Anda di halaman 1dari 5

2014 International Conference on Control, Instrumentation, Energy & Communication(CIEC) 508

Leader Selection in Wireless Sensor Networks —


An Energy Efficient Approach

Joydeep Banerjee & Mrinal Kanti Naskar Utpal Biswas Omar Alfandi & Dieter Hogrefe
Department of Electronics Department of Computer Institute of Informatics
& Telecommunication Engineering Science & Engg SensorLab, Telematics Group
Jadavpur University University of Kalyani Georg-August-Universitt Göttingen
Kolkata 700037 Nadia, West Bengal 741235 Göttingen, Germany
Email: iamjoydeepbanerjee@gmail.com & Email: utpal01in@yahoo.co.in Email: alfandi@cs.uni-goettingen.de &
mrinalnaskar@yahoo.co.in hogrefe@informatik.uni-goettingen.de

Abstract—Wireless Sensor Networks have found its applica- In this paper we propose a leader selection algorithm
tion in numerous domains. Limited energy of the sensor nodes is for routing protocols. Parameters to judge the potential of a
the main constraint in operation of the network. Several energy node to be a leader is analyzed to devise an energy efficient
efficient routing protocols have been proposed to deal with this leader selection algorithm. The main parameters taken into
problem. In this paper we propose a leader selection algorithm consideration are distance of the node from base station,
for these routing protocols to enhance the network lifetime to a
degree of connectivity of the node and trust level of the node.
greater extent. Distance of the node from base station, degree of
connectivity of the node and trust level of the node is individually These parameters are coupled with remaining energy of the
analyzed and is linearly combined with weights associated with node for load balancing among leaders. The efficiency of the
each parameter. This gives the overall potential function of a algorithm is compared with traditional energy-distance based
node to become a leader. The number of rounds after which leader selection made in routing protocols. The metric used
leader selection takes place is also equated. Through simulation for this comparison is the number of rounds of successful data
the algorithm is compared with traditional energy-distance based transfer (i.e. if a selfish node is chosen as a leader and sends
method for routing protocols LBEERA, OREC, PEGASIS and ambiguous packets to base station then it is not considered as
SHORT. Significant improvement of network lifetime is obtained a round of data transfer but energy of the network decreases)
using this leader selection algorithm. Highest improvement of to the base station. Though this algorithm is mainly focused
more than 8% increase in network lifetime is observed when
distance of the node from the base station is given a higher
on centralized data transfer but it can also be extended to
linear combination weight. distributed data transfer like LBEERA [5], OREC [11] with
more than one leader in the network. The major contribution
Keywords—Wireless Sensor Network, Leader Selection, Network of this paper are listed below—-
Lifetime, Energy Efficient.

I. I NTRODUCTION • Potential of a node in the network to become a leader


is first analyzed for each individual parameters in
Wireless Sensor Network (WSN) finds its application in separate. Mathematical expressions are derived which
numerous domains [1]. It consists of sensor nodes deployed gives potential of a node to be a leader with respect
in a region following a topology and defined connectivity. to distance of the node from base station, degree of
The main constraint that WSN faces is the limited battery connectivity of the node and trust level of the node.
capacity (energy) which cannot be replaced in most applica-
tions. Several energy efficient clustering algorithma [2] and • The parameters are superposed linearly to get the
load balancing routing protocols [3], [4], [5], [11] have been overall potential of a node to become a leader. The
proposed to overcome this issue. algorithm follows from this which described the whole
Choosing leaders in routing protocol and cluster heads in leader selection protocol framed.
clustering algorithms for routing the data to the base station
is crucial for increasing the network lifetime. There can be • The algorithm is compared with energy-distance based
significant decrease of energy efficiency if the leader selection leader selection algorithm for different routing algo-
is not done appropriately. Efficient cluster head selection has rithms. This provides the efficiency of the proposed
been extensively studied in various research articles [6], [7], routing algorithm.
[8], [9], [10]. The techniques used are mainly fuzzy based
and can be applied to clustering algorithms like LEACH [2].
However, its applicability to routing protocols like PEGASIS The paper is divided as follows — Section 2 gives the re-
[4], SHORT [3], LBEERA[5], OREC [11] etc. has not been lated work, Section 3 introduces the design of leader selection
looked upon. Trust level of a sensor node to identify selfish protocol with respect to different parameters, Section 4 frames
behavior of a node should also be taken into consideration in the algorithm based on the design in Section 3, Section 5 gives
developing a leader selection algorithm. the simulation results and we conclude in Section 6.

