An Agent Based Clustering Protocol for Energy

Efficient Wireless Sensor Network

Kumar Vivek, Ashok Kumar, Nilesh D. Makwana
Electronics & Communication Engineering Department
National Institute of Technology, Hamirpur
Himachal Pradesh, India
krvivekcsn@gmail.com, ashok@nitham.ac.in, ndmnith@gmail.com
Abstract Limited power resource is one of the major
constraints for wireless sensor networks (WSNs). As the
network life time is directly depends on power efficiency of
a network, so the prolonging the network life time is one of
the most important challenge for research community
from the beginning of this technology. Various techniques
have already proposed for increasing power efficiency of
WSNs based on duty cycling, data driven techniques, or
mobility based approaches. In this paper we have
proposed an Agent Based clustering Protocol (ABCP) for
WSNs which will not only enhance the energy efficiency of
the network but also increases the overall throughput of
the network as compared to other clustering-based
schemes such like Low Energy Adaptive Clustering
Hierarchy (LEACH). Simulation results show that ABCP
is better in terms of energy efficiency and throughput as
over its comparative.
Index Terms Duty cycling, Data driven, Mobility,
Throughput.
I. INTRODUCTION
Typically wireless sensor network composed of large
number of low cost sensor node having limited power, low-
communication distance (as energy is limited), limited
processing capability and small in size are deployed in a region
to sense the data like temperature, humidity, pressure etc. from
an area of interest and send them to sink by means of single
hop and/or multi hop technique. Sink is nothing but the
ultimate destination of data sensed by any node. The way of
communication between sink and a sensor node is depends on
the routing techniques, there are several routing protocol for
WSNs [1, 2, 3]. If there is a cluster head present the sensor
node will send the data to cluster head and then cluster head
can send the data to sink, or in other hand sensor node may
communicate directly to sink node in some scenario. In some
cases the sensor node may processed the some of the sensed
data and/or cluster head can process the data too, as the energy
availability is limited, so unnecessary transmission can be
avoided by processing the sensed data. For example consider
the case of a network where node is mean to be implemented
for sensing the temperature as temperature can increase or
decrease gradually or may remain constant. So transmitting the
same temperature information or small difference in
temperature periodically may not desirable because by doing
this the senor node may exhaust their energy, leading to
decreased life time of the network. So sensor node or/and
cluster head may process the data and see if there is significant
change in temperature then it will send the data to destination
otherwise it remain inactive with respect to transmission.
By use of clustering we enable a sensor network to work in
more efficient way in terms of energy efficiency and thus
leading to increased life time of the network [4]. In this case
the main function of a cluster head is to act as a link between a
sensor node and the sink, i.e. all node of a cluster first sense the
data and send it to cluster head and cluster head forwarded the
data to sink with or without data processing as requires. In this
case a cluster head issues a time division multiple accesses
(TDMA) for its member node to avoid collision as discussed in
[3]. For case of homogeneous network where each node is
same with respect to power source, processing capacity,
memory etc. and cluster is formed by considering some
definite number of nodes together and a cluster head is selected
among them. All other node of the cluster will communicate
with the cluster head and cluster head is responsible for
sending the all data gather from the nodes of the cluster to sink
then due to more data transmission by cluster head it will soon
exhaust its energy. This is a major problem where the cluster
head is fixed. This problem is largely studied by research
community and accordingly there are various protocols to
solve this particular problem. In LEACH [5] protocol the
author came to a solution in which the each time a cluster head
are selected among the member of a cluster by use of priori
optimal probability. This approach provides the guarantee that
the energy load is well distributed among all the nodes of a
cluster by changing the cluster head role uniformly among all
the nodes, leading to almost same residual energy of a node
after a long run of time. Although LEACH performs well in
homogenous network however this protocol is not considered
good for heterogeneous networks as shown in [6]. In this paper
we are proposing a protocol for enhancing network life time by
using heterogeneous clustering approach using some agent
node. Particularly there are number of WSNs applications
which can be benefited by this technique. We will deploy some
agent node with the normal sensor node in the network field.
We are assuming that the agent node is not limited by power
(or at least the energy of agent node is much more than the
normal sensor node) i.e. either there is some energy harvesting
techniques are available or the battery are replaceable.
