www.ietdl.

org
Published in IET Wireless Sensor Systems
Received on 19th October 2010
Revised on 1st February 2011
doi: 10.1049/iet-wss.2010.0085

ISSN 2043-6386

Clustering and fault tolerance for target tracking

using wireless sensor networks
S. Bhatti1 J. Xu1 M. Memon2
1

School of Computing, University of Leeds, Leeds, West Yorkshire, UK

Department of Computer Science, Graduate School of Systems and Information Engineering, University of Tsukuba,
Japan
E-mail: [email protected]

Abstract: Using wireless sensor networks to track a moving object provided a practical solution to a wide variety of applications
including, for example, wild life, military operations, intruder tracking and monitoring in indoor ofce buildings. While much
work has been done in this area, failures are not considered in most of the existing solutions. However, failures have to be
handled carefully in target tracking applications because of their unpredictable and dynamic nature of communication, such as
sensor energy depletion, severe environment conditions, unstable communication links and malicious attacks. Traditional
approaches of fault tolerance are not well suited to address these new challenges. Therefore the authors propose a novel faulttolerant target tracking (FTTT) protocol based on clustering. Also, the results of an investigation in terms of performance
overheads and scalable nature of the FTTT protocol via comparative simulations with LEACH are discussed. Overall, in all
the cases, FTTT consumes at least 25% less energy than LEACH.

Introduction

Moving object tracking applications using a wireless sensor

network (WSN) have gained noteworthy proliferation in
recent years [1 7]. During tracking of a moving object, the
type of sensor to be used and the form of signal to be sensed
depend on the type of object to be tracked. Regardless of
differences in signals to be sensed and varying targets to be
tracked, there are common features shared by tracking
applications. First, the tracking system should be able to
monitor the moving object accurately and then should report
to a base station (BS) in a timely manner. Secondly, the
sensor nodes (SNs) must be able to cooperatively collaborate
to manage redundant and correlated data before sending it to
the BS. This results in the reduction of collision, interference
and communication overheads in the shared media [5].
Owing to the applicability of WSN in a variety of dynamic
scenarios, SNs may fail as a result of energy depletion,
hardware failure, communication link errors, malicious attacks
and so on [8]. Therefore energy-efcient fault-tolerant
algorithms and protocols must be designed for WSNs. One
approach to manage energy consumption is cluster-based
network management, which also improves scalability and
resource allocation. The two-tier architecture is actually a
cluster-based network [6, 9, 10]. The lower tier consists of
SNs randomly distributed in the eld of interest. They are able
to gather the required information and send data packets
directly to their cluster heads (CHs). The higher layer employs
some powerful CHs in the corresponding cluster by collecting
information from SNs and sending it to the BS. In a
single-tiered model both SNs and CHs participate in packet
66

& The Institution of Engineering and Technology 2011

forwarding, whereas only CHs participate in packet

forwarding in a two-tiered model [11].
Fault-tolerant approaches developed for traditional wired
or wireless networks are not well suited for WSN because
of many differences between these networks. Fault-tolerant
techniques such as double and triple redundancy techniques
are effectively implemented in traditional environments
because they are not constrained by storage space,
bandwidth and computational power, while these techniques
need careful investigation of design constraints before
deployment in resource constrained WSN [12]. Other faulttolerant techniques of traditional systems like check
pointing and synchronisation can also be not directly
applied to WSN because of their limitations and dynamic
nature. Therefore new or modied fault-tolerant techniques
are required for WSN and some have developed; for
instance, in paper [8] a runtime recovery mechanism is
proposed, which detects faults in gateways and recovers
sensors from failed clusters by assigning them to healthy
gateways without re-clustering the system. This scheme
lacks the results of energy consumption and overheads. The
most frequently used fault-tolerant technique for WSN is
the deployment of redundant/surplus SNs. When redundant
nodes (RNs) are provided, then the BS is able to obtain
data; even if some SNs are failed due to any reason [11,
13]. Despite the best efforts of well-intentioned researchers,
the techniques available for tracking moving object with
fault tolerance are not keeping pace with the challenges of
modern systems [1, 5, 12]. This paper aims to propose and
investigate a novel strategy for CH and SN fault
identication for applications of WSN that involve object
IET Wirel. Sens. Syst., 2011, Vol. 1, Iss. 2, pp. 66 73
doi: 10.1049/iet-wss.2010.0085

www.ietdl.org
tracking. This work is an extension of our work presented in
[14] with more simulations. The model checking of FTTT is
presented in [15].
This paper makes three major contributions. First, we present
a static cluster formation technique, which facilitates CH
recovery procedure. Secondly, RN are identied and enforced
to maintain a sleep state in order to reduce the cost of fault
tolerance and conserve the total energy consumption of the
network. Thirdly, a technique of how failed CH member
nodes can be recovered and when CHs activate RNs, which
are currently in a sleep state. The rest of the paper is organised
as follows. Section 2 briey presents a summary of related
work. In Section 3, details of the proposed FTTT is illustrated.
The mathematical evaluation of FTTT protocol is given in
Section 4. Parameters and metrics for simulation of the
proposed protocol are demonstrated in Section 5. In Section 6,
simulation results are reported and analysed. Finally, Section
7 states concluding remarks.

