See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/220141406

Adding sense to the Internet of Things - An

architecture framework for Smart Object
systems.

Article in Personal and Ubiquitous Computing January 2012

Source: DBLP

CITATIONS READS

46 1,057

4 authors, including:

Damith Ranasinghe Mark Harrison

University of Adelaide Milecastle Media Limited
110 PUBLICATIONS 910 CITATIONS 43 PUBLICATIONS 1,018 CITATIONS

SEE PROFILE SEE PROFILE

Duncan Mcfarlane
University of Cambridge
210 PUBLICATIONS 6,586 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Supply chain analytics View project

Next generation information systems for logistics View project

All content following this page was uploaded by Damith Ranasinghe on 23 October 2014.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Pers Ubiquit Comput (2012) 16:291308
DOI 10.1007/s00779-011-0399-8

ORIGINAL ARTICLE

Adding sense to the Internet of Things

An architecture framework for Smart Object systems

Tomas Sanchez Lopez Damith C. Ranasinghe

Mark Harrison Duncan McFarlane

Received: 15 October 2010 / Accepted: 4 April 2011 / Published online: 3 June 2011
 Springer-Verlag London Limited 2011

Abstract The Internet of Things (IoT) concept is being industry standards in metrics such as network throughput,
widely presented as the next revolution toward massively delivery ratio, or routing distance. Finally, we demonstrate
distributed information, where any real-world object can the feasibility and flexibility of the architecture by detailing
automatically participate in the Internet and thus be globally an implementation using Wireless Sensor Networks and
discovered and queried. Despite the consensus on the great Web Services, and describe a prototype for the real-time
potential of the concept and the significant progress in a monitoring of goods flowing through a supply chain.
number of enabling technologies, there is a general lack of
an integrated vision on how to realize it. This paper Keywords Smart Objects  Sensors  RFID 
examines the technologies that will be fundamental for Internet of things
realizing the IoT and proposes an architecture that inte-
grates them into a single platform. The architecture intro-
duces the use of the Smart Object framework to encapsulate 1 Introduction
radio-frequency identification (RFID), sensor technologies,
embedded object logic, object ad-hoc networking, and The Internet of Things is a concept that encompasses a
Internet-based information infrastructure. We evaluate the variety of technologies and research areas that aim to
architecture against a number of energy-based performance extend the existing Internet to real-world objects [1].
measures, and also show that it outperforms existing Advances in fields such as automatic identification, wire-
less communications, integrated sensing, or distributed
data processing have narrowed the gap between the notion
T. Sanchez Lopez (&)  M. Harrison  D. McFarlane of ubiquitous computing set 15 years ago [2] and a world
Auto-ID Lab, Engineering Department,
of networked, sensing, and intelligent things. The
University of Cambridge, 17 Chales Babbage Road,
Cambridge CB3 0FS, UK potential benefits from the IoT realization are many, both
e-mail: tomas.sanchez@autoidlabs.org.uk for individuals and businesses. Some of the most promising
M. Harrison applications include: improved management of global
e-mail: mark.harrison@cantab.net supply chain logistics, product counterfeit detection, man-
D. McFarlane ufacturing automation, smart homes and appliances,
e-mail: dcm@eng.cam.ac.uk e-government (electronic official documents and currency),
improved integrated vehicle health management, and
T. Sanchez Lopez
e-health (patient monitoring and patient records).
Innovation Works, EADS UK, Quadrant House,
Celtic Springs, Coedkernew, Newport NP10 8FZ, SW, UK Automatic identification technologies such as Radio
e-mail: tomas.sanchezlopez@eads.com Frequency Identification (RFID) are fundamental to the
realization of the IoT because they enable things to be
D. C. Ranasinghe
linked with their virtual identity on the Internet. RFID tags
Auto-ID Lab, The Schoold of Computer Science,
University of Adelaide, Adelaide, SA 5005, Australia attached to objects expose unique identification numbers
e-mail: damith@cs.adelaide.edu.au that can be read wirelessly by interrogating devices and

123
292 Pers Ubiquit Comput (2012) 16:291308

used to obtain information related to individual instances of networks. This approach enables us to integrate the most
objects managed by networked back-end systems. Minia- significant functionalities described earlier to realize the
turised sensors now provide the ability to monitor the capabilities of the IoT vision: unique and automatic iden-
condition of objects and consequently make it possible to tification, the sensed condition of objects, embedded pro-
dynamically act upon changes to the status of objects such cessing for local intelligence and autonomy, object-to-
as that derived from their temperature, humidity, and object networking, and an Internet-based information
chemical composition. Furthermore, historical records infrastructure. The proposed framework introduces the use
including both identification and sensor data can be used of the Smart Objects (SO) conceptual model as a corner-
off-line to trace the evolution of the objects location and stone; Smart Objects model the capabilities of objects
status throughout their life. Low-power radio communi- participating in the IoT, from the basic capacity to provide
cation technologies and the availability of increasingly a unique identification number to more complex abilities
powerful low-cost embedded processing increase the such as the capacity to perform ad-hoc networking and
autonomy of objects by providing them with networking object-centric complex decision-making. The SO frame-
capabilities and local intelligence. Finally, distributed work is completed with an information infrastructure that
information infrastructures using Internet protocols for leverages the capabilities of Smart Objects to provide
communication serve as connection hubs for all the services to end-users such as identification and condition
things, together with other resources such as databases, information through a well-defined set of interfaces. Our
data mining tools, and computer networks. objective in this paper is to demonstrate both the usefulness
Both the vastly different and increasing number of and the capability of this framework to address many of the
technologies involved in the IoT concept suggest that its challenges and applications of the Internet of Things,
success must inevitably require the effective integration of through its ability to integrate a heterogeneous set of
various technologies. Recent efforts in the standardization devices and the functionalities of various technologies.
of RFID technologies have produced a networked infra- The rest of this paper is organized as follows. Section 2
structure where the identities coming from wireless tags introduces related technology and architectural work.
attached to real-world objects can be filtered, stored, que- Section 3 represents the core of our contribution by intro-
ried, and linked to on-line object information [3, 4]. ducing the Smart Object framework architecture and its
Standardization bodies such as the IEEE [5] or the Open key design features. Section 4 evaluates the proposed
Geospational Consortium (OGC) [6] have developed sen- architecture and Sect. 5 goes further by detailing a generic
sor standards that may use Internet protocols to access implementation. Section 6 presents an example scenario
sensor descriptions and values. EU-funded projects such as built on top of the implementation, and Sect. 7 lists the
PROMISE [7] or SENSEI [8] describe the use of tags with most important barriers to the adoption of our work. Sec-
embedded processing capabilities in order to equip the tion 8 finishes the paper by presenting our conclusions and
objects with certain intelligence. Although such works lessons learned.
have produced important advances in technology and
standards that could empower the IoT revolution, none
provides a complete integration framework. Consequently, 2 Technology background and related work
an architecture that is capable of integrating various
functionalities necessary to realize the IoT is still lacking. RFID has become the leading technology for automatic
Many people associate the so-called Internet of Things identification due to its advantages over other technologies
(IOT) with Auto-ID technologies such as barcodes, matrix such as barcodes. An RFID system consists of a tran-
codes, and low-cost passive RFID tags. While passive sponder tag attached to an object, a reader that interrogates
RFID is looking increasingly promising for automating the the tag using the wireless medium and a back-end system
tracking of physical objects, the energy available from to organize the captured data [9]. In this way, RFID tags
power harvested via the antenna of the tag is insufficient containing unique object IDs can be read automatically
for powering sophisticated sensing and logging capabili- without requiring line-of-sight. Passive RFID has tradi-
ties, given their current power requirements. There are tionally provided automatic object identification and the
many situations where it is necessary to monitor not only location at which the objects have been identified. Recent
the location and movement of objects but also their con- developments in active and semi-passive RFID provide
dition. In this paper, we describe and implement an enhanced object identification functionalities with various
architecture for Smart Objects that extends established degrees of autonomy [10]. Rapidly evolving new applica-
concepts for networked RFID with new functionalities that tions, such as food safety and vehicle health management,
can support co-operative Smart Object implementations are based on knowing the condition of objects (e.g., tem-
using complementary technologies, such as wireless sensor perature, stress, strain, shock). Although RFID is an

