Vous êtes sur la page 1sur 6

Proceedings of the Ninth International Conference on Machine Learning and Cybernetics, Qingdao, 11-14 July 2010




Department of Computer Science, Xiamen University, Xia Men 361005, China

E-MIAL: happyspring055@163.com, csyxie@xmu.edu.cn, chyang@xmu.edu.cn, jingzhengwei@hotmail.com

Abstract 1. Introduction

LEACH is a popular hierarchical routing protocol

which efficiently maintains the energy storage of nodes Wireless Sensor Network (WSN) has been widely de
in Wireless Sensor Network (WSN). The nodes using ployed in the applications of military, science research, in
LEACH are divided into clusters. The randomized ro dustry and environment protection [1]. A WSN consists
tation of cluster head in each cluster can save the energy of smart sensor nodes which collect and transmit informa
consumption of nodes. However, the random election tion to the base station (BS) by radio. All communications
of cluster heads without considering nodes' residual en are required to consume low power because the batteries of
ergy may reduce and oscillate the lifespan of network. nodes are impossible or difficult to be replenished. Many
In this paper, we have proposed the Adaptive Cluster routing protocols have taken energy consumption into ac
Head Election and Two-hop LEACH protocol (ACHTH count.
LEACH) to prolong the lifespan of network. It improves The energy-efficient routing protocols in WSN can
LEACH by using an adaptive algorithm of cluster head be classified into three categories: data-centric protocols
election and allowing multi-top transmission among [5][7], location-based protocols [7][10] and hierarchical
cluster heads and Base Station (BS). Nodes are tagged protocols [2][3][7][8][9]. In this paper, we focus on hier
as near nodes or far nodes according to the distances archical protocols which are featured by clustering nodes.
to the BS. The near nodes belong to one cluster while The main idea of hierarchical routing is to efficiently main
the far nodes are divided into different clusters by the tain the energy consumption of sensor nodes by aggregating
Greedy K-means algorithm. The cluster head is shifted data and reducing transmissions[7]. Low-Energy Adaptive
and the node with the maximal residual energy in each Clustering Hierarchy protocol (LEACH) [1][2] is one typ
cluster is elected. During the data transmission phase , ical example which divides sensor nodes into clusters. In
the far cluster heads may select the cluster head in the each cluster, LEACH randomly selects one sensor node as
near area as the next hop or communicate directly to the cluster head in order to evenly distribute the energy stor
the BS. The simulation results have shown that ACHTH age among nodes. The data aggregations of cluster heads
LEACH outperforms several existing protocols in terms also reduce the communications to the BS and save energy.
of network's lifespan. Especially, ACHTH-LEACH can However, LEACH possibly chooses too many cluster heads
achieve more than 2 times longer lifespan than LEACH at a time or randomly selects the cluster heads far away from
and build a more stable routing environment. the BS without considering nodes' residual energy. As a re
sult, some cluster heads drain their energy earlythus reduc
Keywords: ing the lifespan of WSN.
In this paper, we improve LEACH by designing the
WSN; LEACH; Lifespan; Greedy K-means
Adaptive Cluster Head Election and Two-hop LEACH pro
tocol (ACHTH-LEACH). The nodes near to the BS com
pose one near cluster. The nodes far from the BS are di
vided into different far clusters. In order to balance the
total energy dissipation of nodes, the cluster head is ro
"'The correspondence author: Xie Yi, csyxie@xmu.edu.cn tated by selecting the node with the maximal residual en-

978-1-4244-6527-9/10/$26.00 2010 IEEE

Proceedings of the Ninth International Conference on Machine Learning and Cybernetics, Qingdao, 11-14 July 2010