Related work

This section briey discusses some well-known target

tracking protocols, clustering mechanisms and faulttolerance techniques adapted in WSN. An inspiring
protocol that reports the tracking information of a moving
object to a moving source is highlighted by Tsai et al. [5].
In this protocol, rst the face neighbours are identied by a
Gabriel graph (GG). In the target discovery step, the source
sends a query about location of a target to SNs and the SNs
close to the target reply back. To detect a moving target
continuously, the spatial neighbours of the near SN are
woken up. Next, in the target tracking step, the source
sends a query to the beacon node (SN keeping track
information of the target) which replies back the targets
next location and the source moves towards the next beacon
node. Tsukamoto et al. [7] derived approximate expressions
to show the relationship between sensor density, number of
monitoring SNs and tracking probability. In HPS [2],
cluster is formed using Voronoi division and a targets next
location is predicted via least square method, however,
overheads are not well dened.
Optimised communication and organisation (OCO) is a
method for target tracking that provides self-organising and
routing capabilities [1]. OCO has four steps: position
collecting, processing, tracking and maintenance. The
drawback of this approach is the activation of border SNs
permanently so their energy may be depleted rapidly. Chen
et al. [3, 4] developed a new dynamic shape of a
forwarding zone called variant-egg in order to maintain
minimum active SNs to track a target. In variant-egg-based
mobicast (VE-mobicast) routing protocol [3], a distributed
algorithm based on a simple control packet dynamically
adjusts the size and shape of the forwarding zone. HVE
mobicast [4] is a cluster-based VE mobicast routing
protocol. HVE mobicast protocol improves power efciency
of message delivery by providing cluster-to-cluster instead
of node-to-node delivery of a mobicast message.
An interesting example of hybrid static/dynamic clustering
is the continuous object detection and tracking algorithm [6];
however, it lacks a missing track recovery mechanism and
tracking accuracy measurement. The clustering mechanism
developed by Pino-Povedano et al. [16] takes into account
the position and velocity of a moving target based on
prediction and it classies the target using a state vector.
Node clustering is a non-trivial technique for managing
scarce network resources. A survey of target tracking
IET Wirel. Sens. Syst., 2011, Vol. 1, Iss. 2, pp. 6673
doi: 10.1049/iet-wss.2010.0085

protocols using WSNs is presented in [17]. The target

tracking protocols, which are discussed briey in this paper
[1 6, 16], are discussed in detail in [17] along with the
type of the protocol and the features supported by each of
the target tracking protocol.
A comparison of clustering algorithms for WSNs
according to various properties is presented in [18]. The
number of protocols is classied based on cluster
properties, CH capabilities and clustering process along
with their strengths and limitations. In addition, taxonomy
of relevant attributes is discussed.
Localised algorithms for faulty sensor identication are
developed in [13, 19, 20]. In [19], the sensor reading is
compared with its neighbours median reading. If the
difference is large or large but negative, the sensor is assumed
to be faulty. Computational overhead of the algorithm is low
because of simple numerical operations. Each SN identies its
own status to be faulty or non-faulty with the help of
neighbouring sensors [20]. The probabilistic analysis shows a
correct diagnosis but in reality there may not be enough
neighbour sensors because of random network deployment.
Recently, Behnke et al. [13] presented an algorithm which is
an improvement over Dings algorithm [19] in terms of less
computations, communications and complexity. In general, it
detects erroneous SNs using local outliers and can
successfully detect erroneous SNs when at most 25% of total
SNs in the network are faulty.
In all of the above protocols, clustering and fault tolerance
are not addressed together along with tracking of a target.
However, in FTTT protocol, clustering, CH fault tolerance
and SN fault tolerance are tackled in a distinctive manner
for a target tracking application.