123
Pers Ubiquit Comput (2012) 16:291308 293

important technology in the realization of a seamless link IEC standards. Other advocates of the use of 6LoWPAN for
between individual physical objects and their digital rep- the realization of the Internet of Things also tend to focus on
resentations, it can not provide the condition information adapting existing Internet protocols for constrained
that the next generation of real-world applications require. embedded devices, but do not concern themselves with the
Sensor technologies have attracted relatively recent building blocks of true object integration [27]. For example,
popularity due to their ability to gather the real-world con- IP addresses are poor for describing object identity, unlike
ditions used by modern computing applications [11]. Sensor RFID-based identification systems such as the Electronic
transducers are normally incorporated as part of computing Product Code (EPC) which are structured to enable manu-
systems in order to assist or complement data collection. In facturers to manage the creation of identifiers for their
this regard, recent research efforts pursue the transmission of products and to help consumers access serial and class-level
sensor data through radio links aiming to facilitate sensor product information from those identifiers [28]. Finally,
deployment. Wireless Sensor Networks (WSN) are a popular efforts such as the Web of Things [29] extend the IP-based
research area in this direction, where miniaturized, energy- Smart Objects idea and propose the use of a RESTful
efficient battery-powered wireless devices serve as a plat- approach [30] to access sensor and actuator information.
form for transmitting the sensor data [12]. A recent trend is to RESTful approaches leverage the way the World Wide Web
incorporate sensors into RFID tags [13, 14]. works by using HTTP standard methods such as GET and
Both RFID and sensor technologies are key enablers of PUT to access and manipulate information from networked
the IoT because they provide the means to identify objects Internet repositories. Although some work is being done to
and to obtain their condition. However, there is a need for a extend the REST architecture to constrained devices, such
common underlying infrastructure that links the physical as that of the Constrained RESTful Environments (CoRE)
objects to the digital domain and helps to manage their working group from the IETF [31], the Web of Things is
information. The definition of these links and how they bring still just an architectural conceptualization and not a design
together all the enabling technologies forms the integration for an integration architecture.
architecture that might potentially realize the IoT concept. In the rest of the paper, we aim to introduce an archi-
Research in such architectures is in general missing, tectural framework that not only supports the most impor-
although there are some noticeable efforts that, either on tant foreseen features of the IoT, but that also provides all
their own or in combination with others, show a great the details for creating efficient and complete implemen-
potential for filling this gap. Table 1 summarizes the most tations. Our proposed architecture does not assume the use
relevant-related work, either on architectures that do inte- of any underlying communication technology and includes
grate RFID and sensors or on standardization efforts that we new features such as context-based ad-hoc network and
consider key toward the realization of the IoT. As Table 1 clustering of objects, which can potentially improve the
highlights, although the technologies for gathering, pro- quality and meaning of the collected data.
cessing and distributing information about objects either
already exist or are well advanced, there is little or no
integration among them, leading to a lack of a comprehen- 3 The Smart Object architecture framework
sive platform for heterogeneous data processing and sharing.
The term Smart Object is often mentioned in the lit- Based on the main shortcomings of the related work in the
erature, together with other similar concepts such as area, we devise an integration framework that incorporates
intelligent products [22] or smart parts [23]. The all the main features involved in the Internet of Things
realization of these concepts is often as varied as the number concept. Our framework will thus be able to identify and
of terms used to describe them, although they seldom link monitor the condition of objects, using existing standards
RFID and sensor technologies, as well as other fundamental where possible. The integration of ID and sensor data in
features of the IoT such as ad-hoc networking or embedded relation to a particular object will be unambiguous. Objects
object logic. A school of thought that is growing in popu- will be capable of establishing networks, and the mem-
larity associates the Internet of Things and the Smart Object bership and structure of these networks will depend on the
terminologies with IP-based protocols, specially with a context of the situation in which the objects are involved.
modified version of IPv6 for low-power embedded devices Objects inside the networks will cooperate to manage their
called 6LoWPAN [24]. The IP for Smart Objects alliance resources and to take complex decisions on data routing,
[25], for example, promotes the use of the 6LoWPAN and inter-network relationships, event generation, and others.
ROLL [26] for networking Smart Objects. However, the Object data will be made available to users of the IoT via
IPSO alliance does not give a comprehensive definition of well-defined interfaces, and this data will be organized in
what Smart Objects are, nor does it provide a design doc- such a way as to allow user access with various degrees of
ument of SOs beyond the reference to IETF, IEEE, or ISO/ granularity. These features are summarized in Fig. 2.

123
294 Pers Ubiquit Comput (2012) 16:291308

Table 1 Summary of RFID and sensor integration related work

Description Main shortcomings Potential improvements

PROMISE [7] EU project using Product Embedded Information Little alignment with Adoption of ID standards. Consider
Devices (PEID) for monitoring ID and condition standards, no networking of sensor (object) networking
of objects during their life cycle PEIDs
OGC SWE [6] Extensive set of protocols and interfaces to share IDs not globally unique.
sensor information in a standard way over the Sensors are not considered to
Internet. monitor objects or products
SENSEI [8] EU project designing an architecture for the Under development. No ID Adoption of ID standards. The
connectivity of global & heterogeneous sensor standardization project needs to reach a mature
and actuator networks via the specification of state before its results can be
open service interfaces evaluated
EPC Network [3] Emerging industrial RFID standards architecture Does not yet handle sensor Extend current standards with sensor
based on unique item identification via the data data
Electronic Product Code (EPC)
BRIDGE [15] EU project for developing new technologies within Work with sensors does not
the EPC Network extend the EPC Network
standards
EPC SN [16] Auto-ID lab project to extend the EPC Network Too complex to allow a full Compromise in developing a simpler
with sensors architecture extension functional part of the extension
ISO 18000-6 [4], Set of standards dealing with the integration of Under development. Standards need to reach a mature
24753 [17], RFID with sensors Cooperation among state before they can be used
IEEE 1451-7 standardization bodies is
[18] complex and slow
GSN [19] Middleware to collect and share information from RFID and WSN data are not Provisions for collecting both RFID
RFID and WSN over the Internet integrated at object level and sensor data from an object
CoBIs [20] EU project defining reusable services which use
WSN and RFID to represent physical entities
involved in business processes
SARIF [21] Middleware for designing applications requiring Integration is only by spatial Consider more integration methods
RFID and WSN information comparison, no mention of and an explicit connection with
Internet scalability Internet protocols

