Content uploaded by Safdar Hussain Bouk
Author content
All content in this area was uploaded by Safdar Hussain Bouk
Content may be subject to copyright.
Distributed Multiple Criteria based Clustering
Scheme for Wireless Sensor Networks
M. Mustafa†, T. Shah †, Safdar H. Bouk †a, Syed H. Ahmed ‡band N. Javaid†c
†Dept. of EE, COSMATS Institute of Information Technology, Park Road, Chak Shahzad, Islamabad, Pakistan 44000
‡School of Computer Science and Engineering, Kyungpook National University, Korea.
Email:abouk@comsats.edu.pk, bhassan@monet.knu.ac.kr, cnadeemjavaid@comsats.edu.pk
Abstract—Stability and lifetime of Wireless Sensor Networks
(WSNs) mainly depend on energy of each node in the network.
Hence, it is necessary for any technique proposed for WSNs to be
energy efficient. There are different methods to preserve energy
in WSNs and clustering is one of them. Clustering technique
divides whole network into small groups, each having a managing
node, called cluster head (CH) and rest act as members. CH is
responsible to provide communication bridge between members
and the base station. In this paper, we propose a distributed
clustering scheme that uses multiple criteria i.e. residual energy,
node degree, distance to the base station and average distance
between a node and its neighbors, to select a CH. Fuzzy Technique
for Preference by similarity to Ideal Solution (Fuzzy-TOPSIS)
method is used to outrank the potential nodes as CHs. The
realistic multi-hoping communication model is used in both, inter-
cluster and intra-cluster communication, instead of single hop as
in previous schemes. Simulation results show that our purposed
technique performs much better than previous methods in terms
of energy efficiency, network life time, less CH deformation and
control overhead.
Keywords—WSNs, Clustering, Cluster Head, Fuzzy-TOPSIS.
I. INTRODUCTION
Low-power electronic devices have grown interest in recent
years. Wireless Sensor Networks (WSNs) use these many
low-power devices along with communication capabilities for
sensing and monitoring various fields. Major areas of WSNs
include environmental sensing of temperature and humidity,
earthquake monitoring, healthcare monitoring and battle field
surveillance [1]. In some applications sensor nodes are de-
ployed in strategic manner but in most of the applications like
battlefield, these nodes are dispersed randomly.
After deployment in the field, sensor nodes have to be
self-organized without human interference. These sensor nodes
consist of battery operated radio devices, which have limited
memory and processing capabilities [2]. Their batteries cannot
be charged or replaced after deployment. Hence, energy is
one of the major issues in WSNs. The sensor nodes not only
sense data but they also process data and communicate with
the Base Station (BS). The communication and processing
of data are main causes of energy consumption and they are
major requirements of WSNs, therefore, these tow must have
to be energy efficient. Lots of research has been conducted in
WSNs on energy efficiency to prolong network life time and
stability. This includes design of various energy efficient MAC
and routing protocols. Routing protocols may either be flat or
hierarchical. In flat protocols, each sensor node sends data to
BS directly or in a multi-hop fashion [3][4]. On the other hand,
in hierarchical architecture, WSNs are divided into optimum
number of groups or clusters. Inside each cluster a Cluster
Head (CH) is selected to perform management and routing
tasks for that cluster. Research has proved that hierarchical
protocols perform much better than the flat protocols in terms
of energy efficiency and stability. The CH is selected either
randomly or based on some criteria [5]. After selection of a
CH, other nodes join that cluster and act as member nodes.
These member nodes send their data to CH, which then
aggregates data and sends to the BS. Hence, the selection of a
CH largely affects whole network’s performance and stability.
