0 0 − 0  q 
r r
 q2M   R   q2 
1
1
q M   0 0 0 0 −   
q M
=
 3M  =  r r  ⋅  q3  =⋅
A q . (28)
 q4M   − 1 0 0 −
R
0 0   4
q
 q5   r r   q5 
 q6M   0 −
1
0 0 −
R 
0   q6 
 r r
 1 R
 0 0 − 0 0 − 
 r r
Due to the characteristic setup of the driving mechanism the equations for
the kinematics are obtained in similar way:
q A−1 ⋅ q M .
q M= A ⋅ q and = (29)
In accordance with the formulated equations the dynamics of the manipulator
can be calculated easily, and will be presented in a further paper.

z0 z0

q i+3
Ei Ei Ci
R Ai
q i+3
E’i

qi qi

q iM M
q i+3
O y0 O x0

Figure 3: Schematic design of the belt mechanism for one, double drive link.
 Z. Forgó 175

5. Singular configurations
The singularity analysis of this mechanism can be done based on the
matrices from (24) and (25). Inverse kinematic singularities occure in case of
aix biy − aiy bix =
0 (i=1,2,3) which defines the workspace boundaries. An other
possibility is biz = 0 (i=1,2,3) but it can be avoided through geometrical design,
because it is a constant value. Direct kinematic singularities occure when at the
same time it can be stated that bix = 0 or biy = 0 (i=1,2,3), which means that the
CiBi links are parallel. The same type of singularities can be found for
coexistence of eix biy − eiy bix =
0 (i=1,2,3) in case of coliniar C i Bi and
PBi vectors. Both direct kinematic singularity cases can be avoided by careful
geometrical design. The implemented parallel drive mechanisms have no
singular configurations, and this kind of calculations can be omitted.

6. Conclusions
This paper deals with a 6 degrees of freedom manipulator architecture using
the group theory. The mobile platform is connected to the base through three
PRRS limbs, each being double actuated on the first and second joint levels.
The inverse geometrical calculations are performed through equations (16) and
(21), hence the direct modelling is presented through the equation system (22).
The relation between the robot and general velocities is stated by the equation
(23). Some aspects about the singular configurations are introduced in the paper
based on the equation mentioned before. As it can be seen in the figures
presented in this paper the architecture is the extension of the well known planar
Delta robot to a 6 DOF mechanism. The mathematical model of the spatial
manipulator reflects this fact very well. The simple setup of the presented
mechanism assures a good manufacturability and needs a relatively easy control
algorithm considering some other 6 DOF manipulators.

References
[1] Stewart, D. A., “Platform with six degrees of freedom”, in Proceedings on Institution of
Mechanical Engineering, 1965, vol. 180, pp. 371-386.
[2] Dasguta, B. and Mruthyunjaya, T. S., “The Stewart platform manipulator: a review”
Mechanism and Machine Theory, vol. 35, pp. 15-40, 2000.
[3] Hunt, K. H., “Structural kinematics of in-parallel-actuated robot arms”, ASME Journal of
Mechanical Design, vol. 105, pp. 705-712, 1983.
[4] Tsai, L. W., and Joshi, S., “Kinematics and optimization of a Spatial 3-UPU Parallel
manipulator”, ASME Journal of Mechanical Design, vol. 122, pp. 439-446, 2000.
176 Kinematic Analysis of a 6 DOF 3-PRRS Parallel Manipulator

[5] Hervé, J. M., “Design of parallel manipulators via the displacement group”, in Proceedings
of the 9th World Congress on Theory of Machine and Mechanisms, Milano, 1985, pp. 2079-
2082.
[6] Shen, H., Yang., T., and Ma, L., “Synthesis and structure analysis of kinematic structures of
6-DOF parallel robotic mechanisms”, Mechanism and Machine Theory, vol. 40, pp. 1164-
1180, 2005.
[7] Tsai, L. W., “Robot Analysis – The Mechanics of Serial and Parallel Manipulators”, John
Wiley & Sons, 1999.
[8] Hervé, J. M., “The Lie group of rigid body displacements, a fundamental tool for
mechanism design”, Mechanism and Machine Theory, vol. 34, pp. 719-730, 1999.
[9] Hervé, J. M., “The planar-spherical kinematic bond: Implementation in parallel mecha-
nisms”, http://www.parallemic.org/Reviews/Review013p.html.
[10] Lee, C. C. and Hervé, J. M., “Type synthesis of primitive Schoenflies-motion generators”,
Mechanism and Machine Theory, vol. 44, pp. 1980-1997, 2009.
[11] Olea, G., Plitea, N., and Takamusa, K., “Kinematical analysis and simulation of a new
parallel mechanism for robotics application”, in ARK Piran, Piran, 25-29 June 2000, pp.
403-410.
 ACKNOWLEDGEMENT