The rest of this section describes the design of an The previous scenario highlights one of the main drivers
architecture that implements the integration framework for the use of sensor data at the object level. Studies have
features described above. Later sections will demonstrate shown that even in closed 14-m refrigerated containers, the
our key contributions and evaluate the central conceptual variation of temperature across pallets can reach up to 35
components of the architecture framework design. percent [32] and that in chilled delivery vans, the temper-
ature from one package to another can vary up to 4,
3.1 Smart Object definition resulting in the potential growth of bacteria among per-
ishables [33]. Other studies have also shown how the dis-
The integration of object identification and sensor data tribution of cargo in confined spaces can influence the
streams may be realized in multiple ways. For example, it detection of safety-related events such as fires when envi-
is possible to merge ambient sensory data provided by a ronment sensors are used [34].
sensor-rich space to identities of objects entering that space With the new developments on integrated circuitry,
by detecting when objects enter the area. However, this micro-electro-mechanical systems (MEMS), wireless
scenario might prove difficult to implement since detection communications, and embedded technologies in general,
systems must be placed on the boundaries of the space. the vision of an integration that occurs at the hardware
Moreover, architectural logic must be put in place for level is more plausible and logical than ever. As the
merging both independent streams of data (sensor data and amount and type of information that can be embedded into
automatic identification data). Furthermore, sensor data assets increases, we witness an evolution toward object-
could be deemed inappropriate since the transducers could centric systems, where manufactured items take control
be located at a considerable distance from the monitored over the flow of information which was traditionally
objects. retrieved manually by human operators. As a result of the

123
Pers Ubiquit Comput (2012) 16:291308 295

augmented capabilities and intelligence that the object- containers, vehicles), warehouses, retailers facilities, or
centric paradigm supports, this new generation of assets end-user assets may be subject to condition monitoring and
has been called smart products or, more generally, therefore provide valuable information for themselves or
Smart Objects [35]. Our framework for building Smart other objects in their vicinity. In order to monitor their
Objects is based on five fundamental properties: Smart condition, our architecture design makes use of embedded
Objects are those objects that devices with wireless communication capabilities. These
devices would be attached to the objects, becoming a part
possess a unique identity.
of them, the same way a bar-code is integrated into the vast
Are able to sense and store measurements made by
majority of todays products.
sensor transducers associated with them.
The devices attached to the objects are meant to provide
Are able to make their identification, sensor measure-
a global identification, to sense the status of the object and/
ments, and other attributes available to external entities
or their surroundings and to provide a wireless interface for
such as other objects or systems.
data communication. These devices are thus aligned with
Can communicate with other Smart Objects.
the developments in sensor-enabled active tags, with the
Can make decisions about themselves and their inter-
fundamental difference that our devices (which we call
actions with external entities.
simply nodes) hold the logic to not only communicate
The Smart Object properties directly contribute to the with each other but to form ad-hoc networks for the better
realization of the framework features described throughout management of resources and data capture. Our nodes are
this section. Figure 1 provides an overview of the archi- basically WSN devices with a standardized unique identi-
tecture design, and its conceptual components are descri- fication. However, unlike traditional WSN, nodes are
bed in the following subsections. Figure 2 summarizes the expected to move together with the objects they monitor
architecture design components that address each of the and to continuously interact with other nodes. This dyna-
Smart Object framework features and their relationship to mism is not a common characteristic of such restricted
the SO properties. devices and coping with it constitutes a major part of our
work in this area. Algorithms and strategies for dynamic
3.2 Object identification and sensor integration object networking and administration are described in
Sects. 3.3 and 3.4
Objects such as consumer goods, product parts, assembly The use of the same device for capturing both ID and
machinery, logistics and transportation items (e.g., pallets, sensor data provides integration at the hardware level with
no ambiguity. Sensor and ID data are matched the instant
they are captured and travel together in communication
packets and events. Consequently, they are stored and
discovered as one.

3.3 Object networking

RFID and sensor technologies that capture data about

specific objects have traditionally used point-to-point
communication between the device representing the object
(e.g, RFID tag) and the data reader (e.g., RFID reader).
This form of communication is sufficient for simple
applications, but might be insufficient when large numbers
of devices coexist or a certain degree of intelligence on the
data gathering is required.
Smart Object networks utilize ad-hoc networking for
communications and clustering for energy management.
Clustering extends the network lifetime by electing a rep-
resentative network member, or Cluster Head (CH), which
collects all the communication within the network and
forward it to the infrastructure. Cluster heads are beneficial
because they handle the burden of communicating with
external entities, a task that is usually more power-inten-
Fig. 1 Smart Object system architecture sive than intra-network communication. Furthermore, CHs

123
296 Pers Ubiquit Comput (2012) 16:291308

Fig. 2 Mapping between properties, features and architecture components of the Smart Object framework

can provide a centralized control over network function- No global cluster information Nodes do not need to
alities when needed. They may also be elected according to store information about all the members of a cluster.
their particular capabilities, such as radio ranges, and This property is vital for the scalability of the
processing power. On the downside, CHs consume more algorithms.
energy that the other network members and their role must No need for synchronization Cluster nodes do not need
be periodically rotated in order to avoid the premature to be synchronized in order to elect the CH or to
exhaustion of their battery power. operate inside the cluster. Synchronization among
Clustering is a relatively popular technique for balanc- network nodes is time-intensive and energy-consuming.
ing the communication load and the energy expenditure in Mobility support The algorithms support both the
wireless networks. What makes the approach used for movement of nodes within the network as well as
Smart Objects different from existing approaches is the merging and splitting of networks. Section 3.4
combination of a number of techniques aimed at serving describes in more detail how the SO architecture
mobile and independent groups of nodes: supports these features.
Low algorithm complexity The number of nodes fully
Energy-based CH election The residual energy of the
participating in a given CH election is always lower
network nodes (i.e, the amount of energy left in its
than the total number of nodes in the cluster, making
battery to enable its operation) is used to calculate the
our approach less complex than any other clustering
best candidate when the CH role is rotated.
algorithm.
Variable CH period Residual energy is also used to
calculate the amount of time that a node will have the The first two items on the list outline the algorithm for
CH role until the next rotation occurs. This technique selecting new cluster heads. When a node ends its CH
effectively controls the energy expenditure of CHs period, it sends an Election message to announce a new
thereby maximizing the number of alive nodes. election procedure and request other nodes to bid for
Double clustering Physically large Smart Objects (e.g., becoming a new CH. This message contains a computed
containers, logistic vehicles) might be equipped with time, which is the proposed period for staying as a CH.
more than one node to provide accurate status across This proposal is based on the current residual energy of the
the physical space. In the same way that a network of CH node, so that the greater the residual energy the greater
Smart Objects elects a CH (NetCH) to balance the the proposed time period. When a cluster member receives
communication burden of the SO network with other an Election message, it sends back an Election
infrastructure components, a Smart Object with various Response message only if its own computed time period
nodes elects one of the nodes as a Smart Object CH is higher than the CH proposal. After waiting for a fixed
(SoCH) to balance the load of communicating with period of time for counterproposals from the rest of the
other Smart Objects. As we will show in Sect. 4, this cluster members, the CH compares the time periods of all
double clustering mechanism provides benefits in terms the received responses and sends an Election Set
of energy consumption distribution and network message to all the cluster members setting the node with
lifetime. the highest proposed period as the new CH. The new CH