In most of the previous clustering schemes, CH selection is
based on single criterion. In single criterion, CH is mostly
selected randomly or based on residual energy, node density
or distance form BS. If CH is selected on the basis of residual
energy only, then problem arises when a node with higher
residual energy but located far away from BS is selected as
CH. That node consumes more energy to forward aggregated
data to BS. Similarly if CH selected no the bases of shortest
distance form BS, then similar type of problem arises if node
near to BS is selected as CH, but with not sufficient residual
energy to communicate with BS. Hence, single criterion is not
suitable for CH selection. Therefore we use all four criteria,
which are necessary for efficient CH selection.
In this paper we propose a technique in which we use
distributed algorithm of CH selection i.e. nodes themselves
decide whether to become CH or not. Fuzzy Technique for
Preference by similarity to Ideal Solution (Fuzzy-TOPSIS)
method is used to outrank the potential nodes as CHs. Fuzzy
logic and fuzzy set theory is applied to decision making
process. TOPSIS is a solution for multi-criteria optimization
problem. TOPSIS was initially proposed by Hwang and Yoon
[6]. Fuzzy-TOPSIS consists of decision matrix with malter-
natives and each alternative has nattributes. This technique is
applied is scientific and engineering problem solving. Fuzzy-
TOPSIS uses relative importance of attributes instead of using
precise values, because is some situations precise assignment
is not possible due to any reason. We consider four criteria for
CH selection which are residual energy, number of neighbors,
average distance from neighbors and distance between node
and BS. In our proposed scheme we avoid quick deformation
of CH, which in result reduces control overhead. Because of re-
duction in control overhead energy consumption is minimized.
In our proposed scheme we use realistic communication model
by introducing multi-hop communication model in both intra-
cluster (communication between normal nodes and CH) and
inter-cluster (communication between CH and BS).
Rest of the paper is organized as follows: Related work in
the area of clustering in WSNs is briefly discussed in Section
II. Section III describes the proposed scheme. In section IV,
simulation results are discussed in detial. Finally, Section V
concludes the manuscript.
II. RELATED WORK
Proper clustering and CH selection largely affects network
lifetime and stability in WSN. Lots of research has been
devoted in this regard and many clustering protocols have been
purposed. This section briefly reviews the previously proposed
clustering schemes.
In [7], authors proposed a Low-Energy Adaptive Clustering
Hierarchy (LEACH) protocol for WSN, which assumed that all
nodes in the network are homogeneous (i.e. having same initial
energy). BS or sink node provides a fixed probability value to
all nodes in the network and each node has equal chance to
elect itself as a cluster head. The CH responsibility is rotated
in every round among all nodes to balance the communication
overhead. LEACH operates in three phases. First phase is the
advertisement phase in which nodes are elected as CHs and
the CH status is broadcasted by each elected CH within their
neighborhood or transmission range. All the other nodes within
that range determine to which cluster they have to join and
become member of that cluster. In second phase, CHs perform
scheduling to properly carry out data communication in their
respective clusters. Member nodes only turn ON their radio
when they have data to send, which reduces energy utilization.
The third phase is the data transmission phase. When CHs
complete data collection from all member nodes within their
respective clusters, they perform data compression and send it
to the BS to save energy by avoiding multiple transmissions.
In result, the network lifetime and stability of LEACH is much
better than the flat routing protocols. To enhance performance
of LEACH, authors in [8] proposed Centralized LEACH (C-
LEACH) protocol, in which BS performs CH selection process.
This protocol performs better than LEACH, however, change
of CH in every round is not a feasible solution and increases
the signaling overhead. In [9], authors propose a clustering
algorithm that considers single criterion e.g. residual energy
of each node, to elect CHs.
All above mentioned protocols are based on single cri-
terion. As discussed previously that ideal CH is one that is
selected on multiple criteria. In [10] authors have proposed a
centralized clustering scheme that considers remaining energy,
node density or number of neighbors and distance from BS
and uses fuzzy-TOPSIS to select CHs. In that clustering
scheme, BS performs CH selection process (centralized). Due
to centralized scheme, nodes periodically send their data to
BS, and then BS decides the CHs. After selection of CHs, BS
broadcasts that information to all nodes in the network and
generates high control overhead. CHs broadcast advertisement
message and normal nodes send join that cluster as member
nodes by sending join request message. The CHs send ac-
knowledgment packets to confirm the membership. In their
proposed scheme, CH is changing in every round, which also
increases control overhead packets.