Fault-tolerant target tracking protocol

Before designing a protocol for a WSN, it is crucial to think

about its scenario of applicability. It assists the designer to
understand the requirements of the application and explore the
application-specic assumptions. The FTTT protocol is
designed for tracking a moving target and is not restricted for
any particular type of target. Let the network consists of a
given number of CHs, which are placed at particular locations,
and m degree of redundant SNs which are randomly
distributed in the area of interest. Let a target be moving
according to random way point model [21] enters this area.
It is assumed that SNs have the capability to receive the
signals emitted from a target, such as seismic signals.
Let M SNs and N CHs be randomly distributed in a twodimensional (2D) space in sufciently dense collection so
that entire area of interest is covered. The CHs are not
energy constrained like SN with g times more energy than
SN and their responsibility is to report back to the BS. It is
assumed that in FTTT each SN knows its position using
GPS or any localisation algorithm. Four operational modes
are dened for the radios of CH and member node. The CH
and member node can be either in transmit, receive, idle
and sleep modes, and each mode is characterised by a
different amount of power consumption. The SNs can
switch between active, idle and sleep states. The CHs can
communicate with their NCHs. Following are the steps of
FTTT protocol which are designed in such a manner that
initial phases support forthcoming phases:
1. cluster formation;
2. redundancy reduction;
3. target tracking strategy;
67

& The Institution of Engineering and Technology 2011

www.ietdl.org
Table 1

Notational conventions

MNTj  MAMij

Notations

Descriptions

CHN
SNM
LSi (xi , yi)
LHj (xj , yj)
LT (xt , yt)
ELi
CHID
SNID
SCHID
RCHID
MNTj
NNTj
Edij
w1
w2

set of N CHs
set of M SNs
2D location of SNi
2D location of CHj
2D location of target
energy level of SNi
CH identity
SN identity
sending CH identity
receiving CH identity
member node table of CHj
NCH member node table maintained by CHj
distance between SNi and CHj
threshold value used during cluster formation
threshold value used during redundancy
reduction and SN fault tolerance
threshold value used during target tracking
packet sending time
size of packet
number of SNs sending data to the CHj
maximum limit of SNs sending data to the CHj
set of SNs unable to join any CH
threshold for maximum overlaps
faulty CH

w3
T
size
Ct
K
mi
maxO
CHf

4. CH fault identication and recovery;

5. SN fault identication and recovery.
For convenience, notational conventions are dened in
Table 1.
3.1 Cluster formation
The concept of clustering presented in [8] is extended to apply to
a target-tracking scenario. The cluster is formed based on the
communication range of the SNs and the CHs. The SNs that
are within the communication range of CHj become its
members. The steps numbers in Fig. 1 are similar to the steps
numbers detailed in boxes. The details of communication
between CHj and SNi during cluster formation are illustrated as
follows, where CHj(MM)  SNi means CHj sends MM to SNi
CHj (MM)  SNi : MM = (CHID , (xj , yj ), size)

(1)

SNi (MAM)  CHj : MAM = (SNID , CHID , (xi , yi ), size, ELi )

(2)

NNTj  MNTj+1 ,
CHj (SCM)  SNi ,

NNTj+1  MNTj

SCM = (CHID , joined SN, size)

(3)
(4)
(5)

The CHj will send a membership message (MM) to SNi

which are within its communication range (1). The SNi
calculates its distance Edij from the CHj and will send a
membership accept message (MAM) to the CHj as step (2)
based on the following condition
SNi (MAM)  CHj ,

if j [ CHN , i [ SNM

The CHj maintains MNTj which consists of three elds

about its member nodes: (i) Sensor ID (ii) Sensor
location and (iii) Energy level, based on MAM (3). If a
member node is within communication range of two or
more CHs, then based on w1 , it will decide which CH it
can join. The w1 is controlled according to transmission
ranges of SNs and CHs. The CHj shares MNTj with
their NCHs which is maintained along with the ID of the
CHj and the table is termed NNTj (4). The CHj sends
time division multiple access (TDMA) schedule (SCM) to
its member nodes as step (5). There is a possibility that
mi SNs are unable to join any CHj .
3.2

Redundancy reduction

It is assumed that each point within the area of interest is

covered by at least m sensors so that m degree of redundancy
is achieved in the area. In redundancy reduction step of
FTTT, it is ensured that if two SNs have overlapping sensing
area, then only one active SN will forward data to the
corresponding CH. The FTTT protocol checks to see if there
are any neighbour SNs within the sensing range of a SN by
using threshold value w2 , then a sleep message (SM) is sent
to the overlapped SN.
The CHj builds a graph structure to nd out RNs in MNTj .
The vertices of the graph represent member nodes and the
edges represent distance between them. Each SN determines
its overlap with every other SN in the MNTj by measuring
the distance and then compares its value with w2 to nd out
overlapping area with that particular SN, where w2 depends
on transmission range of SNs. If the distance value is less
than w2 , it is found that the two SNs are overlapping. The
CH then identies the SNs with maximum overlaps and

Fig. 1 Fault-tolerant target tracking protocol

68

& The Institution of Engineering and Technology 2011

IET Wirel. Sens. Syst., 2011, Vol. 1, Iss. 2, pp. 66 73

doi: 10.1049/iet-wss.2010.0085

www.ietdl.org
drainage or because of damage.

sends them a sleep message (SM) as step (6).