123
Pers Ubiquit Comput (2012) 16:291308 297

will remain in its role for the timer period stated in its The addressing and routing mechanism developed for
proposal. the SO architecture, called Sequence Chain (SC) [36], have
Networking of SO also differs from other wireless net- been integrated into the algorithms that handle the forma-
working approaches in that it is additionally employed as a tion of clusters and are therefore within the processes that
filter to data capture. Rather than forming networks with manage the splitting and merging of networks. Sequence
random objects, SO may only interact with one another if Chain assigns addresses following a hierarchical structure
they share a common context (e.g., participate in the same and therefore the routing of packets between network
shipment, are parts of the same composite object, are stored nodes follows the resulting tree (i.e, tree routing).
in the same place). To achieve this, a number of SO Sequence Chain has the following main properties:
attributes and algorithms for classifying and prioritizing
Unique and reusable addresses Addresses are unique
SO interactions are also included within the logic of the
and thus require no duplicate address detection. The
nodes. This processes effectively influences how the clus-
address of a node that leaves the network can be reused
ters are formed, transforming the network clustering into
for nodes joining the network at a later time.
more than a balancing technique, and shaping the network
Network merge and split support A network can
structure into logically connected groups of objects.
recover from the leaving or merging of both single
nodes and networks.
3.4 Network administration
Fully distributed Each address is calculated locally
without the need for any central authority. Network
Smart Objects are mobile objects that may move inde-
recovery after a split or merger is also managed locally
pendently of their network and therefore may cause
by the affected portions of the network.
unexpected network changes such as the departure from
Low overhead and scalable Addresses are not limited
their current network (i.e, network split) or the arrival to a
in size but grow with the addition of new nodes. The
new network (i.e, network merge). Traditional WSN
local decision-making results in low latencies and
research has focused on data routing and resource man-
processing complexity that do not increase with the
agement on static systems. However, the dynamic nature of
number of network nodes. Routing decisions are taken
SO interactions differs from existing WSN research and
based only on address comparison and do not require
requires further considerations to be embedded into the
route discovery.
network protocols. As a part of the research in Smart
Object systems, mechanisms to handle spontaneous addi-
tions and departures of SO to/from the network were 3.5 Event generation and data sharing
developed.
Just as computer networks require each network node to A Smart Object itself is not enough to achieve the expected
have an address in order to route communications between benefits of smart products. There is also a need for
any two computers, SO networks also require each object flexible and efficient ways of managing the Smart Object
to have an address. The address of an object should be information and making it available to end-users. This
different from its ID because object IDs are selected with a functionality is achieved by providing an information
business context in mind (e.g., bar codes, EPCs), while infrastructure with which Smart Objects can communicate.
object addresses are selected depending on the topology of The SO architecture presented here is event driven.
the network (e.g IP addresses). As mentioned in Sect. 2, a Events are generated by changes in the SO network con-
number of initiatives have emerged aiming to introduce IP ditions and are forwarded from their point of origin toward
protocols in the Smart Object arena. Although implemen- the information infrastructure. Two family of events exist:
tation of these protocols such as 6LoWPAN [24] have those that are created to reflect changes of the network
achieved small network stacks suitable for some embedded structure (e.g network split or merge) and those that are
devices, resource constrained hardware implementations used to periodically report sensor data (i.e, Smart Object
(e.g with up to 4 kb of RAM) are still unable to hold both condition). The former set of events are used to create and
the IP stacks and the clustering algorithms presented in this update a virtual network structure on the infrastructure
paper. We believe that cheap (and thus feature-limited) side, which we call the Smart Object network structure
WSN-like devices are the future in the mass production of repository. This repository provides clients with the ability
Smart Objects, and therefore can not require the use of IP to discover the Smart Objects that form part of the same
protocols for our architecture. For this reason, a new network, their particular relations with other Smart Objects
addressing and routing mechanism was developed specially and the sensors and output types supported by each SO.
tailored for the dynamic and distributed nature of the Smart As described in previous sections, the formation of
Object networks. networks is controlled by a mechanism that considers

123
298 Pers Ubiquit Comput (2012) 16:291308

contextual Smart Object relationships. In this way, clients adapt to various implementation technologies and its sca-
accessing the SO network structure repository can deter- lability in terms of supporting increasing loads of Smart
mine which objects form part of the same contextual sit- Object data and concurrent clients.
uation and decide which object or sets of objects would A mathematical energy model of the SO nodes was
produce the sensor data most relevant to them. This design developed and plugged into a purpose-built Smart Object
effectively provides a mechanism whereby a client has network simulator. The reasons for developing a simulator
access to a vast variety of logically related data sources tool were fourfold:
with a very narrow entry point (e.g., by only knowing the
1. To be able to accurately control and simulate the
ID of one of them)
particular attributes of Smart Objects and their net-
Events sent from the Smart Object networks to the
works, which differ in many ways from regular
Information Infrastructure will be received and trans-
Wireless Sensor Networks.
formed into messages understandable by the Information
2. To have a graphical representation of the complex and
Infrastructure components. Traditional WSN use base
distributed SO networks, including each SO real-time
stations to collect wireless sensor data and to connect
status and the messages exchanged between them.
wireless sensor nodes to other computing infrastructures. A
3. To serve as a monitor for our implementation trials,
base station would normally receive wireless messages,
whereby a sniffer node would listen to the messages
process them, and either perform only local computation
exchanged in an implemented SO network and trans-
(e.g display the result of analysing their contents) or send
mit them to the simulator tool. In this monitoring
their information to other components over a network for
mode, the tool would interpret the messages from the
further analysis and processing. Much in the same way, the
sniffer node and draw a graphical representation of the
gateway component of the SO architecture takes the role of
status of the network, in real-time.
physically receiving messages from the protocols used by
4. To be able to extend an implemented network with
the SO wireless nodes, decoding those messages and
simulated Smart Objects. In this way, a hybrid
interpreting their contents, and finally putting those con-
simulation/implementation scenario can be estab-
tents into infrastructure-native messages that can be for-
lished, providing an effective way of testing the
warded toward the rest of the Information Infrastructure
implemented network with communication loads that
components. The need for a gateway node is the result of a
would not be possible due to the lack of hardware
generic framework description, which is not bound to any
resources.
specific transport protocol, on both the SO network and the
Information Infrastructure. Simulations used a number of variables that influence
Infrastructure clients access the SO network structure the performance of the SO networks, such as the number of
repository through a query interface and may subscribe to nodes per cluster, the CH rotation time, the SO priority
sensor data events through a capture interface. The design within the cluster membership queues, or the radio range of
of these interfaces is such that it allows access to infor- the nodes. The simulation assumptions are detailed on
mation with various degrees of granularity. For example, a Table 2. Other simulation details will be given as necessary
client might subscribe to the data from a single sensor, a when describing particular results.
Smart Object, or a whole network. Subscriptions might also The key results of our evaluation are discussed in the
be specified for any data from a particular sensor type. The following subsections.
flexibility of these interfaces, coupled with the logical
relationships among the objects of a network, results in a 4.1 Improved network lifetime and balanced energy
powerful yet simple design. distribution

The double clustering mechanism and the variable rotation

4 Architecture evaluation time of CHs results in a balanced energy distribution
among network nodes and longer network lifetimes.
Both the Smart Object network algorithms and the infra- Figure 3 shows the average decrease in the residual
structure design were evaluated. Apart from confirming the energy of the Smart Object CHs when the number of nodes
design features described in previous sections, the network per SO is varied. The number of SOs in the network was
protocols were evaluated in terms of energy consumption fixed to four. The zigzagging patterns reflect the CH
and the variables that influence it. The addressing and rotation mechanism, with decreasing residual energy as the
routing mechanism was also compared with the de-facto simulation progresses. In Fig. 3a, the double clustering is
industry standard ZigBee [37]. The information infra- used, and both the roles of NetCH and SoCH are rotated.
structure design was evaluated in terms of its flexibility to As the number of nodes per SO increases, CHs will use

123
Pers Ubiquit Comput (2012) 16:291308 299

Table 2 Simulation assumptions

Assumption Description

Simulation area 500 m2

Initial node position Randomized
Addressing tree construction The root of the tree is assigned to the closest node to the center of the simulation area, and the rest
of the nodes are joined to the network in an increasing distance from the center
Initial residual energy 100 % if not explicitly stated
Communication range Enough to reach other network nodes, if not stated otherwise
Energy consumption for transmitting (etx) etx ecircuitry  k eamplify  k  d c ; erx ecircuitry  k; where etx is the energy used while
and receiving (erx) transmitting data, erx is the energy used while receiving data, ecircuitry is the energy needed to feed
the tx/rx circuits, eamplify is the energy required to amplify the tx signal, k is the length of the
packet to be transmitted, d is the distance between source and target and c is the path loss
exponent [38].