The rest of this paper is organized as follows, in Section II
we have discuss the surveys and related work with respect to
proposed routing protocol. Section III will give the energy and
network model for proposed protocol. In Section IV we will
discuss our proposed model. Simulation and results are
provided in Section V. At last in Section VI we will conclude
our paper.
II. RELATED WORKS
We have already discussed that in wireless sensor network
it is desirable that a routing design needs to guarantee the
delivering of data from sensor node to sink in energy efficient
way because energy consumption and network life time are the
most important features in designing of the wireless sensor
network. There are several clustering based protocols are
available.
LEACH [5] is a clustering based routing protocol where the
overall network field is divided into number of cluster and in
each cluster head is elected based on optimal priori based
probability by doing this the energy loads of the cluster head
can evenly distributed among all the nodes of a cluster, and for
long run the residual energy of all the sensor nodes of a cluster
will be almost equal, leading to larger lifetime of the network.
It is well suited for homogeneous network but not for
heterogeneous network as LEACH depends on spatial density
of the network [6].
TL-LEACH [7] the author proposed a two levels Hierarchy
for low energy adaptive clustering hierarchy. There are two
level of cluster heads present primary cluster heads and
secondary cluster heads. Secondary cluster head connect with
their corresponding sensor nodes of the cluster they are
associated with and primary cluster head is cluster head of the
cluster formed by secondary cluster heads. This mechanism
provides the well distributed load of the network to each sensor
node.
In PEGASIS [8] the author proposed a chain based routing
protocol by forming chain form each node or by base station
for sending data from sensor node to sink. Each node only
communicate with its close neighbor and turn wise transmit to
base station thus reducing the amount of energy spent each
round. But this protocol requires the global knowledge of the
network topology, makes this protocol difficult to implement.
The SEP [9] is a two level heterogeneous network which
has two types of node, the advance node and the normal node.
The rotating epoch and election probability is depends on the
initial energy of the nodes. But SEP is not good for multilevel
heterogeneous network, when this type of heterogeneity is due
to the network operation of the WSN.
DEEC [10] is another protocol which is used for energy
saving in heterogeneous network using distributed energy
efficient algorithm, in which the cluster head election
probability is based on the ration of residual energy of the node
and the average energy of the network.
In EEUC [11] the author stress the fact that as the cluster
heads cooperate with each other to send the data to the sink so
the cluster head closer to the sink has more burden that the
cluster head situated far away from the sink as the nearer
cluster have to forwarded the more data coming from other
cluster head. So the nearer cluster head may exhaust their
energy soon and die. The author came to solution in which he
divide the cluster in unequal size i.e. cluster closer to sink have
small size than the cluster farther to sink. By doing so the
protocol balances the energy consumption well among all the
sensor nodes.
III. RADIOENERGYMODEL ANDNETWORKMODEL
In this section we will discuss energy model used for our
proposed protocol and also the network model of proposed
protocol.
A. Radio Energy Model
We are considering a very popular radio energy model [5, 9,
12] for our proposed protocol. For reception and transmission
of a bit there is always energy expenditure associated with it
also if we are using sleep mode scheduling for a particular
model then the turning on and turning off of transmitter and
receiver also requires some energy also processing of data also
requires energy. Here we are considering the energy used for
reception and transmission of data as major chunk of energy is
utilized in this only. According to energy model proposed in
Fig. 1 in order to maintain an acceptable Signal-to-Noise
(SNR) in transmitting N-bit message over a distance of d, the
energy require by the transmitter is given by:
, )
2
N.E +N.E .d ifd<d
elec fs 0
E N,d =
T
x
4
N.E +N.E .d ifd d
elec amp 0

>

(1)
Here E
elec
is the energy loss per bit to run the transmitter or
receiver circuit and it depends upon factor like coding
technique used, modulation, filtering and spreading of the
channel.
fs
E and
amp
E depends upon transmitter model
used. d is distance between transmitter and receiver. In this
model we have consider both model of energy consumption
describe in the above equations.