978-1-4799-2044-0/14/$31.00©2014IEEE
2014 International Conference on Control, Instrumentation, Energy & Communication(CIEC) 509

II. R ELATED W ORK To derive the dependence metric for this parameter we
consider a two dimensional field with nodes represented as
Cluster head selection in clustering algorithms for in- points relative to axis X and Y . The axis can be drawn in
creasing energy efficiency in WSN has been researched quite any fashion as the metric is independent on the position or
deeply. In [6] the authors proposed a fuzzy logic for cluster orientation of the axis. We define a metric Pld (i, t) as the
head selection to reduce dependence on energy and distance potential of a node i to become a leader at any round t with
overheads in large sensor networks. Fuzzy approach to select respect to its distance from the base station as —
cluster heads using node energy, location of a node and quality
of service had been done in [7] and with distance from base 1
Pld (i, t) =
station and sum of distance between nodes in [8]. Authors (xi − x1 − λ1 x2 )2 + (yi − y1 − λ2 y2 )2 + k 2
in [9] discussed about distributed cluster head election for (1)
heterogeneous network. [10] also extended the fuzzy based
approach in cluster head selection with residual energy of the Where (x1 , y1 ) is the co-ordinate of the base station,
node, number of neighbors and distance from the base station. (x2 , y2 ) is the co-ordinate of the center most point in the
topology (the center most point is determined by joining the
Though existing research work considers the parameters nodes towards the end of the topology to get an approximate
residual energy, distance of the node from base station and boundary and finding the center most point in structure defined
number of neighbors yet they have not been extensively by the boundary) and (xi , yi ) is the co-ordinate of the ith node
analyzed individually. Moreover research work in leader se- in the network with i = 1, 2, ...N (where N is the total number
lection in energy efficient routing protocol is not prominent of nodes in the network). The weights λ1 and λ2 on the co-
in literature as most works considers cluster head selection for ordinate (x2 , y2 ) is determined by putting the individual terms
clustering protocols like LEACH [2]. Our work deals with this (xi − x1 − λ1 x2 )2 and (yi − y1 − λ2 y2 )2 to 0. This is done
problem and also includes the notion of trust in designing the as when a node in the topology (if present) corresponds to the
leader selection algorithm. center most point in the topology then the metric Pld should
be maximum for that node and would be equal to k 2 . This
gives λ1 = x2x−x 1
and λ2 = y2y−y 1
. The value of k is taken as
III. D ESIGN OF L EADER S ELECTION P ROTOCOL  2 2
2 2
(x2 − x1 ) + (y2 − y1 ) which is the distance between the
We identify three major parameters in designing the pro- center most point in the topology and the base station. So the
tocol as listed below. maximum value of the metric is given as Pldmax = k12 .
• Distance of the node from the sink The parameter Pld (i, t) is normalized with respect to Pld
of all other nodes in the network. This normalized parameter
• Degree of the node
is termed as ωD (i, t) and is expressed as —
• Trust level of the node
Pld (i, t)
ωD (i, t) = N (2)
j=1 Pld (j, t)
To frame the algorithm of the protocol we develop the
dependency of leader selection on the above mentioned pa-
rameters individually.

A. Distance of node from the sink


The energy dissipated by a leader is approximated to be
inversely proportional to the square of distance of leader node
from base station [4], [3], [5], [11]. For arriving to a more
accurate relation we perform simulation in MATLAB taking
a square grid topology (as in Figure 1). The experimental
setup is built with SHORT protocol [3] used for data routing
with each node in the network forced to be a leader without
any constraint for 100 rounds of data transfer. The energy
dissipated by the whole network is calculated for each node
being a leader. The total number of nodes in the square grid
topology are 49 with grid space of 10m. The distance between
sink node and nearest node from the sink is also taken as 10m.
The energy dissipated in routing of packets is calculated using
the power control algorithm proposed in [11]. Figure 2 gives
the variation of the total energy dissipated per node by the
network with distance of the leader node from the sink. A
close investigation of Figure 2 yields that nodes with distance
midway between the closest and farthest node in the network
from base station has more energy efficient data transfer when
selected as the leader node. This essentially disproves the
Fig. 1: Square grid topology and its connectivity
inverse square relationship of distance of node from sink with
its potential to be a leader.
2014 International Conference on Control, Instrumentation, Energy & Communication(CIEC) 510