more energy in each period in order to cope with the the period increases. Long periods postpone the CH rota-
increasing packet forwarding needs. This is reflected in the tion excessively, provoking the death of the CH before the
graphs by steeper peaks. However, the greater number of election takes place. This is a sign of poor energy distri-
nodes also contributes to an increase in the lifetime of the bution, and a balance should be found between long net-
network by providing more candidates for the CH election. work lifetimes and ensuring that all the nodes of a network
As the graph shows, the simulation with 10 nodes per SO keep enough energy to operate.
continues for a longer time than the other simulations with Optimal balanced results can be obtained by incorpo-
fewer nodes. Figure 3b shows the same set of simulations rating the number of nodes of the cluster in the calculation
but only activating the NetCH rotation mechanism. As the of the CH rotation period in every CH election process. An
graph shows, the increase in the number of SO nodes now energy model was developed in [39], and its solution
shortens the network lifetime. produced a unified equation that network nodes might use
Figure 4 shows the average decrease in the residual to calculate the optimal CH rotation period as a function of
energy of the network when the time between CH rotations the current number of nodes in the cluster. The number of
is changed. The short lines perpendicular to the x-axis nodes in a cluster is a known variable calculated in a dis-
represent the time that the first node of the network com- tributed manner throughout the life of the network [36].
pletely exhausted its battery. Two observations can be Figure 5 shows the average decrease in the residual energy
deduced from these graphs: Firstly, the network lifetime of the network when the time between CH rotations is
tends to increase as the period increases. Secondly, the calculated dynamically using the aforementioned equation.
distribution of energy among the network nodes worsens as The time of the death of the first node is 12.16 min from

100 100
2 nodes 2 nodes
5 nodes 5 nodes
8 nodes 8 nodes
10 nodes 10 nodes

80 80
Residual Energy (%)
Residual Energy (%)

60 60

40 40

20 20

0 0
0 5 10 15 20 0 5 10 15 20
Time (m) Time (m)
(a) (b)
Fig. 3 Average energy consumption pattern of the network with 3 (a) and without 3 (b) the Smart Object cluster head (SoCH) rotation
mechanism activated

123
300 Pers Ubiquit Comput (2012) 16:291308

Fig. 4 Average energy consumption pattern of the network when increasing the CH period

the start of the simulation. This time is comparable with the protocol optimization line shows the number of exchan-
highest time obtained from Figure 4, which is 12.3 min ged messages using the Smart Object protocols. In this
for a rotation period of 2 min. However, the new results protocol, not all the cluster members answer the CH
produce a network lifetime of almost 20 min, which is far Election message, but only those that have a higher
better than the static 2 min period, which barely reached residual energy than the current CH. This strategy avoids
the 13 min mark. We can therefore conclude that the the exchange of unnecessary messages, yielding increasing
dynamic calculation of the optimal CH rotation period savings with increasing numbers of cluster members.
depending on the number of nodes present in the cluster To emphasize the improvement of the proposed proto-
maximizes the network lifetime while balancing the energy col, Fig. 6b shows the average number of messages
consumption among the cluster members. exchanged in each CH election procedure, obtained by
dividing the total number of messages by the number of
CH election procedures in each simulation. Based on these
4.2 Lower overhead and greater scalability
results, we can conclude that the protocol optimization
strategy delivers a highly scalable architecture.
The protocol for electing a new CH not only generates a
small amount of messages but its benefits over the number
of exchanged messages increase with the number of nodes
100
within the cluster. Figure 6a shows the total number of
messages exchanged in the CH election process in a
20 min simulation. The # election procedures line plots 80
the number of election procedures that were undertaken for
Residual Energy (%)

each variation of the number of nodes. Higher number of

60
nodes results in longer CH periods due to the better dis-
tribution of the available energy and therefore in lower
number of election procedures. A similar effect was 40

observed in Fig. 5 and explained in Sect. 4.1 The No

protocol optimization line plots the messages exchanged 20
if a simple request/response protocol is used. In this pro-
tocol, a CH broadcasts an Election message within its
0
cluster at the beginning of new rotation period. This mes- 0 5 10 15 20

sage would be answered by all the cluster members, and a Time (m)

new CH would be elected according to their responses. Fig. 5 Average energy consumption pattern of the network when
This strategy results in a linear increase in the number of dynamically calculating the CH rotation period according to the
messages with the number of cluster members. The With number of cluster nodes

123
Pers Ubiquit Comput (2012) 16:291308 301

Fig. 6 Number of messages exchanged in the CH election process during a simulation time of 20 min

4.3 Low latencies Smart Objects (i.e, network latency), are also short. Since
the packet latency is associated with its delivery ratio (i.e,
The time that a new Smart Object will have to wait from the number of packets that reach their destination), Fig. 8a
the moment it sends a request message until the Smart plots both performance measures averaged throughout 10
Object becomes part of a new network (i.e, merger latency) repetitions, with random origin and destination nodes and
is short, even when several requests are queued. Figure 7a varying the number of nodes between 20 and 100. In order
shows the merger latency with three priority levels. A SO to gain an understanding of how well the Sequence Chain
will be classified within a priority level according to the (SC) protocol performs, we measured the same perfor-
filtering algorithms mentioned in Sect. 3.3 The cross-mark mance indicators in the ZigBee tree routing protocol [37]
represents the average waiting time with processing cycles within the same simulation environment. Figure 8b and c
of 5 seconds, while the vertical lines represent the maxi- plot the results setting the nwkMaxDepth variable of
mum and minimum values across 10 repetitions. Although ZigBee to 2 and to 7, respectively. While SC
the latency seems large for 2nd and 3rd priority levels, the addressing trees are not restricted in their depth, ZigBee
probability that requests will have to wait several cycles is address trees need to fix their depth in order to calculate the
small. As an example, Fig. 7b shows the distribution of the addresses of their nodes. They achieve this by setting the
probability that a SO classified in the 2nd priority level will nwkMaxDepth variable to the desired maximum tree
have to wait several processing cycles. depth [37]. 2 was chosen as the minimum value with
The routing latencies inside a network, i.e, the time which a tree can be formed with 20 nodes, and 7 was
taken for a packet to navigate a route inside a network of chosen as the value showing the best results in terms of tree

(a) (b)
Fig. 7 Cluster membership processing latency

123
302 Pers Ubiquit Comput (2012) 16:291308

3500 100 3500 100 3500

100
Latency Latency
Delivery Ratio 3000 Delivery Ratio 3000
3000
80 80 80

2500 2500

Delivery Ration (%)

2500

Delivery Ration (%)

Latency (ms)

Delivery Ration (%)

Latency (ms)

Latency (ms)
60 60 60
2000 2000 2000

1500 1500 1500

40 40 40

1000 1000 1000

20 20 20
500 500 Latency
500 Delivery Ratio

0 0 0 0 0 0
20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100
Number of Nodes Number of Nodes Number of Nodes

(a) (b) (c)