CHj (SM)  Si
3.3

j [ CHN , i [ SNM

Target tracking

The CHj is sending TDMA schedule to its member nodes after

clustering and the member nodes are working according to
TDMA schedule. The active SNs in a cluster can sense the
target when it is within their sensing range. To reduce the
energy loss, it is important to restrict more than K SNs
sensing the target, to transmit information to the CHs. By
limiting the number of reporting SNs up to K, energy
consumption is reduced considerably because of avoidance of
collisions, idle listening and redundant packet transmission.
The SNi measures Edit from the target when it is getting
location information of the target LT(xt, yt) as step (7). The
data message (DM) is sent to the CHj as step (8) based on the
following condition
SNi (DM)  CHj ,
j [ CHN ,

CHj+1 = faulty,

(6)

if Edit f3 > ct , K

i [ SNM

where ct represents count of the number of SNs sending data

to the CHj at every instant. Finally, the data are sent to the
BS as step (9) and (10). This process will continue until
the target is within the sensing region. The messages
passed during target tracking demonstrated in Fig. 1 are
detailed as
SNi  (xt , yt )

(7)

|NCH| 2
3.5

CHj (DM)  BS: DM = (CHID , t, size, (xt , yt ))

(10)

3.4

CH fault identification and recovery

There can be three possible fault scenarios for a CHj . First is the
link failure with only one particular NCH due to any type of
interference. Second is energy depletion and the third is
complete damage. The CHj sends an activate message (AM)
to NCH in order to check their status periodically (11). If a
NCH does not send a reply message (RM) to CHj as dened
in step (12), then it is assumed to be a faulty CH, CHf . The
details of communication during fault tolerance of CHs
revealed in Fig. 1 are demonstrated as follows
CHj (AM)  CHj+1 : AM = (SCHID , RCHID , t, size)
CHj+1 (RM)  CHj : RM = (RCHID , t, size)

(11)
(12)

The rst step is to identify which fault has occurred. The CHj
would consult its NCHs to send an activate message to CHf
and if they get RM from CHf then it is identied that the
communication link between CHj and CHf has failed. Thus,
no recovery procedure is required as other CHs can
communicate with the CHf and it can take part in the
tracking process. If NCHs also do not get RM, then it is
identied that CHf has failed, either because of energy
IET Wirel. Sens. Syst., 2011, Vol. 1, Iss. 2, pp. 6673
doi: 10.1049/iet-wss.2010.0085

SN fault identification and recovery

There can be three possible scenarios for a member node

failure: rst is the energy depletion; second is the hardware
failure and third is the complete damage. When the CHj
knows that a SNi has died, it activates overlapped SNs
whose neighbour SNi has died. From the ID of the dead SNi ,
the CHj nds out its location from the MNTj as step (13).
The CHj then determines any overlapped SNi which is in a
sleep state near the dead SNi as dened in step (14) and
sends an activate message to it (15) using the following
condition
CHj (AM)  SNi ,

if Edij , f2

The details of communication during fault tolerance of SNs

revealed in Fig. 1 are demonstrated as follows

(8)
(9)

j [ CHN

where cluster N represents count of the number of CHs sending

RM to CHj . In that situation NCHs come to know that CHf
is not responding and will use NNTj to nd member nodes
of CHf and send MM to its member nodes. Next, cluster
formation procedure dened in Section 3.1 is repeated from
step (1) to step (5). The mutual CH consulting mechanism
for fault tolerance restricts applicability of FTTT for a given
number of CH failures. FTTT is applicable if the following
condition related with the neighbouring CH set is valid

SNi (DM)  CHj : DM = (SNID , CHID , t, size, (xt , yt ))

SNi (DM)  BS: DM = (SNID , BS, t, size, (xt , yt ))

if cluster N , NCH

MNTj  LSi (xi , yi )

(13)

MNTj  LSi (xi , yi ), SNID

(14)

CHj (AM)  SNi : AM = (SCHID , SNID , t)

(15)

The fault tolerance of SNs is based on the deployment

redundancy of SNs in the eld. The CH would be able to
substitute a faulty SN with the condition that there is an
overlapped SN near the faulty SNs.