TABLE I: average energy consumption per node in the network


verses degree of the node
κ(i, t) Average Energy Consumption (J)
1 49.7
2 43.6
3 44.3
4 48.3

of 3d
2 is considered. So, using this definition, the value of κi, t
for any node is restricted to either 2, 4 or 6. For each node in
the topology the average energy consumption of the network
for 100 rounds of data transfer to base station via the selected
node as leader is calculated. Power calculations is done using
[11] and SHORT [3] protocol is used for routing. Table 1 gives
Fig. 2: Energy consumed per node vs distance of leader node the node ids corresponding to the degree κ of the node and the
from base station plot average energy consumption of the network for those nodes in
the network.
Based on the observations from the experiment we define
B. Degree of connectivity of the node a relationship between the node degree of a node i κ(i, t) and
its potential to be a leader Pln (i, t) as
The instantaneous degree of connectivity a node κ(i, t) also
dictates the energy efficiency of the network in leader selection.
The energy dissipated by a node depends on the size of data 1
Pln (i, t) = (max(κ(i,t))+min(κ(i,t))
packet it receives. So a leader node with high degree would (κ(i, t) − max(κ(i,t))+min(κ(i,t)) 2
) + )
2 2
decrease the energy of the node drastically after each round (3)
of data transfer due to large amount of energy dissipated for
receiving large number of data packets. Moreover, for a leader The equation 3 is normalized with respect to the sum of Pln
node with low degree would increase the energy consumption values in the network. The dependence of degree parameter
in the network as for some packets requiring to be transmitted denoted by ωN (i, t) is given by the equation 4.
through a greater distance(more than or equal to two hops)
which would otherwise be directly transmitted to the leader Pln (i, t)
node. ωN (i, t) = N (4)
i=1 Pln (i, t)

The potential parameter for degree connectivity of the


node should allocate higher potential for nodes which have
connectivity as arithmetic mean of highest and smallest de-
gree node (as given by simulation result shown in Table 1).
Equation 3 does that with a maximum potential Plnmax =
1
max(κ(i,t))+(min(κ(i,t)) .
2

C. Trust level of the node


A selfish leader would result in frequent resending of same
packets to base station due to incorrect or no packet received
from a given request. This would result in decrease of the
energy efficiency of the system. We calculate the trust metric
Fig. 3: Triangle topology and the degree of connectivity of from [12] and for any node i we denote it by Pt (i, t). The
each node dependence of trust is reflected by the normalized parameter
ωτ (i, t) and is given in equation 5.
We use simulation results using MATLAB to quantify this Pt (i, t)
ωτ (i, t) = N (5)
dependency. We consider a triangular topology with n(n+1) 2 i=1 Pt (i, t)
numbers of nodes. The value of n is the number of nodes in
the base of the triangle. We have taken n as 7 i.e. 28 nodes are IV. L EADER S ELECTION P ROTOCOL A LGORITHM
taken into consideration (as shown in Figure 3) and each node
are given a node√id i, i = 1, 2, ...28. The nodes are separated To get an expression for the effective potential of a node
by a distance of 2d or 2d depending whether the vertical or to be a leader, the parameters should be combined taking into
horizontal nodes are considered respectively. To establish the consideration their dependency. The parameters ωD , ωN and
connectivity of the graph from the given topology (i.e. to get ωτ can be superposed linearly with individual weights α, β
κ(i, t) value for each node) a maximum transmission distance and γ respectively associated with them. This is so because
2014 International Conference on Control, Instrumentation, Energy & Communication(CIEC) 511