Fig. 8 Comparison of latency and delivery ratio inside SO and ZigBee networks a Smart Object networks, b ZigBee networks with
nwkMaxDepth=2 (c) ZigBee networks with nwkMaxDepth=7

connectivity (see Fig. 9a). As results show, ZigBee pre- nwkMaxDepth variable. All the graphs in Fig. 9 show SC
sents a very large trade-off between delivery ratio and as horizontal lines since the measurements do not depend
latency, where delivery ratios comparable to those of SC on the variation of any variable, and adapt to the random
produce large latencies, and where small latencies com- simulated deployment scenarios. Values shown here aver-
parable to those of SC produce very low delivery ratios. age the results of over 10,000 simulations, where the
Judging by these results, we can conclude that SO networks number of nodes was varied from 0 to 100 and the radio
generate very low network latencies, outperforming the range of the nodes was varied from 0 to 300 m. For each
well-established routing protocol ZigBee for the same combination, 100 repetitions with randomized node
delivery ratios. deployment locations were conducted.
Orphan nodes (Fig. 9a) are those nodes that could not
4.4 Stable topologies connect to the tree during the tree formation phase either
because no neighbor was found within its radio range or
The last subsection showed how ZigBee topologies can because the neighbors found are unable to accept more
produce very disparate performance results depending children. Fewer orphans result in better network connec-
on very specific deployment decisions. The value of tivity. The depth of the tree (Fig. 9b) is an indicator of the
the nwkMaxDepth variable is supposed to be selected tree efficiency, since increasing the number of links, a
according to the forecast topology for a particular message has to traverse from its origin to the trees root
deployment scenario. Since the mobility of Smart Objects requires more time and energy. Similarly, the number of
makes it impossible to determine how the topology of SO hops (Fig. 9c) is the number of links between random
networks would evolve, a stable performance, independent origin and destination nodes. Results in Fig. 9 highlight not
of deployment scenarios, becomes very important. Figure 9 only the variability of the performance results of ZigBee
shows additional indicators of the stability of both SC and according to its topologies, but also the performance
ZigBee topologies, with a particular focus on how ZigBee superiority of SC in tree connectivity, tree depth, and
performance is affected with the selection of the number of routing hops.

40 7 3.2
ZigBee
ZigBee Sequence Chain
ZigBee Sequence Chain
Sequence Chain
3
Average number of orphan nodes

35 6
Average number of hops

2.8
30 5
Average depth

2.6

25 4
2.4

20 3 2.2

2
15 2

1.8
10 1
2 4 6 8 10 12 14 2 4 6 8 10 12 14 2 4 6 8 10 12 14

nwkMaxDepth nwkMaxDepth nwkMaxDepth

(a) (b) (c)

Fig. 9 Comparison of topology stability between SC and ZigBee tree networks a average orphan nodes, b average maximum tree depth,
c average number of hops

123
Pers Ubiquit Comput (2012) 16:291308 303

4.5 Infrastructure flexibility and performance 5 Architecture implementation

The information infrastructure design was evaluated in The described architecture features a flexible design which
terms of its flexibility for use as a reference model for does not restrict the technology used for its implementa-
implementation. The design was thus implemented with tion. In order to provide a prototype implementation that
various popular technologies, including XML and JSON- would allow us to test a variery of scenarios, the archi-
based web services, REST and binary distributed object tecture was implemented using Wireless Sensor Networks,
paradigms such as Java Remote Method Invocation (RMI). Web Services and relational databases (Fig. 11). One such
The various architecture interfaces and events were suc- scenario, based on a supply chain application, is presented
cessfully converted into these technologies with no loss of in Sect. 6.
functionality, and it was consequently concluded that the For empowering objects into Smart Objects presented in
proposed design is flexible and adaptable for the Internet of Sect. 3, we chose Wireless Sensor Networks as the logical
Things concept. solution given the current state of wireless, low-power
The information infrastructure was also evaluated in embedded technologies. Smart Objects were implemented
terms of data access performance. In particular, the design using the ANTS sensor network platform [40]. Each node
of real-time push of sensor data toward the infrastructure featured an 8 bit l-controller, a 2.4 Ghz transceiver,
subscribers was tested and its benefits quantified in terms 128 kb of Flash memory and 4 kb of SRAM and a variety
of scalability and latency. Two strategies were trialed. of sensors including pressure, humidity, temperature
Firstly, the sensor data were stored in the SO network sensors and accelerometers. The implementation used
structure repository along with the network structure data. the memory and processor of each node to execute the
Secondly, the sensor data were pushed in real-time from its algorithms described earlier, and the radio module to
reception in the capture interface to the subscribed clients. communicate with other nodes and with the information
Figure 10 plots the latency of 100 sensor data events infrastructure.
between their reception by the gateway until their delivery As shown in Fig. 11, the SO network (a) communicates
to the data subscriber, with the two strategies described with the infrastructure via a gateway node (b). In the
earlier. In order to test the performance with different load implementation, the gateway node serializes the SO events
conditions, several concurrent clients (i.e, data producers) and transmits them to a Network Translator (c), whose role
were tested while maintaining the same amount of total is to convert those messages into HTTP client requests to a
data events. This test was performed using the implemen- Java Servlet server that acts as the point of entry to
tation described in Sect. 5. Results show very clearly that the information infrastructure. From all the technolo-
whereas the latency increases linearly with the number of gies evaluated for the infrastructure implementation,
concurrent clients when the repository is used, it remains
constant when real-time push is utilized. These results
where retrofited into the design process to devise efficient
strategies for delivering large amounts of sensor data to a
large number of concurrent subscribers.

Fig. 10 Comparison of sensor data event latencies with and without Fig. 11 Smart Object information systems architecture
using the Smart Object network structure repository implementation

123
304 Pers Ubiquit Comput (2012) 16:291308

XML-based Web Services (WS) were chosen due to their environmental conditions such as temperature, are partic-
ability to provide cross-platform and cross-language com- ularly relevant for our discussion [43]. On top of these, the
munications over a network. XML and XSD were used to shift in the world economy toward a low carbon future
encode the events and event data. WSDL was used to through the reduction of carbon dioxide emissions, energy
describe the capture (d) and query interfaces (e). SOAP and consumption, and wastage caused by the transport pro-
HTTP were used as messaging protocols to transfer the cesses are also seen as related and significant goals [44].
XML-encoded events between the different infrastructure Consequently, there is an imperative need for ensuring
components. A relational database was used as a repository food quality across the supply chain. Achieving the
(f) in order to store the structure of the Smart Object net- objectives of food quality partly relies on physical trace-
works. Due to the flexible and interoperable nature of the ability throughout the chain. As a result, the management
XML-based WS, infrastructure clients could use a variety of the supply chain, more significantly the cold chain
of methods for encoding requests sent to the Capture and related to the manufacture, distribution and sale of per-
Query Interfaces. In our implementation architecture, both ishable, and condition-sensitive products, are seen as high
Web Browser (h) and Java based clients (g) were built. A priority applications.
screeshot of one such client for the supply chain scenario Today, these goals are addressed to a limited extent
presented in Sect. 6 is shown in Fig. 13. through central planning and optimization. Large logistics
networks consist of numerous destinations and vehicles.
Due to the complexity and scale of transportation pro-
6 Example scenario cesses, an optimal solution for managing such a distributed
and large scale supply network cannot be achieved cen-
While there are a large number of applications that could trally in a real-time manner to effect real change on the
benefit from the proposed framework, ranging from pre- ground. Dynamic changes and unforeseen situations, such
venting counterfeiting to defeating bio-terrorism, we con- as traffic jams, machine failures (e.g., refrigeration units),
sider an application in transport logistics to illustrate how and changes to delivery quantities pose a significant chal-
users can detect, track, trace, and manage complex business lenge to central planning and control facilities.
problems using the Smart Object framework. In this section, we argue that the integration of the Smart
An increasing number of drivers, such as achieving Object framework approach in transportation logistics to
greater supply efficiencies through the elimination of manage a supply network in a distributed manner can
waste, legislative drivers such as 2001/95/EC and 178/2002 potentially reduce waste, handle dynamic situations
to both reduce waste and make available fresh and safe autonomously, and ensure freshness of products as well as
food for consumption by the general public [41], and complying with legislative requirements. This is demon-
investments by retailers to meet customer expectation of strated by the following scenario, illustrated in Fig. 12.
quality [42], have created the impetus to re-examine ways Although the following application scenario is based on
of managing supply chains. More specifically, those drivers sample data (instead of data from real-life RFID imple-
associated with the transport of perishable goods, such as mentations), it is still a high impact and high priority
meat, poultry, fruits, and vegetables, which are sensitive to application.