In order to overcome these deficiencies as much as possi-
ble, we propose a new distributed technique base on multiple
criteria. In our proposed technique, CHs are selected in a
distributive manner. To reduce the control overhead in our
proposed scheme CHs does not change periodically in every
round. We implement multi-hop communication between CHs
and BS and also between normal nodes and CHs, which
reduces energy consumption and results in a longer network
lifetime.
III. PROPOSED SCHEME
We propose a multiple criteria based distributed CH selec-
tion technique based on fuzzy-TOPSIS method. We improve
deficiencies previous fuzzy based CH selection technique.
Due to using distributed algorithm, nodes themselves take
decision to be selected as CH, hence nodes join CH with
maximum resources because all nodes have index value of
their neighboring nodes (which is a rank value obtained using
multiple criteria, final CH selection is based on this value).
We define a threshold value for change of CH, so in our
proposed scheme CHs are not changing in every round, due
to this, control overhead is much reduce as compared to
previous schemes. We consider four criteria including residual
energy, node density or number of nodes in neighborhood,
distance from BS and average distance between a node and its
neighbors. Our proposed scheme consists of three phases, i.e.
neighbor discovery, CH selection and cluster formation.
After updating packet form all neighbors, the nodes per-
form multiple criteria technique to calculate their respective
rank index, and share it with all neighbor nodes through Hello
packet. Following are steps to calculate rank index value based
on fuzzy-TOPSIS:
Step 1: The initial step of our proposed distributed cluster-
ing scheme is to perform neighbor discovery. Initially, all nodes
broadcast a Hello packet, which contains node’s ID, residual
energy, C1, node density C2, distance to the BS, C3, average
distance between this node and its neighbors, C4and location
information. Initially, C2and C4fields in the Hello packet
will be empty because each node has no information about
its neighbors. However, after sharing node ID and location
information with its neighbors, each node can easily compute
C2and C4and exchange it in the next Hello packet. All the
other nodes in the transmission range Trof that node, receive
Hello packet. After receiving Hello packet from all neighbors,
a node updates its neighborhood table (T) with neighboring
node’s ID, C1to C4as well as its own information. Suppose,
there are nneighbors of node k, then Tkwill be an (n+1)×4
matrix, as in eq.(1):
Tk=
C1C2C3C4
a1u(1,1) u(1,2) u(1,3) u(1,4)
a2u(2,1) u(2,2) u(2,3) u(2,4)
a3u(3,1) u(3,2) u(3,3) u(3,4)
: : : : :
a(n+1) u(n+1,1) u(n+1,2) u(n+1,3) u(n+1,4)
(1)
Step 2: It is evident that values of all criteria Cido not lie
in the similar range, e.g. range of values in C1is not similar
to the C2. Therefore, these criteria must be normalized to the
similar range to fairly select a CH. Note that there are some
criteria whose larger value is suitable for a node to be selected
as a CH e.g. C1and C2. These criteria are called Positive
criteria, Benefit criteria or Positive Ideal Solution (PIS) and
are normalized as in eq.(2). On the other hand, the criteria
with smaller value is appropriate for a node to be selected as
a CH, e.g. C3and C4. This type of criteria are called Negative
criteria, Cost criteria or Negative Ideal Solution (NIS) and are
normalized as in eq.(3).