Mathematical derivation

In this section, we mathematically calculate the energy ETotal ,

the tracking probability Pt(k) , energy saving Esav and average
latency Lavg .
Theorem 1: The total energy consumed by CHN and SNM
during the execution FTTT protocol is
ETotal = EMN + ENCH + EB

(16)

Proof: According to [9], the energy dissipated by SNi for

transmitting and receiving b bit data packets over distance
Edij is Ec (100 + 0.1Ed2ij)b. It is extended to compute the
total energy consumption during communication between
CHj and SNi according to [22] as
EMN =

N 
M


(100 + 0.1Edij2 ) b

j=1 i=1

69

& The Institution of Engineering and Technology 2011

www.ietdl.org
Let Dj is the distance between CHj or mi and the BS, while
dj is the distance between CHj and CHj+1 . The energy
consumed during communication among CHs is
ENCH =

N


5
(100 + 0.1dj2 ) b

j=1

In FTTT only mi SNs which are unable to join any CH are

assumed to communicate with the BS. Thus, energy
consumed during communication with BS is
EB =

mi
N 


(100 + 0.1D2j ) b

j=1 i=1

Thus, total energy dissipation is given as

ETotal =

N 
M


(100 + 0.1Edij2 )b +

j=1 i=1

N


(100 + 0.1dj2 )b

j=1

mi
N 


(100 + 0.1D2j ) b

j=1 i=1

A
Theorem 2: The probability Pt(k) as the total R packets are
received by BS in response of Pj packets send by CHj and
Pi packets send by mi SNs is given as
Pt(k) =

SRa=1 Pa
mi
SCHj
j=1 Pj + Si=1 Pi

(17)

Proof: Let the only packet received by the BS is Pa in

response of packet Pj send by CHj , thus Pt(k) Pa/Pj .
Similarly, the total packets received by the BS send by CHj
and mi SNs can be represented as the sum by (17).
A
Corollary 1: Let the energy difference gained by identifying
and switching RNs to sleep state is
Esav =

The time complexity of (16) is O(M 2) and it is O(M ) for

(17), (18) and (19).
A

ERA
CHj
Sj=1 Ej

ERS
SN

+ Si=1i Ei

(18)

Proof: The energy dissipation with RN in active state ERA and

RN in sleep state ERS are computed according to Theorem
1. Let the initial energy assigned to CHj and SNi are Ej and
Ei , respectively. Thus, Esav can be given by (18).
A

The performance of FTTT is evaluated via simulation

comparison with LEACH [9], which is one of the
non-trivial representatives of the clustering-based method
with energy-efcient load balancing. Other reasons of
selecting LEACH for comparison are that the FTTT
protocol uses a clustering mechanism as its base and a
number of recent studies have compared their results with
LEACH [1, 10]. The FTTT uses the radio model of
LEACH. In LEACH SNs employ randomised rotation to
choose CHs depending on the amount of energy left to
evenly distribute the energy load among altered SNs at
different times in the network. A limitation of this scheme
is that it assumes that all SNs have enough power
to communicate with the BS which is not true in the large
WSNs.
The simulations are performed using J-Sim [23, 24], which
is a Java open-source, component-based compositional
network simulation environment equipped with a package
to simulate WSNs. Simulation study shown in [24]
demonstrates better performance of J-Sim than ns-2 [25] in
terms of simulation time and scalability in memory usage.
Another reason for selecting J-Sim in this research is its
exible component-based architecture. The simulations of
the FTTT are performed by varying the number of SNs and
CHs. The experiments are performed with 100, 150 and
200 SNs which are randomly distributed over
400 400 m2 region. In the results, FTTT3, FTTT4 and
FTTT5 indicate the eld with 3, 4 and 5 CHs, respectively,
which are placed at particular locations. The CHs are
required to cover overlapping areas to support fault tolerance.
The J-Sim simulation parameters are bandwidth 1 Mbps;
frequency 914 MHz; receiver threshold 6n W; carrier
sense threshold 1n W and capture threshold 10 dB. The
initial energy of each SN is set 1 J and each CH is set 2 J.
The communication range of SNs is set as 50 m and sensing
range is set as 40 m. CH range is set as 150, 130 and 120 m
with FTTT3, FTTT4 and FTTT5, respectively. The target
moves according to a random waypoint model by varying
speed between 10 and 20 m/s and generates stimulus every
3 s. The sensor channel uses seismic propagation model and
MAC CSMA is adapted as underlying MAC protocol.
The parameters of LEACH are taken same as FTTT, that is,
100, 150 200 SNs; among them CH is identied
dynamically with initial energy of 1 J. This set of parameters
is common for FTTT and LEACH to avoid giving
misleading conclusions.