Fig. 12 Example scenario for

condition monitoring of
refrigerated products in the
supply chain

123
Pers Ubiquit Comput (2012) 16:291308 305

Thousands of products such as seafood, milk, and fresh 6.1 Real-time condition track and trace
vegetables must be refrigerated upon manufacture. These
products are packaged and transported on pallets. These The condition of the pallets, resulting from the interpreta-
pallets travel from the point of manufacturer to retailers tion of sensor measurements taken by their integrated tags,
through various intermediaries such as distribution and can be tracked in real-time along the entire supply chain.
storage locations. As is often the case, these refrigerated Tracking is independent of the location of the client, the
goods are transported between locations using third party truck, and the pallets as long as the identity of the pallets
logistics (3PL) providers in refrigerated trucks or contain- are known and their network CH has access to a gateway
ers. In the prototype application scenario, illustrated in connected to the Internet. Partners such as the manufac-
Fig. 12, each pallet is a Smart Object equipped with a node turers that released the product, retailers that are waiting
that can sense ambient temperature. Furthermore, the for their shipment, or the logistic companies that are
trucks employed by the logistics services provider incor- responsible for the transportation of the goods can enable
porate temperature and humidity sensors. Although the tracking and tracing by simply subscribing to the capture
prototype application illustrates pallets and a truck, the interface with the unique identification numbers (the EPCs
application supports more complex and multiple Smart used in the application) of the pallets as illustrated in
Object networks (e.g where each box in each pallet is a Fig. 13.
Smart Object).
As indicated in Fig. 12, all the Smart Objects in the 6.2 Dynamic service discovery
scenario such as pallets and the refrigerated truck are
identified through a unique identifier, an Electronic Product The SO networking design enables the discovery of addi-
Code (EPC), stored in the memory of embedded wireless tional sensor sources relevant to the context of a particular
sensor nodes. The sensors installed in the node of the truck object and the automatic selection of the most appropriate
measure the temperature and humidity of the climatic zone data source from multiple data providers.
within the truck, and the node has access to Internet-based By querying the infrastructure repository with the EPC
services through a mobile gateway (e.g., 3G cellular net- of a specific pallet on the truck, parties throughout the
work). The enhanced infrastructure provided by the Smart supply chain can retrieve the EPCs of all the Smart Objects
Object framework is used in this application to achieve the participating in the current network. This is a dynamic
real-time management of the products supply network in a discovery mechanism that allows parties to determine that
distributed manner, and to reduce waste, help ensuring the truck contains a humidity sensor as well as a temper-
freshness of the products as well as complying with the ature sensor. Where humidity is also relevant to the con-
regulations in place. dition of the refrigerated products, the parties can also

Fig. 13 Data subscription Web

Service Java client software
running the example scenario.
The service location string
points to the capture interface
endpoint. The granularity of
query can be selected using the
drop-down list and SensorID
fields. The duration of the
capturing process and the
polling period can also be
specified. Graphs show the
sensor data for the available
transducers, on the requested
Smart Object with the required
granularity. Smart Objects are
identified by EPCs. The raw
XML files being exchanged
with the server via SOAP can
also be visualized for debugging
purposes

123
306 Pers Ubiquit Comput (2012) 16:291308

subscribe to the humidity data stream (see Fig. 13). Fur-

thermore, leveraging the EPC Network standards and
infrastructure, it is possible to query on-line information
services, such as EPC Information Services [45], to
obtain additional information related to specific Smart
Objects. For example, a trading partner in the prototype
application can determine the accuracy or suitability of
the temperature sensors of both the truck and the pallet
using sensor hardware metadata made available though
the EPC Information Services of the manufacturer or
the logistics company to select the more accurate or
suitable sensor data stream. In situations where each
item on a pallet carries a node, pallet networks would
be formed prior to the truck network and could thus
build an effective hierarchical network structure aiding
the discovery of meaningful contextual condition
information.

6.3 Level of service provisions

The SO architecture design permits assignment of priorities

and restrictions to control SO interactions. As a result,
logistic companies can ensure that only products from Fig. 14 Examples of messages exchanged between the architectural
selected suppliers, for instance gold service members, components a example of response by the Query interface, b example
make use of the resources of the company (e.g only the of a DataEvent
node of pallet 1 in Fig. 12 stores the appropriate credentials
and can therefore use the sensor and gateway of the truck). 7 Challenges
As a result, the framework not only supports advanced
application scenarios but also provides the opportunities to 7.1 Economic challenges
develop value added services.
Figure 13 shows a screenshot of the Data Subscription It is only in recent years that simple passive RFID tags
client software, monitoring all the sensor sources from the have become available at sufficiently affordable prices
SO network formed inside the truck from the example (around 7c per tag in large volumes) that many industry
scenario, for a period of 30 min, in 60 s intervals. sectors are considering widespread adoption of RFID.
Figure 14 shows two examples of messages exchanged Sensor-enabled active tags are likely to cost considerably
between the various architectural components of the SO more than simple passive RFID tags because of the addi-
framework for the scenario presented in this section. Fig- tional cost of the sensor, memory capacity, and batteries.
ure 14a shows the body of a SOAP message returned by For this reason, many of the early trials apply them to
the Query interface to an IDquery request for the SoID reusable assets (e.g food trays, pallets) rather than indi-
urn:epc:id:stgin:1111111.1111111.0001. vidual items in order to amortize the cost over a much
The SOAP header has been removed for clarity. The longer period of service.
response includes a list of the nodes that the SO contains,
together with sensors and their characteristics, as well as 7.2 Security and trust issues
the identifier of the network where the Smart Object is
located. Figure 14b shows a single DataEvent as sent by RFID usually requires the assignment of unique identifiers
the same Smart Object toward the Capture interface. The for each object. This results in fine-grained visibility and
example illustrates how the sensor data are encoded within tracking information, but means that an individual object is
the DataEvent messages. Data events like the one shown no longer anonymous as simply another instance of a
in Fig. 14b are parsed by the client software pictured in particular product type. At the same time, complete sensor
Fig. 13 in order to plot the graphs displayed in the information for an object is realistically likely to be frag-
screenshot. The Events and Sensor XML schema describe mented and distributed across its lifecycle, with each
the encoding of the events and the meaning of the sensor organization holding only the sensor information that was
data, respectively. collected while the object was within their custody. In