U(i,j)=u(i,j)−min∀i(u(i ,j))
[max∀i(u(i,j))−min∀i(u(i,j))] (2)
U(i,j)=max∀i(uj)−u(i,j )
[max∀i(u(i,j))−min∀i(u(i,j ))] (3)
Each element of the Tk, is normalized using eq.(2) and (3)
and the normalized matrix at node k,Vk, will be:
Uk=
U(1,1) U(1,2) U(1,3) U(1,4)
U(2,1) U(2,2) U(2,3) U(2,4)
U(3,1) U(3,2) U(3,3) U(3,4)
::::
U(n+1,1) U(n+1,2) U(n+1,3) U(n+1,4)
(4)
Step 3: The preferences or weights are assigned to each
criterion. These, weights are application specific, however, for
our proposed scheme, the weights assigned to the selected
criteria are shown in Table I. We assign slightly higher weight
to residual energy, because it is the main constituent for CH
selection.
TABLE I. CRITERIA WEIGHTS
Criteria Weight
Residual Energy (w1) 0.4
Node density (w2) 0.2
Distance form Base Station (w3) 0.2
Avg. Distance between Neighbors (w4) 0.2
Step 4: The normalized values of fuzzy membership func-
tions are shown in Table II. These values are assigned ac-
cording to standard fuzzy model. These values are used to
determine preference classes for criteria values.
TABLE II. FUZZY MEMBERSHIP FU NCTIO NS
(VH) Very High (0.75, 0.88, 0.95, 1.00)
(H) High (0.55, 0.68, 0.75, 0.85)
(M) Medium (0.35, 0.48, 0.55, 0.65)
(L) Low (0.15, 0.28, 0.35, 0.45)
(VL) Very Low (0.00, 0.08, 0.15, 0.25)
Step 5: Calculate weighted decision matrix based on fuzzy
membership functions.
U=
U11 U12 U13 U14
U21 U22 U23 U24
U31 U32 U33 U34
: : ... :
U(n+1)1 U(n+1)2 U(n+1)3 U(n+1)4
(5)
Step 6: PIS and the NIS from weighted decision matrix, in
eq.(5), are computed as:
P IS = (U+
1, ..., U +
n) = [(maxiUij |i= 1, .., 4)andj = 1, ..., (n+ 1)]
(6)
NIS = (U−
1, ..., U −
n) = [(miniUij |i= 1, .., 4)andj = 1, ..., (n+ 1)]
(7)
Step 7: After computing the PIS and NIS, the separation
measures using n-dimensional Euclidean distance is calculated
as:
D+=
4
X
i=1
v
u
u
t
(n+1)
X
j=1
(Uij −U+
j)2(8)
D−=
4
X
i=1
v
u
u
t
(n+1)
X
j=1
(Uij −U−
j)2(9)
Step 8: Calculate Rank Index (RI) according to following
formula:
R.I =D−
D++D−(10)
In this manner, rank index value of a node and its neighbors
is computed and available at each node without transmitting
any extra information to the neighboring nodes. If a node has
highest rank index value within its RI table (potential node
within neighborhood), then it announces itself as a CH. On
the other hand, all the other nodes that have less rank index
value send join request to the node with higher value in RI
or to the neighboring node that announced itself as a CH.
After successful reception of the join request message, the
CH acknowledges to all requesting nodes and accept them as
members and forms the cluster. In this manner, whole network
is divided into clusters and each cluster is served by a potential
CH.
After successful clustering round, all member nodes in
clusters, start normal communication through their respective
CHs. Along with the normal communication, they also com-
pare their index value in their neighborhood table. If any
node’s index value is greater than index value of CH plus
the hysteresis parameter or threshold value (in our case 0.1),
then the CH will no more be eligible to act as CH, and nodes
will perform re-election process within the cluster only by
following the steps discussed above. The significance of using
this threshold value is to avoid CH change frequency or re-
election in every round. This process continues until there is
any surviving node in the network.
The multi-hoping communication model is considered by
our proposed scheme because it is the more realistic and
practical. The nodes within five meters range of CHs, send
their data directly to CH, however, the member nodes that
have more than five meters distance between itself and its
CH, communicate in multi-hoping fashion with the CH. Same
condition is also applied on CHs when they communicate
with the BS. The CHs within twenty meters range of BS,
communicate directly to BS, whereas the remaining CHs use
multi-hop communication mode via other CHs. The purpose
of using multi-hoping is to increase network stability and life
time. The intra-cluster and inter-cluster communication are
depicted in Fig. 1.