6
Corollary 2: Let the Lavg in terms of the time difference
between target information packet reception by CHj or SNi
and packet delivery to the BS is given as
Lavg =

N
SM
i=1 ti + Sj=1 tj

SRa=1 Pa

(19)

70

& The Institution of Engineering and Technology 2011

Simulation results and analysis

The evaluation of FTTT is measured by six metrics: (i) total

energy consumption; (ii) tracking accuracy; (iii) network
throughput; (iv) energy reduction with RNs in sleep state;
(v) average latency and (vi) fault tolerance. The results are
plotted after taking the mean of ve simulation runs of 800 s.
6.1

Proof: Let ti , tj are the time intervals when a target information

packet received by SNi , CHj , respectively, are delivered to the
BS. Let the latency for a packet Pa send from CHj is L tj/Pa .
Similarly, Lavg is computed as the sum of all times taken by a
packet send from CHj or SNi to reach the BS.

Performance evaluation

Total energy consumption

The ETotal consumed by FTTT protocol is computed using

(16). Fig. 2 reveals the ETotal consumption comparison of
FTTT and LEACH for different target speeds and varying
number of SNs and CHs. The effectiveness of FTTT is
IET Wirel. Sens. Syst., 2011, Vol. 1, Iss. 2, pp. 66 73
doi: 10.1049/iet-wss.2010.0085

www.ietdl.org

Fig. 2 ETotal with varying number of SNs and CHs

a Total energy with 100 SNs
b Total energy with 150 SNs
c Total energy with 200 SNs

shown by ETotal consumption difference between LEACH and

FTTT. With the increase in number of SNs, ETotal consumed
by LEACH increases as compared to all the cases of FTTT.
FTTT5 (20 m/s) is presenting the highest ETotal than all
other cases of FTTT irrespective of the number of SNs.
However, the scalability of FTTT is conrmed by ETotal
difference of 150 and 200 SNs which is minimum. Both the
protocols demonstrate discrepancy in ETotal by variation in
target speed and show the direct proportionality between
the two. With 200 SNs FTTT consumes 36, 32 and 24%
less energy as compared to LEACH in case of FTTT3,
FTTT4 and FTTT5, respectively.
6.2

Tracking packet delivery

The Pt(k) is computed using (17). Thus, the BS is receiving

two types of data, one from SNM directly and another from
CHN . The Pt(k) rises in FTTT with the number of SNs
while reduces in LEACH. Both methods are able to sense
the target, but LEACH packet delivery probability is less in
all the cases because of dynamic clustering [9]. The Pt(k)
value is relatively high with 200 SNs in all the cases. The
tracking probability has a direct relationship with the
number of SNs as demonstrated in study [7]. The small
values of standard deviation in Fig. 3 conrms that the Pt(k)
is close to the mean in all cases.
6.3

Fig. 3 Pt(k) with varying number of SNs and CHs at different target
speeds

Energy reduction with RNs in sleep state

The energy reduction by stopping overlapping SNs to sense

the target and send information to CHj is shown in Fig. 4.
The Esav saving is calculated as follows (18): RN: active
state corresponds to energy dissipation by FTTT protocol
without identifying overlapped SNi and allowing RNs to
send sensed information to CHj and RN: sleep state
IET Wirel. Sens. Syst., 2011, Vol. 1, Iss. 2, pp. 6673
doi: 10.1049/iet-wss.2010.0085

Fig. 4 Esav with varying number of SNs and CHs

71

& The Institution of Engineering and Technology 2011

www.ietdl.org
indicates FTTT with all the steps. It is obvious that a
signicant amount of energy saving is achieved in case of
the RN: sleep state. One of the reasons for the greater
energy consumption by the RN: active state is redundant
packet transmission towards BS. In the RN: active state
more SNs are detecting the target and are sending data to
CH at the cost of additional energy while in the RN: sleep
state, RNs are in sleep state as indicated and only K SNs
are allowed to send data to the CHj .
6.4

Network throughput

The network throughput in terms of total packets received at

the BS is dened as NTP (SRa21Pa/Ts), where Ts is the
simulation time. Fig. 5 illustrates signicant differences in
the NTP of both protocols with varying speeds of moving
target. In RN: active state, there is a continuous increase in
the NTP because of higher packet reception by BS, thus
consuming more energy as compared to FTTT3, FTTT4
and FTTT5. The NTP increases with the increase of number
of SNs in FTTT3, FTTT4, FTTT5 and LEACH with
highest rate in RN: active state having approximately
double rate in case of FTTT3, FTTT4 and FTTT5,
respectively, with 200 SNs. The NTP drops off when target
speed is increased from 10 to 20 m/s in LEACH, FTTT3,
FTTT4 and FTTT5; however, increases with the increase of
number of SNs from 100 to 200. The NTP of LEACH is
considerably less as compared to FTTT because of dynamic
clustering which raised collisions at MAC layer during
simulation.
6.5