123
Pers Ubiquit Comput (2012) 16:291308 307

order to gather complete sensor information from across 3. EPCglobal Inc (2009) The EPCglobal architecture framework, v
multiple organizations, it may be necessary to provide 1.3, standard specification, March 2009
4. ISO/IEC 18000-6 REV1 (2009) Information technology radio fre-
authentication and business relationship credentials in quency identification for item management Part 6: parameters for air
order to justify the request, both for querying data from the interface communications at 860 MHz to 960 MHz, August 2009
Smart Objects and when they attempt to connect to the 5. Song EY, Lee K (2008) Understanding IEEE 1451 networked
infrastructure gateways. smarttransducer interface standard. IEEE Instrumentation &
Measurement Magazine, April 2008
6. Open Geospatial Consortium Inc (2007) OGC sensor web en-
7.3 Scalability challenges ablement: overview and high level architecture. OGC White
Paper, OGC 07-165, v3
A further challenge to the collection of sensor data is the 7. Kiritsis D et al (2008) Product lifecycle management and infor-
mation tracking using smart embedded systems, final report,
capacity available for storage of historical data. Consider- PROMISE FP6-507100, EU, June 2008
ation should be given to mechanisms that allow thresholds 8. Carrez F et al. (2009) Deliverable D3.2, reference architecture.
and also more complex exception criteria to be defined in a SENSEI FP7 215923, EU, January 2009
standardized manner. This may ultimately lead to a dif- 9. Ranasinghe DC, Leong KS, Ng ML, Engels DW, Cole PH (2005)
A distributed architecture for a ubiquitous RFID sensing network.
ferent paradigm for information sharing, in which alerts In: Proceedings of the ISSNIP05, Dec 58
and details of exceptions are returned by queries, rather 10. Finkenzeller K (2003) RFID handbook: fundamentals and
than large volumes of low-level location and sensor data. applications in contactless smart cards and identification. Wiley,
This strategy may allow compression of historical data for London
11. Beigl M, Krohn A, Zimmer T, Decker C (2004) Typical sensors
long-term storage and could also address some of the data needed in ubiquitous and pervasive computing. In: Proceesdings
sharing concerns since organizations may prefer to provide of INSS
object information only when exceptions and alerts arise. 12. Akyildiz IF, Su W, Sankarasubramaniam Y, Cayirci E (2001)
Wireless sensor networks: a survey. Computer networks. Else-
vier, Amsterdam
13. Sample AP, Yeager DJ, Powledge PS, Smith JS (2007) Design of
8 Conclusion a passively-powered, programmable sensing platform for UHF
RFID systems. In: Proceedings of IEEE international conference
We have presented an architecture that uses Smart Objects on RFID
14. Mitsugi M (2006) Multipurpose sensor RFID tag. In: Proceedings
to integrate technologies such as automatic identification, of the APMC 2006 workshop on emerging technologies and
sensor systems, embedded processing, context-aware ad- applications of RFID, Yokohama, Japan, 1215 December, IE-
hoc networking, and Internet-based services, which are ICE, Japan, pp 143148
identified as central to the realization of the IoT concept. 15. EU FP6 BRIDGE Project BRIDGE: building radio frequency
identification solutions for the global environment, 20062009.
Practical experience gained with the evaluation and http://www.bridge-project.eu
implementation of the architecture demonstrates that it is 16. Sung J, Sanchez Lopez T, Kim D (2007) The EPC sensor network
both feasible and flexible to adapt to a variety of applica- for RFID and WSN integration infrastructure. PerComW 2007,
tions and off-the-shelf technologies. Key observations IEEE Computer Society, pp 618621
17. ISO/IEC CD 24753.2 standard draft (2009) Information tech-
showed that the proposed architecture has good perfor- nology Radio frequency identification (RFID) for item manage-
mance in terms of network lifetime, overhead, and scala- ment application protocol: encoding and processing rules for
bility, as well as producing low latencies in the various sensors and batteries, March 2009
processes of the network operation. We can thus conclude 18. IEEE P1451.7/D.07 (2009) Draft standard for smart transducer
interface for sensors and actuatorstransducers to radio fre-
that not only does the proposed architecture succeed in quency identification (RFID) systems communication protocols
incorporating a number of key technologies but it does so and transducer electronic data sheet formats. January 2009
from a practical standpoint. Finally, a number of identified 19. Aberer K, Hauswirth M, Salehi A (2006) Global sensor networks.
challenges suggest that the adoption of the IoT, in general, Technical report LSIR-REPORT-2006-001
20. The CoBIs Consortium (2007) Deliverable D104, final project
and our architecture, in particular, is not only limited by report. CoBIs FP6-004270, EU, March 2007
developments in technology but also by social and trust 21. Cho J, Shim Y, Kwon T, Choi Y (2007) SARIF: a novel
issues. framework for integrating wireless sensors and RFID networks.
IEEE Wire Commun 14:5056
22. Wong CY, McFarlane D, Zaharudin AA, Agarwal V (2002) The
intelligent product driven supply chain.In: Proceedings of the
IEEE international conference on systems, man and cybernetics,
References vol 4, p 6
23. Wycisk C, McKelvey BH (2008) ulsmann, M smart parts supply
1. Uckelmann D, Harrison M, Michahelles F (2011) Architecting networks as complex adaptive systems: analysis and implications.
the Internet of Things. Springer, Berlin Int J Phys Distrib Logist Manage 38(2):108-125
2. Weisser M (1999) The computer for the 21st century. ACM 24. Montenegro G et al (2007) IPv6 over low-power wireless per-
SIGMOBILE Mobile Comput Commun Rev 3(3):311 sonal area networks (6LoWPANs): overview, assumptions,

123
308 Pers Ubiquit Comput (2012) 16:291308

problem statement, and goals. Network working group, IETF, 36. Sanchez Lopez T, Huerta Canepa G (2009) Distributed and
August 2007 dynamic addressing mechanism for wireless sensor networks.
25. The IPSO Alliance. http://www.ipso-alliance.org. Accessed 25 Taylor & Francis (accepted to the Int J Distrib Sensor Netw)
October 2010 37. ZigBee Specification (2008) Document number 053474r17, Zig-
26. IETF ROLL Working Group. http://tools.ietf.org/wg/roll. Accessed Bee Alliance, January 17 2008
25 October 2010 38. Heinzelman WR, Chandrakasan A, Balakrishnan H (2000)
27. Shelby Z, Bormann C (2009) 6LowPAN: the wireless embedded Energy-efficient communication protocol for wireless microsen-
Internet. Wiley, London sor networks. HICSS00;January 47; Maui, USA. USA: IEEE
28. EPCglobal Inc (2010) EPCTM Tag data standards version, v 1.5, Computer Society; 2000. p 10
standard specification, August 2010. 39. Sanchez Lopez T, Kim D, Huerta Canepa G, Koumadi K (2008)
29. Trifa V, Guinard D (2009) Towards the web of things, White- Integrating wireless sensors and RFID tags into energy-efficient
paper v 1.0 and dynamic context networks. Comput J. doi:10.1093/comjnl/
30. Fielding RT (2000) Architectural styles and the design of net- bxn036
work-based software architectures. Ph.D disseration, University 40. Kim D, Sanchez Lopez T, Yoo S, Sung J (2005) ANTS: an
of California, Irvine evolvable network of tiny sensors, embedded and ubiquitous
31. Shelby Z et al. (2009) CoAP feature analysis. 6lowpan Internet computing, LNCS. Springer, 3824/2005, pp 142151
draft, IETF 41. Trienekens JH, Beulens AJM (2001) The implications of EU food
32. OConnor MC (2006) Cold-chain project reveals temperature safety legislation and consumer demands on supply chain infor-
inconsistencies. RFID J. http://www.rfidjournal.com/article/ mation systems. In: 11th Annual world food and agribusiness
view/2860/. Accessed 25 October 2010 forum, Sydney.
33. Estrada-Flores S, Tanner D (2005) Temperature variability and 42. Deasy DJ (2002) Food safety and assurance: the role of infor-
food spoilage during delivery of online retail products. Acta mation technology. Int J Dairy Technol 55(1):3567
Hortic (ISHS) 674:6369 43. The Chill-On project (2005) EU DG Research OOD-CT-2005-
34. Blake D (2009) Effects of cargo loading and active containers on 016333. http://www.chill-on.com/
aircraft cargo compartment smoke detection times. Technical 44. European Commission (2004) European CO2 capture and storage
report DOT/FAA/AR-09/52, U.S. Department of Transportation, projects, project synopses, sixth framework programme, Euro-
Federal Aviation Administration pean communities
35. Sanchez Lopez T, Ranasinghe DC, Patkai B, McFarlane D (2009) 45. EPCglobal Inc (2007) EPCTM Information services, v 1.0.1,
Taxonomy, technology and applications of smart objects. standard specification, September 2007
J Inform Syst Front 120. doi:10.1007/s10796-009-9218-4

123

View publication stats