Fig. 1. Intra-cluster Communication
IV. SIMULATION ANALYSIS
The simulation analysis of the proposed scheme in contrast
with LEACH [5] and fuzzy based scheme [7] is discussed
in this section. In simulation, nnumber of sensor nodes are
randomly dispersed in a field of 100m∗100m. The BS is
located at corner of the WSN deployment field. Following
assumptions are used during these simulations. First, the sensor
nodes are continuously monitoring the environment and always
have data to be sent to the BS. Second, the wireless channel
is free from collision and interference. The last assumption
is that the sensor nodes are static throughout the simulation
period. Table III summarizes the simulation parameters. We
TABLE III. SIMULATION PARAMETERS
Parameter Value
Network Area 100m x 100m
Number of Nodes n100
Base Station Position (50,100)
Initial Energy 0.5 J
Range of Sensor Node 10m
Data Aggregation Energy 50pj/bit/report
Size of Data Packet 4000 bits
Size of Hello Packet 200 bits
Transmit Amplifier (Eamp) 100 pJ / bit / m2
Receiver (Rx) Electronics (EelecRx) 50 nJ / bit
Transmitter (Tx) Electronics (EelectTx) 50 nJ / bit
evaluate the stability of the network by examining the numbers
of rounds until first node dies. Following graphs show simu-
lation results of our purposed scheme compared with previous
schemes.
Network stability and lifetime of proposed technique are
shown in Fig. 2. It is clear from graph that in LEACH, first
node dies around 170 rounds and previous fuzzy model first
node dies around 530 rounds, where as our proposed scheme
first node dies 1600 rounds. Similarly the network lifetime for
LEACH goes to around 1000 rounds, previous fuzzy based
model goes to 1100 rounds, where as our proposed scheme
die out at 2400 rounds. Hence network lifetime and stability
in our purposed scheme is much better than previous clustering
techniques the Reason for is that LEACH is a single criteria
based technique, whereas previous fuzzy-TOPSIS method, BS
0 500 1000 1500 2000 2500 3000
0
10
20
30
40
50
60
70
80
90
100
Number of Rounds
Number of Dead Nodes
LEACH
Previous Fuzzy
Proposed
Fig. 2. Network Stability and Lifetime
is performing CH selection process, which does not depend
on geographical conditions of nodes, whereas in our purposed
scheme every node itself take decision for CH, considering the
knowledge of neighbor nodes.
0 500 1000 1500 2000 2500 3000
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Number of Rounds
Total Energy consumption per round
LEACH
Previous Fuzzy
Proposed
Fig. 3. Energy consumption per round
Energy consumption of network per round is shown in
Fig.4. It evident from the graph that the proposed scheme
consumes very less battery power than previous schemes,
because of efficient CH selection. In previous schemes, energy
variation at different points is observed and it due to re-election
of CHs in each round. This CH re-election is minimized in the
proposed scheme by introducing the CH changing threshold or
hysteresis parameter and that is the main reason of less energy
consumption in our proposed scheme.
0 500 1000 1500 2000 2500 3000
0
2
4
6
8
10
12
14
16
18
20
Number of Rounds
Number of CHs per Round
LEACH
Previous Fuzzy
Proposed
Fig. 4. Total No. of Cluster Heads per round
In previous scheme CHs are changing in every round that
0 500 1000 1500 2000 2500 3000
0
2
4
6
8
10
12
14
Number of Rounds
Number of CHs changes Per Round
Previous Fuzzy
Proposed
Fig. 5. Cluster head change per round
0 500 1000 1500 2000 2500 3000
0
1
2
3
4
5
6
7
8
9x 104
Number of Rounds
Total number of Hello packets
LEACH
Previous Fuzzy
Proposed
Fig. 6. Number of hello packets
results in higher control overhead. In our proposed scheme
we introduced the hysteresis parameter and the CH will only
change when index value of that CH plus the threshold value
is less than the RI of any of the member node within that
cluster. Less variations in CH changes or cluster deformation,
results in less control overhead and that is depicted in Fig. 5
and 6.