Fig. 6 Average latency with varying number of SNs at different

target speeds

Average latency

The Lavg is calculated using (19). The average latencies does

not show consistent improvement or decline because of the
random time calculations considered during different steps
of experiment in LEACH and FTTT. Fig. 6 highlights the
average latencies of all the methods with standard deviations.
6.6

Fault tolerance

Fault tolerance is measured by the number of overlapping

SNs, Noverlap , which are in sleep state and total number of
SNs in the network as (Noverlap/M ) 100. The graph shows
uctuation in the percentage of fault tolerance with the
increase of network size in different iterations. FTTT5
offers the best fault tolerance among the three versions of
FTTT due to the identication of more RNs by CHs as
illustrated in Fig. 7.

Fig. 7 Fault tolerance of SNs

Overall, the simulations verify the scalability of FTTT in

terms of trade-off between number of SNs, energy
utilisation and packet reception at BS. By experimental
results, the performance of FTTT protocol is proved to be
better than LEACH while minimising energy consumption
during target tracking and fault tolerance.

Fig. 5 NTP with varying number of SNs at different target speeds

72

& The Institution of Engineering and Technology 2011

Conclusion

In this paper, we proposed a cluster-based target tracking

strategy allowing fault tolerance with a minimum energy
consumption and high tracking probability by varying the
number of SNs, CHs and target speed. By organising
the network into clusters, our scheme is capable of tracking
the moving target, as well as recovering SNs and CHs from
failures. The protocol supports target tracking even when
the system is fragile while prolonging network lifetime.
The FTTT protocol reduces energy consumption by
identifying RNs working in the network and minimising the
number of SNs sending data to the CHs. Further, the CH
fault-tolerance mechanism is applicable for a specic
number of CH failures and SN fault tolerance is applicable
with the condition of at least m degree of deployment
IET Wirel. Sens. Syst., 2011, Vol. 1, Iss. 2, pp. 66 73
doi: 10.1049/iet-wss.2010.0085

www.ietdl.org
redundancy. Simulations demonstrate that FTTT protocol
achieves better results when compared with LEACH via
less energy consumption, higher tracking probability and
fault tolerance. It is observed that ETotal , Pt(k) and NTP show
the direct relationship with the number of SNs and CHs.
Future work includes exploring the tracking quality and
recovery of target losses in FTTT. It would be useful to
explore the upper bound on the degree of redundancy to
achieve fault tolerance.

Acknowledgments

The authors would like to thank PC-I scheme for

strengthening and development of MUET approved by
CDWP through HEC, Pakistan scholarship for funding this
research. Special thanks to Dave Webster from Leeds for his
comments and suggestions intended for the improvement of
this paper.

References

1 Tran, S.P.M., Yang, T.A.: OCO: optimized communication and

organization for target tracking in wireless sensor networks. Proc.
IEEE Int. Conf. on Sensor Networks, Ubiquitous, and Trustworthy
Computing, (SUTC06), 2006, pp. 428435
2 Wang, Z., Li, H., Shen, X., Sun, X., Wang, Z.: Tracking and predicting
moving targets in hierarchical sensor networks. IEEE Int. Conf. on
Networking, Sensing and Control, ICNSC, 2008, pp. 1169 1173
3 Chen, Y., Ann, S., Lin, Y.: VE-mobicast: a variant-egg-based mobicast
routing protocol for sensornets, Wirel. Netw., 2008, 14, (2), pp. 199 218
4 Chen, Y.-S., Liao, Y.-J., Lin, Y.-W., Chiu, G.-M.: HVE-mobicast:
a hierarchical-variant-egg based mobicast routing protocol for wireless
sensornets, Telecommun. Syst., 2009, 41, (2), pp. 121 140
5 Tsai, H.-W., Chu, C.-P., Chen, T.-S.: Mobile object tracking in wireless
sensor networks, Comput. Commun., 2007, 30, pp. 18111825
6 Chang, W.-R., Lin, H.-T., Cheng, Z.-Z.: CODA: a continuous object
detection and tracking algorithm for wireless ad hoc sensor networks.
Fifth IEEE Consumer Communications and Networking Conf.,
CCNC, 10 12 January 2008, pp. 168174
7 Tsukamoto, K., Ueda, H., Tamura, H., Kawahara, K., Oie, Y.: Design
of wireless sensor network for multi-point surveillance of a moving
target based on the relationship between tracking probability and
sensor density. Sixth IEEE Consumer Communications and
Networking Conf., CCNC, 10 13 January 2009, pp. 15
8 Gupta, G., Younis, M.: Fault-tolerant clustering of wireless sensor
networks, IEEE Wirel. Commun. Netw., 2003, 3, pp. 15791584
9 Heinzelman, W.R., Chandrakasan, A., Balakrishnan, H.: Energyefcient communication protocol for wireless microsensor networks.
Proc. 33rd Annual Hawaii Int. Conf. on System Sciences, 4 7
January 2000, pp. 3005 3014