ergy in each cluster. For avoiding the overloading of clus increases the area of sensor nodes. That is, BCDCP pro
ter heads, ACHTH-LEACH combines the greedy method vides an energy-efficient routing scheme in a vast range of
with K-means algorithm [4] to build the far clusters so that sensing applications. But both LEACH-C and BCDCP re
the number of sensor nodes in each cluster is approximately quire that all nodes transmit the information to the BS. This
equaL Since the long distance transmission consumes con increases the complexity of routing protocoL
siderable energy, the far cluster heads implement the two Recently, some protocols are proposed to implement
hop transmission to the BS by using the near cluster head load balance as well as prolong the lifespan of network. For
as the next hop. Compared with LEACH, ACHTH-LEACH example, Thein et aL [9] put forward the stochastic clus
has been shown to increase the lifespan of network by more ter head selection algorithm for LEACH. The probability of
than 2 times. each node being a cluster head is determined by the ratio of
the node's residual energy to its initial energy and the opti
1.1. Related work mal numbers of cluster heads. For the sake of load balance,
CALRE [3] maintains the different sizes of clusters based
The main task of WSN is to transmit the data collected
on the distances to the BS. The maximal size of one clus
by sensor nodes to the BS. The simplest method is using
ter is small if the cluster is near to the BS. In each cluster,
direct transmission [1] where each node sends data to the
the node with the most residual energy becomes the clus
BS. A lot of energy is wasted on sending similar data to the
ter head to further balance the energy consumption among
BS. If a node is far away from the BS, it will soon die be
nodes. Additionally, the cluster heads near to the BS not
cause much energy is consumed on long distance transmis
only fuse data but also relay the data from the other cluster
sions. In order to reduce energy consumption and data re
dundancy, the cluster-based protocols have been proposed.
The above observations show that the load balance
They organize the sensor nodes into clusters. The cluster
among nodes and the energy reduction in the communica
head aggregates the received data and sends the aggregated
tions are efficient to prolong the lifespan of network. There
data to the BS. The original method Static-Clustering[1]
fore, we have proposed ACHTH-LEACH which utilizes the
only solves the problem of data aggregation. But the fixed
Greedy K-means algorithm in clustering, selects the cluster
cluster heads become network bottlenecks because of their
heads by nodes' residual energy and allows two-hop trans
considerable energy consumption. To solve this problem,
mission to the BS. The far cluster heads in CALRE must
Heinzelman et al.[1] [2] proposed an adaptive clustering
select the next hop among existing cluster heads in every
routing protocol called LEACH. It increases the lifespan of
transmission. But ACHTH-LEACH reduces the complex
WSN by using the cluster heads' randomized rotation and
ity of routing since it makes all near nodes as one cluster
reducing transmissions by fusing data in cluster heads.
and its cluster head relays data for other cluster heads.
The idea proposed in LEACH has been an inspiration for
The rest of this paper is organized as follows. Section
many hierarchical routing protocols [2] [3][7][8][9]. Using
two describes the network model. Section three briefly
Power Efficient Gathering in Sensor Information Systems
introduces LEACH and shows its defects. Section four
(PEGASIS) [7], the nodes are organized into a chain using
presents ACHTH-LEACH including the designs of active
the greedy algorithm so that each node transmits and re
algorithm and two-hop transmission. Section five evalu
ceives data from the nearest neighbor. The aggregated data
ates ACHTH-LEACH by comparing it with Direct Trans
moves from node to node and ultimately reaches the BS via
mission, Static Clustering and LEACH. Finally, we draw
one randomly chosen leader in the chain. PEGASIS reduces
our conclusions and discuss our future work.
the amount of energy spent per round but results in the ex
cessive delay of distant nodes on the chain. Moreover, if the
single leader breaks, the whole network will be invalidated. 2. Network Model
LEACH-C[2] improves LEACH by using a central control
algorithm to build clusters. Each node sends information The network model and wireless communication model
about its current location and residual energy to the BS. (radio model) under our consideration are similar with the
The BS computes the average energy of all nodes and then ones in the LEACH protocol[1] [2]. We assume that sensor
the nodes whose energy storage is over this average may nodes are randomly distributed in a square area of a WSN
be selected as cluster heads. Using the global knowledge which has the following properties:
of network, LEACH -C can build optimal clusters and select Include two kinds of nodes: 1) Sensor nodes with lim
appropriate cluster heads to save energy. Base-Station Con ited energy can sense their own residual energy and have
trolled Dynamic Clustering Protocol (BCDCP)[8] is also a the same architecture; 2) One BS without energy restriction
centralized routing. It utilizes a high-energy base station to is far away from the area of sensor nodes.
operate the cluster head selection, routing path construction All sensor nodes are immobile. They use the di-
and TDMA scheduler. BCDCP outperforms LEACH-C and rect transmission or multi-hop transmission to communicate

Proceedings of the Ninth International Conference on Machine Learning and Cybernetics, Qingdao, 11-14 July 2010

with the BS.

Sensor nodes sense environment at a fixed rate and al
ways have data to send to the BS.
T(n) =

Sensor nodes can revise the transmission power of wire o otherwise (3)
less transmitter according to the distance.
The lifespan of WSN is the total amount of time before 3.1 Defeats of LEACH
the first sensor node runs out of power [6][11].
Although LEACH saves nodes' energy and prolongs net
Equations (1) and (2) of energy consumption are
work lifespan, it still has several shortcomings. At the be
adopted to calculate the transmission cost ET",(k,d) and
ginning of each round,LEACH randomly selects the cluster
the receiving cost ERx (k) of k-bit message between two
heads without considering the residual energy of nodes. As
nodes where their distance is d. The parameter Eelec is the
a result, the number and location of cluster heads are un
per bit energy dissipations for transmission and reception.
stable. Table 1 records the minimal, maximal and average
We also use the free-space and two-ray models according
residual energy of cluster heads (denoted as Emin , Emax
to the distance between the transmitter and receiver. do is
and Eavg) at the rth round in one simulation. The num
a threshold transmission distance. If d<do, the free-space
ber of cluster heads with the notation of m widely varies
model will be employed; otherwise, the two-ray model will
round by round. The residual energy of cluster heads differs
be employed.E /s and Eamp are the amplifier parameters
greatly since that the difference between Emin and Emax
of transmission corresponding to the free-space and the
is significant. Moreover, the unbalance of residual energy
two-ray models, respectively.
grows worse since that the standard deviation of residual
ETx(k, d) ETx-elec + ETx-amp(k, d)
{ energy of cluster heads (denoted as SDE) increases with r
= =

(Eelec + E, /sd2 ) * k, d < do (do = J f,:p) . If the node with little residual energy is elected as a cluster
(Eelec + E,ampd4) * k, d >= do (1) heacl,it will run out of power quickly because of the heavy
workload and then speed up the lifespan of network.