Now equating the above to equation for d d
0
we get
E
fs
=
0
d
E
amp
(2)
To receive bit message receiver requires energy of
E =N.E
Rx elec
(3)
Figure 1.Radio Energy model
If the packet size of massage is of bits then residual energy
( E
Residual
) of a node after one round of transmission can be
found by the following equation
, )
, )
2
E -{E +E +E d }B ifd<d
0 T DA fs 0
X
E =
Residual
4
E -{E +E +E d }B ifd d
0 T DA amp 0
X

>

(4)
Where,
0
E is the initial energy provided to each sensor node.
E E E
,
T f
,E ,
s
x A
amp
D
is per bit transmission, compression,
free space loss and multipath loss respectively.
B. Network Model
In this section we will describe ABCP network model for
wireless sensor network which is composed of some agent
node i.e. heterogeneous network with respect to the initial
amount of energy. We are deploying sensor node in an area of
interest with some agent node as shown in Fig. 2.
Let us consider we are deploying N number of sensor nodes
in MM region as shown in Fig. 2. Where, M denotes the
length and breadth of the network field. Let us say node is
represented by S
i
= {S
1
, S
2
, S
3
, S
4
.S
N
}. Also we are
deploying P number agent nodes at fixed places, which will act
as local sink for the some of the nodes let us say agent nodes
are represented as A
j
= {A
1
, A
2
, A
3
, A
P
}. We are taking
negligible number of agent node as compared to sensor node.
I.e. P << N. We will place the agent nodes in network field in
such a way so that sink can create some logical area by
dividing the network field in smaller sub-field, which will
efficient in terms of energy saving. That sink is associating
maximum number of nodes which can directly send its sensed
data to an agent node or sink.
The assumptions used in the proposed routing techniques
are as follows:
- Sensor nodes are randomly deployed in our
network field and it is using uniform random
distribution, and each of the sensor nodes has same
computational power as well as energy.
- The Sink is located outside of our network field and
we know the co-ordinate of it.
- Agent nodes are rechargeable and the placement of
agent node is not random (we will fix the coordinate
of agent nodes) i.e. either the node is deployed
manually or the agent node is movable.
- Each sensor node assigned a unique ID.
IV. AGENT BASED CLUSTERING PROTOCOL
We will discuss our proposed routing models in this
section. As sensor node has to send the sensed data to sink. If
the sensor node send the data to directly to sink or with help of
multi hop technique the sensor node may exhaust their energy
soon as they have to send their own data as well as they have to
work as transit node too for the data forwarding for data of
other nodes. Although clustering technique like LEACH is
good and able to increase the life time of network considerably
but with the help of our proposed model we will able to
increase the life time of network almost double as compared to
LEACH. For increasing the network life time and throughput
we deploy some agent node at some fixed location in the
network field such that a virtual maximum area network sub-
field can created such that all the nodes within that sub-field
can communicate with an agent node directly as well some
sensor node can communicate with the sink directly while
considering energy efficiency. The function of agent nodes is
to collect data from the nodes near to it and can also collect the
data from the cluster heads of the other cluster forming by the
nodes which have not any agent node nearby. The agent node
work is forwarding the overall aggregated data collected by
other cluster heads and the data gathered from the sensor node
associated with it to the sink. Our result will reflect the increase
life time and throughput with the expense of addition of agent
nodes because adding rechargeable agent node is cheaper than
the price of sensor node itself. We can broadly identify our
proposed network implementation as following steps:
A. Initial Phase
We will use homogeneous sensor node with respect to
energy and deployed it in our network field in uniformly
random distributed way. The sink broadcast a hello packet; in
response of this packet each sensor node will forward its
location and other information to the sink. Now the sink has
sensor node ID, energy information of the node, and
information about sensor node euclidean distance from itself
and also from agent nodes.
Figure 2.Network field
B. Middle Phase
Now as sink has required information for dividing the
whole network field into logical region based on location of
sensor node in the network field. Sink now divided the whole
region in three type of region the first type of region will
directly send its gathered data to sink, the second type of region
will communicate directly with agent nodes and the third type
of region will not associated with any agent node or sink
directly but they are divided into number of clusters and they
choose their cluster head depending upon priori based
probability, and the cluster head can communicate and send
their data to agent node or sink node directly i.e. via single hop
or multi hop as required.