the individual parameters are independent of each other (i.e. Data: network topology,number of nodes n, initial
one parameter doesn’t influence the value of other), the only energy of each node, length of data packet,
difference being that one parameter may have higher priority routing protocol R1 , α, β, γ
in assigning leader selection protocol than the other. So the Result: leader selected and number of rounds for which
effective potential ω(i, t) of a node to be a leader is expressed it is selected (numberof round)
as in equation 6 Initialization: K, ωD , ωN , ωτ =n × 1 elements of 0,
round=0, recompute=0, leader=0;
Compute ωD , ωN , ωτ for all nodes at round=0;
ω(i, t) = E(i, t) × (αωD (i, t) + βωN (i, t) + γωτ (i, t)) (6) while At least 20% of node in the network is present do
recompute=0;
Select node with highest ω as leader;
In equation 6 the term E(i, t) which gives the energy of Compute numberof round = min(K − kleader );
the node i at any particular instant is introduced so as to while numberofround>0 do
distribute the leaders through out the network and relief any Send data packet to base station using the
particular node to be a leader every time resulting in fast energy selected node as leader and with routing
depletion. The linear combination parameters follows the rule protocol R1 ;
α +β +γ = 1 and (α, β, γ > 0) To relax the base station from numberof round=numberof round-1;
leader selection computation at each round of data transfer round=round+1;
we forecast the number of rounds after which leader selection end
would occur. An array K = [k1 k2 ...kn ] is computed whose ith for p=1:n do
element is zero which corresponds to the current leader node if E(i, round) < Eth then
and other entries kj gives the number of rounds after which the E(i, round)=0;
node j can be a leader when node i is selected as the leader. set recompute=1;
Using energy calculation as in [11] the energy cost of sending end
a packet from the node i (leader node) of defined length to the end
base station can be calculated and is denoted as Esend (i, t). if recompute then
For all other nodes the energy cost is calculated considering Recompute ωD , ωN ;
the the distance to which it transmits the packet at the time end
instant t and is denoted as Esend (j, t). The array components Recompute ωτ ;
kj can be calculated as the the number of rounds after which a end
leader node i has its potential reduced and becomes the same Algorithm 1: Leader Selection Algorithm
as that of node j when the leader node i is elected at time
instant t and is given as in equation 7.

V. S IMULATION AND R ESULTS


ω(i, t)
(E(i, t) − kj Esend (i, t)) × = The simulation is done in MATLAB. Grid topology as
E(i, t)
ω(j, t) in Figure 1 is considered with spacings as described in
(E(j, t) − kj Esend (j, t)) × (7) Section 3.1. The communication energy model as in [11] is
E(j, t) considered. The packet length is taken to be as 200 bits. 100
nodes are considered in the network with initial energy of
The value kj for each element of the array K − ki is given 500mJ and threshold energy Eth = 10mJ. The simulation is
as in equation 8. run using routing protocols LBEERA, OREC, PEGASIS and
SHORT with leader selection protocol as in Algorithm 1 and
four different linear combination parameters. Figure 4 — 7
ω(i, t) − ω(j, t) compares the energy-distance based leader selection algorithm
kj = ω(i,t) ω(j,t)
(8)
Esend (i, t) × E(i,t) − Esend (j, t) × E(j,t)
with Algorithm 1 for different routing protocols and different
linear combination values. The comparison metric is taken as
the number of rounds of data transfer (network lifetime) after
The minimum value of K − ki gives the number of data which 80% of the nodes in the network dies.
transfer rounds after which leader selection is again done. The
parameters ωi and ωN are computed if there is a change in the
network topology (i.e whenever any one or more of the node The network lifetime increases across all the experiments.
in the network dies). ωτ is computed at every point whenever But for greater value of α, i.e. potential with respect to distance
the leader selection is done to include the dynamics of trust of the node from base station, increases the network lifetime to
behavior of the nodes.The algorithm of the leader selection more than 8% for all routing protocol as evident from Figure
is shown in Algorithm 1. It is considered that the network 6. The lowest network lifetime increase is obtained when β
becomes dysfunctional when 20% of the node dies. is given a higher weight. This shows that leader selection
protocol depends loosely on the degree of connectivity of
Algorithm 1 is polynomial time solvable and hence no the nodes. Optimization of the linear combination values for
approximation algorithm is required. The parameter Eth in different routing protocols (it may happen that different routing
Algorithm 1 is the minimum energy at which a node becomes protocol have different optimal linear combination value) is not
dysfunctional. included in this article and is left for future work.
2014 International Conference on Control, Instrumentation, Energy & Communication(CIEC) 512