The Editors would like to acknowledge the contributions of all who

coordinated and performed the double-blind peer review of the
manuscripts submitted to the journal. Besides the members of the
Editorial Board, the following researchers made significant efforts to
complete this process:

László BAKÓ
Sándor Tihamér BRASSAI
József DOMOKOS
Katalin GYÖRGY
Piroska HALLER
Tünde JÁNOSI-RANCZ
Lajos KENÉZ
Nimród KUTASI
László Ferenc MÁRTON
Márton MÁTÉ
István PAPP
Sándor PAPP
László SZABÓ
László SZILÁGYI
Tamás VAJDA

177
 Acta Universitatis Sapientiae
The scientific journal of Sapientia University publishes original papers and surveys
in several areas of sciences written in English.
Information about each series can be found at
http://www.acta.sapientia.ro.

Editor-in-Chief
Antal BEGE
abege@ms.sapientia.ro
Main Editorial Board

Zoltán A. BIRÓ Zoltán KÁSA András KELEMEN

Ágnes PETHŐ Emőd VERESS

Acta Universitatis Sapientiae

Electrical and Mechanical Engineering
Executive Editor
András KELEMEN (Sapientia University, Romania)
kandras@ms.sapientia.ro
Editorial Board
Tihamér ÁDÁM (University of Miskolc, Hungary)
Vencel CSIBI (Technical University of Cluj-Napoca, Romania)
Dénes FODOR (University of Pannonia, Hungary)
Dionisie HOLLANDA (Sapientia University, Romania)
Maria IMECS (Technical University of Cluj-Napoca, Romania)
Zsolt LACZIK (University of Oxford, United Kingdom)
Géza NÉMETH (Budapest University of Technology and Economics, Hungary)
Ştefan PREITL (”Politehnica” University of Timişoara, Romania)
Gheorghe SEBESTYÉN (Technical University of Cluj-Napoca, Romania)
Iuliu SZÉKELY (Sapientia University, Romania)
Imre TIMÁR (University of Pannonia, Hungary)
Mircea Florin VAIDA (Technical University of Cluj-Napoca, Romania)
József VÁSÁRHELYI (University of Miskolc, Hungary)

Sapientia University Scientia Publishing House

ISSN 2065-5916
http://www.acta.sapientia.ro
 Information for authors
Acta Universitatis Sapientiae, Electrical and Mechanical Engineering
publishes only original papers and surveys in various fields of Electrical and Me-
chanical Engineering. All papers are peer-reviewed.

Papers published in current and previous volumes can be found in Portable Document
Format (PDF) form at the address: http://www.acta.sapientia.ro.

The submitted papers must not be considered to be published by other journals.

The corresponding author is responsible to obtain the permission for publication of
co-authors and of the authorities of institutes, if needed. The Editorial Board is dis-
claiming any responsibility.

The paper must be submitted both in MSWord document and PDF format. The
submitted PDF document is used as reference. The camera-ready journal is prepared
in PDF format by the editors. In order to reduce subsequent changes of aspect to
minimum, an accurate formatting is required. The paper should be prepared on A4
paper (210 × 297 mm) and it must contain an abstract of 200-250 words.

The language of the journal is English. The paper must be prepared in single-column
format, not exceeding 12 pages including figures, tables and references.

The template file from http://www.acta.sapientia.ro/acta-emeng/emeng-main.htm

may be used for details.

Submission must be made only by e-mail (acta-emeng@acta.sapientia.ro).

One issue is offered to each author free of charge. No reprints are available.

Contact address and subscription:

Acta Universitatis Sapientiae, Electrical and Mechanical Engineering
RO 400112 Cluj-Napoca
Str. Matei Corvin nr. 4.
E-mail: acta-emeng@acta.sapientia.ro

Publication supported by

Printed by Gloria Printing House

Director: Péter Nagy