C. Final Phase
Now sink divided the network field into three types of
logical region. The node residing in first type of region, which
will be close to sink itself, can send their sensed data to sink
directly leading to energy saving. The node residing in second
type of region i.e. region closed to any agent node will forward
their data directly to an agent node which leads to energy
saving. The third type of area which is not associated with any
agent node or sink directly, is divided in to small regions
(cluster) by sink. So the cluster formed by third type region
needs a cluster head for energy efficient communication which
is chosen by LEACH protocol. Each node of particular cluster
has probability of p to act as cluster head, i.e. each node will
elect as cluster head at least once after 1/p rounds as proposed
in LEACH. The
1/p
round is referred as epoch. The 1/p round
is referred as epoch. Initially all the nodes of third type of
region has equal energy level and has equal chance to became
cluster head and the probability of being so is p. On an average
(NK).p of cluster head present in the network field apart from
agent nodes, Where K is number of sensor nodes which are
associated directly agent nodes or sink. Nodes that are elected
to become cluster head in the current round cannot become
cluster head in same epoch. At the start of each round a node Si
from the third type of region autonomously choose a number
between 0 and 1. If chosen number for node Si is less than a
predefined threshold T(S) then that node will be act as cluster
head for that round.
p
ifS C
1-p(rmod(1/p)) T(S)=
0otherwise

(5)
Where, r is the current round number starting from round 0.
C is the set of nodes which have not been cluster head in the
last 1/p rounds. Using this threshold level each node of a
particular cluster head have a chance to became cluster head
once in each round.
All sensor nodes transmit their data to cluster head in
steady state phase. The cluster head collect the data from
member nodes and send it to nearest agent node. Agent node
received the data processed it and send it to sink.
V. SIMULATION AND RESULTS
We have simulated our setup on MATLAB and check
performance of the protocol and compare it with one of
standard routing protocol of WSNs LEACH.
A. Performance Parameters
In this subsection we have consider the performance
parameter for our simulation. We have check and compare our
protocol based on following performance parameters:
1) Network Lifetime: Here we have considered the network
lifetime here as number of round till the last node of the
network died.
2) Throughput: It is one of the performance analyses for our
proposed protocol where we check the total number of packet
received by the sink.
3) Residual Energy: We will check the total amount of
energy left in the network. Because on this parameters our
network lifetime depends.
B. Simulation Parameters
In order to analyze and compare the above mentioned
performance parameters with LEACH protocol with ABCP we
consider that there are 100 numbers of sensor nodes which are
deployed randomly in network filed of area
Four agent node is deployed at the
corner of a square sub-field of the
consider network field, and considered sub-filed diagonals are
whose diagonals bisecting each other at the same point where
the original network field diagonals are bisecting each other.
The sink is not situated in network field and sink and agent
node are stationary after deployment. Table 1 contains the
other simulation parameters used for performance analysis of
our proposed protocol.
C. Simulation Result and Analysis
In this subsection we will discuss the simulation result
obtained and compare it with leach and see how ABCP turn out
to be better compared to LEACH with respect to residual
energy, network lifetime and overall throughput at sink.
1) Residual Energy: We will analyze and compare the total
energy remaining in the network field after each round
assuming that each sensor node has equal energy of 0.5 J.
Since, the network has total 100 sensor nodes, so total energy
of the network will be 50 J, shown in Fig. 3.
Table 1.Simulation parameters
Parameters Value
0.5 J
10 pJ/bit
0.0013 pJ/bit
87.70 m
50 nJ/bit
5 nJ/bit
2000 bits
Figure 3.Number of rounds versus residual energy
Figure 3. shows the number of round versus residual energy
plot of the proposed protocol and its comparison with LEACH
protocol.
We can see from the Fig. 3 that by using our proposed protocol
the network has almost 8 J of energy remaining with it when
by using LEACH protocol the network is out of energy,
i.e.16%of total energy of the network is still there when all the
sensor node using LEACH protocol exhausted their energies.
2) Network Lifetime: We can compare the overall network
lifetime of network as the total number of rounds till the last
node of the network is alive. Figure 4 shows the number of
rounds and percentage of alive nodes at that time.