0 500 1000 1500 2000 2500 3000
0
0.5
1
1.5
2
2.5
3
3.5
4x 104
Number of Rounds
Number of packets to BS
LEACH
Previous Fuzzy
Proposed
Fig. 7. Number of packets to base station
Total number of packets to the BS is shown in Fig. 7. Due
to proper CH selection and communication model, network
throughput of our proposed scheme is much greater than
previous schemes.
V. CONCLUSION
A multi-criteria and fuzzy TOPSIS based distributed clus-
tering technique is proposed in this work. Multiple criteria i.e.
residual energy, number of neighbors or node degree, distance
between a node and BS, and average distance between a
node to its neighbors, are considered to select a cluster head.
Our proposed scheme considers more practical communica-
tion model that is multi-hop communication in both inter-
cluster and intra-cluster communications. Cluster deformation
frequency is minimized by introducing the hysteresis param-
eter or threshold value. Simulation results show that multiple
normalized criteria for a CH selection improves throughput,
consumes less energy, minimum variations in CH re-elections
frequency (cluster stability), network lifetime, and very less
control overhead, compared to the previous clustering schemes.
REFERENCES
[1] T. Shah, N. Javaid, and T. N. Qureshi “Energy Efficient Sleep Awake
Aware (EESAA) Intelligent Sensor Network Routing Protocol.” In Proc.
of 15th IEEE International Multi Topic Conference (INMIC12), 2012.
[2] A. Hergenroder, and J. Horneber, “Facing challenges in evaluation
of WSN energy efficiency with distributed energy measurements,” In
Proc. of 7th IEEE International Wireless Communications and Mobile
Computing Conference (IWCMC), 2011
[3] S.D. Muruganathan, D.C.F. Ma, R.I. Bhasin, and A. Fapojuwo “A
Centralized Energy-Efficient Routing Protocol for Wireless Sensor Net-
works” IEEE Communications Magazine, Vol. 43, Iss. 3, 2005, pp.S8-13.
[4] I. F. Akyildiz “Wireless Sensor Networks: A Survey,” Elsevier Sci. B.V.
Comp. Networks, Mar. 2002 Communications 29, 2006, pp 2230- 2237.
[5] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energy-
efficient routing protocols for wireless microsensor networks,” In Proc.
33rd Hawaii Int. Conf. System Sciences(HICSS), Maui, HI, Jan. 2000.
[6] S. Opricovic and G.H. Tzeng, “Compromise solution by MCDM meth-
ods: A comparative analysis of VIKOR and TOPSIS.” European Journal
of Operational Research, Vol. 156, Iss. 2, 2004, pp. 514529.
[7] W.R. Heinzelman, A.P. Chandrakasan, H. Balakrishnan, “Energy-efficient
communication protocol for wireless microsensor networks,” Proceedings
of the 33rd Annual Hawaii International Conference on System Sciences,
2000, vol.2, pp. 4-7, Jan. 2000
[8] W.R. Heinzelman, A.P. Chandrakasan, and H. Balakrishnan, “An
application-specific protocol architecture for wireless microsensor net-
works,” IEEE Transactions on Wireless Communications, Vol.1, No.4,
pp.660-670, Oct 2002.
[9] O. Younis and S. Fahmy, “HEED: a hybrid, energy-efficient, distributed
clustering approach for ad hoc sensor networks,” IEEE Transactions on
Mobile Computing, Vol.3, No.4, pp.366-379, Oct.-Dec. 2004.