Fig. 4: Comparison chart with α = 0.33, β = 0.33 & γ = 0.33 Fig. 6: Comparison chart with α = 0.25, β = 0.25 & γ = 0.25

Fig. 5: Comparison chart with α = 0.25, β = 0.25 & γ = 0.5 Fig. 7: Comparison chart with α = 0.25, β = 0.5 & γ = 0.25

[5] Yu, Yongchang and Wei, Gang Energy Aware Routing Algorithm Based
VI. C ONCLUSION on Layered Chain in Wireless Sensor Network. International Confer-
In this paper an energy efficient leader selection algorithm ence on Wireless Communications, Networking and Mobile Computing
(WiCom), 2701-2704, 2007, IEEE.
for routing protocol is designed and is compared with existing
[6] Kim, Jong-Myoung and Park, Seon-Ho and Han, Young-Ju and Chung,
energy-distance based leader selection. Linear combination Tai-Myoung CHEF: cluster head election mechanism using fuzzy logic
of three parameters — distance of the node form the base in wireless sensor networks. 10th international conference on advanced
station, degree of connectivity of the node and trust level of communication technology, 654–659, 2008, IEEE.
the node coupled with its instantaneous energy determines the [7] Gao, Teng and Jin, Ren Cheng and Song, Jin Yan and Xu, Tai Bing and
potential of a node to be a leader. The combination parameters Wang, Li Ding Energy-efficient cluster head selection scheme based on
α, β, γ shows that maximum network lifetime enhancement multiple criteria decision making for wireless sensor networks. Wireless
Personal Communications,vol.63,no.5, 871–894, 2012, Springer.
can be achieved using higher combination value α for the
[8] Shen, Yan and Ju, Hui Energy-Efficient Cluster-Head Selection Based
potential parameter distance from the node ωD . In all other on a Fuzzy Expert System in Wireless Sensor Networks. International
cases considered the leader selection algorithm dominates the Conference on Green Computing and Communications (GreenCom),
existing algorithm for leader selection in routing protocols. 110–113, 2011, IEEE.
Future work can be extended in getting an optimal value [9] Kumar, Dilip and Aseri, Trilok C and Patel, RB Distributed Cluster
for linear combination values at which the network lifetime Head Election (DCHE) Scheme for Improving Lifetime of Heterogeneous
enhancement is maximum. Sensor Networks. Tamkang Journal of Science and Engineering, vol.13,
no.3, 337–348, 2010.
[10] Sharma, Vidushi Cluster Head Selection in Wireless Sensor Networks
R EFERENCES under Fuzzy Environment. ISRN Sensor Networks, 2013, Hindawi Pub-
lishing Corporation.
[1] Akyildiz, Ian F and Su, Weilian and Sankarasubramaniam, Yogesh and
Cayirci, Erdal A survey on sensor networks. Communications magazine, [11] Mitra, Swarup Kumar and Banerjee, Joydeep and Chakraborty, Arpita
IEEE,vol.40, no.8, 102–114, 2002, IEEE. and Naskar, Mrinal K, Data gathering in wireless sensor network using
realistic power control. Proceedings of the 2011 International Conference
[2] Heinzelman, Wendi Rabiner and Chandrakasan, Anantha and Balakr-
on Communication, Computing & Security, 124–127, 2011, ACM.
ishnan, Hari Energy-efficient communication protocol for wireless mi-
crosensor networks. Proceedings of the 33rd Annual Hawaii International [12] Trakadas, Panagiotis and Maniatis, Sotiris and Karkazis, Panagiotis
Conference on System Sciences, 2000, IEEE. and Zahariadis, Theodore and Leligou, Helen-Catherine and Voliotis,
Stamatis A novel flexible trust management system for heterogeneous
[3] Yang, Yang and Wu, Hui-Hai and Chen, Hsiao-Hwa SHORT: shortest
wireless sensor networks. International Symposium on Autonomous
hop routing tree for wireless sensor networks. International Journal of
Decentralized Systems, 1–6, 2009, IEEE.
Sensor Networks, vol. 2,no.5, 368-374, 2007, Inderscience.
[4] Lindsey, Stephanie and Raghavendra, Cauligi S PEGASIS: Power-
efficient gathering in sensor information systems. Aerospace conference
proceedings, vol.3, 3–1125, 2002, IEEE.

Anda mungkin juga menyukai