We found that after 3000 round there is no alive node in in
the network filed where the nodes are communicating using
LEACH protocol whereas by using our proposed technique
network can extend in terms of alive node till approximately
5000 rounds which is approximately 66 % more than LEACH.
I.e. we are significantly increasing network lifetime using
proposed routing protocol. Also we can see that there is sudden
increase in the number of dead nodes after 4000 round it is due
to fact that node communicating directly with agent nodes will
exhaust their energies almost at the same time.
3) Throughput: Here we will see the overall throughput of
the network i.e. the total number of packets received by sink.
Figure 5 shows that at the end of network lifetime the total
number of packets received by sink using proposed protocol is
almost 5 times more than the total number of packets received
by sink using LEACH protocol.
Figure 4.Number of round versus percentage of alive nodes
Figure 5.Total number of packets received by sink
VI. CONCLUSION
In this article we have proposed an agent based protocol for
energy efficient WSN. As our result shows that the proposed
protocol is efficient in terms of lifetime, energy efficiency and
overall throughput. But flip coin of the side is that we are
deploying agent node in the network field which has better
energy resource or have energy harvesting technique with it,
also we have to manually deployed the agent node or we have
to take mote as agent node, which is costlier than general
sensor nodes but we know that providing better energy source
to a node is cheaper than sensor node itself. So we can use our
proposed routing protocol for better energy efficiency and high
throughput in expense of a little costlier deployment cost.
REFERENCES
[1] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, A
survey on sensor networks, IEEE Communications Magazine,
vol. 40, no. 8, pp. 102114, August 2002.
[2] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci,
Wireless sensor networks: a survey, Elsevier Computer
Network, vol. 40, no. 8, pp. 393422, 2002.
[3] Ahmed A. Ahmed, Hongchi Shi, and Yi Shang , A survey on
network protocols for wireless sensor networks , IEEE, pp.
301305, 2003.
[4] Jamal N. Al-Karaki, Ahmed E. Kamal, Routing Techniques in
Wireless Sensor Networks: A Survey, IEEE Wireless
Communications, Volume: 11, Issue: 6 , 26-28, December
2004.
[5] W.Heinzelman, A. Chandrakasan, and H. Balakrishnan,
"Energy-efficient communication protocol for wireless sensor
networks," in the Proceeding of the Hawaii International
Conference System Sciences, Hawaii, January 2000.
[6] Smaragdakis, Georgios, Ibrahim Matta, and Azer Bestavros.
SEP: A stable election protocol for clustered heterogeneous
wireless sensor networks. Boston University Computer Science
Department, 2004.
[7] Loscri, V., G. Morabito, and S. Marano. A two-levels hierarchy
for low-energy adaptive clustering hierarchy (TL-LEACH).
IEEE Vehicular Technology Conference. Vol. 62. No. 3. IEEE;
1999, 2005.
[8] Lindsey, Stephanie, and Cauligi S. Raghavendra. PEGASIS:
Power- efcient gathering in sensor information systems.
Aerospace conference proceedings, 2002. IEEE. Vol. 3. IEEE,
2002.
[9] G. Smaragdakis, I. Matta, A. Bestavros, SEP: A Stable Election
Protocol for clustered heterogeneous wireless sensor networks,
in: Second International Workshop on Sensor and Actor
Network Protocols and Applications (SANPA 2004), 2004.
[10] Li Qing , Qingxin Zhu, Mingwen Wang, Design of a
distributed energy-ecient clustering algorithm for
heterogeneous wireless sensor networks, Elsevier Computer
Communication, pp. 2230-2237, 2006.
[11] Chengfa Li; Mao Ye; Guihai Chen; Jie Wu, "An energy-
efficient unequal clustering mechanism for wireless sensor
networks," Mobile Adhoc and Sensor Systems Conference,
2005. IEEE International Conference on , vol., no., pp.8 pp.,604,
7-7 Nov. 2005
[12] W. Hcinzelman, A. Chandra!wan, and H.
Ealakrirhnan, "An application-specific protocol
architeenut for wireless microsensor networks," IEEE
Tramctiom on Wireless Commicalions, vol. 1, No. 4, pp.
660-670, October 2002.