IET Wirel. Sens. Syst., 2011, Vol. 1, Iss. 2, pp. 6673

doi: 10.1049/iet-wss.2010.0085

10 Kim, K.T., Yoo, H.K., Son, B.H., Youn, H.Y.: An energy

efcient clustering scheme for self-organizing distributed wireless
sensor networks. Proc. 10th IEEE Int. Conf. on Computer and
Information Technology, Bradford, UK, 29 June01 July 2010,
pp. 1468 1473
11 Liu, H., Nayak, A., Stojmenovic, I.: Fault tolerant algorithms/protocols
in wireless sensor networks, in Misra, S. et al. (eds.): Guide to wireless
sensor networks, computer communications and networks (SpringerVerlag London Limited, 2009), pp. 265295
12 Paradis, L., Han, Q.: A survey of fault management in wireless sensor
networks, J. Netw. Syst. Manage., 2007, 15, (2), pp. 171 190
13 Behnke, R., Salzmann, J., Simanowski, S., Timmermann, D.: Efcient
localized detection of erroneous nodes (ELDEN). Proc. 2010 10th IEEE
Int. Conf. on Computer and Information Technology (CIT 2010), 2010,
pp. 2884 2890
14 Bhatti, S., Xu, J., Memon, M.: Energy-aware fault-tolerant clustering
scheme for target tracking wireless sensor networks. Proc. Seventh
Int. Symp. on Wireless Communication Systems, York, The
University of York, UK, 1922 September 2010, pp. 531 535
15 Bhatti, S., Xu, J., Memon, M.: Model checking of a target tracking
protocol for wireless sensor networks. Proc. 10th Int. Conf. on
Computer and Information Technology, Bradford, UK, 29 June 1
July 2010, pp. 2867 2872
16 Pino-Povedano, S., Gonzalez Serrano, F.-J.: Distributed tracking and
classication of targets with sensor networks. 16th Int. Conf. on
Software, Telecommunications and Computer Networks, SoftCOM
2008, 25 27 September 2008, pp. 213217
17 Bhatti, S., Xu, J.: Survey of target tracking protocols using wireless
sensor network. Proc. Fifth Int. Conf. on Wireless and Mobile
Communications, France, 23 29 August 2009, pp. 110115
18 Abbasi, A.A., Younis, M.: Survey on clustering algorithms for wireless
sensor network, Comput. Commun., 2007, 30, pp. 28262841
19 Ding, M., Chen, D., Xing, K., Cheng, X.: Localized fault-tolerant event
boundary detection in sensor networks. Proc. IEEE INFOCOM 2005,
Miami, March 2005, pp. 902 913
20 Chen, J., Kher, S., Somani, A.: Distributed fault tolerant detection of
wireless sensor networks. Proc. 2006 Workshop on Dependability
Issue in Wireless Adhoc Networks and Sensor Networks, Los
Angeles, September 2006
21 Alaybeyoglu, A., Dagdeviren, O., Erciyes, K., Kantarci, A.:
Performance evaluation of cluster-based target tracking protocols for
wireless sensor networks. Proc. 24th Int. Symp. on Computer and
Information Sciences, ISCIS, 2009, pp. 357 362
22 Halgamuge, M.N., Guru, S.M., Jennings, A.: Energy efcient cluster
formation in wireless sensor networks. Proc. Tenth Int. Conf. on
Telecommunications, ICT 2003, 23 Feburary 1 March 2003, vol. 2,
pp. 1571 1576
23 Sobeih, A., Chen, W., Hou, J., et al.: J-Sim: a simulation and emulation
environment for wireless sensor networks, IEEE Wirel. Commun.,
2006, 13, pp. 104119
24 Tyan, H.: Evaluation of J-Sim, October 2003 [accessed 9 November
2008]. Available at: http://sites.google.com/site/jsimofcial/j-sim-vsothers
25 The network simulator Ns-2 [accessed 10 July 2008]. Available at:
www.isi.edu/nsnam/ns/

73

& The Institution of Engineering and Technology 2011