(2) Table 1: Residual energy of cluster heads

r m Emin(J) Emax(J) Eavg(J) SDE(J)
We use the same parameters of [2]: Eelec=50nJlbit, 10 8 1.718 1.994 1.917 0.1038
E/s=lOpJlbitlm 2,Eamp= 0.0013 pJlbitlm4, and the energy 20 24 1.593 1.987 1.817 0.1490
for data aggregation Eda=50nJlbitJsignal. In the follow 30 4 1.642 1.945 1.742 0.1612
ing simulations, the 100 nodes are randomly deployed in an 40 13 1.319 1.832 1.677 0.1753
area of 100m*100m where each node is assigned an initial 50 6 1.120 1.836 1.534 0.3057
energy of 2 Joules. The BS locates at a point of (50,175). Next, we repeat the SImulation of LEACH 500 times to
Each data message is 500 bytes where the packet header is compare the lifespan of network (denoted as L). Six net
25 bytes. work topologies created by different settings of node distri
bution are considered individually. The minimal and max
imal samples of network lifespan are denoted as Lmin and
3 Introduction of LEACH
Lmax respectively. Lavg is the average of 500 samples and
SDL is their standard deviation. Table 2 demonstrates the
The operation of LEACH is broken into rounds. Each oscillating routing of LEACH since that the samples of net
round contains two states: cluster setup state and steady work lifespan vary greatly under each node setting. For ex
state. In the cluster setup state, each node chooses a ran ample, SDL is near one fifth of Lavg in Setting 1. The main
dom number between 0 and 1. If the number is less than reason is that the number and locations of cluster heads are
the threshold T(n), the node will turn into a cluster head for randomly selected in LEACH. When the more cluster heads
the current round. The threshold T(n) is shown in Equa are far from the BS, they will more quickly consume energy
tion (3) where p is the desired percentage of cluster heads, on long transmissions and then the lifespan of network will
r is the current round and G is the set of nodes which have decrease.
not acted as cluster heads in the last lip rounds [1]. Nodes Each cluster head using LEACH directly communicates
are subsequently divided into clusters based on the received with the BS no matter the distance is near or not. When the
signal strength and then they notify the corresponding clus network is huge, the communication between cluster heads
ter heads. During the steady state, each cluster head gathers and the BS consumes much energy for the long distance
and fuses the data from nodes before directly sending it to transmission. So the lifespan of WSN would be shortened.
the BS. Then, the network reselects the cluster heads in a If something is wrong with the charrnel, the data can not
new round. The lifespan is accounted by rounds. arrive at the BS in time.

Proceedings of the Ninth International Conference on Machine Learning and Cybernetics, Qingdao, 11-14 July 2010

Table 2: Lifespan of network under different settings

1. Select k initial group centroids pk,(k = 1 , .... ,K.) among N
Time \Set 1 2 3 4 5 6 nodes, the location of Pk is(xk,Yk)
Lmin 20 32 35 24 39 35 Randomly choose one node as t he first centroid Pl
Lmax 74 74 90 79 84 74 For k=2 to K
Lavg 56 59 64 57 62 58 For any node ai rf- {Pl, P2, .... , Pk- d
SDL 9.49 9.02 9.06 8.61 8.94 8.82 Calculate
Dist(ai) min k {(Xi -Xt)2+(Yi -Yt)2}.
4 Improvement on LEACH: ACHTH . . . . -l
arg max{ dist(ai) h=l,

ACHTH-LEACH is an improved protocol to abate the End

defects of LEACH. Firstly, each node is marked as a near 2. Form K clusters by assigning all nodes to the closest
node or a far node according to its distance to the BS. One centriod
near cluster includes all near nodes. And the Greedy K 3. Recomputed the centroid of each cluster
means algorithm is designed to build the clusters of far 4. Repeat step 2 and step 3 until the num
nodes so that the number of common nodes in each clus

ber of iterations reach the maximal value or
ter is approximately equaL In order to evenly distribute
the energy among nodes, the node with the maximal resid JLf=l[(Xi-X)2+(Yk-y)2l K IS less than

ual energy is elected as the cluster head. ACHTH-LEACH

the threshold 6 where p is the value of Pk in the last
also adopts two-hop transmissions to avoid consuming high
transmission power in the cluster heads. If the near cluster
head is alive, the cluster heads in the far area will select it
as the next hop; otherwise they will communicate directly
to the BS.
Figure 1. The Greedy K-means algorithm
The operation of ACHTH-LEACH includes three major
phases: clustering, selecting cluster head and data transmis
sion. Clustering phase is a prelude. One selecting cluster the node with maximal energy. This is helpful to evenly
head phase and one data transmission phase compose one distribute the load among the nodes in one cluster. The
round. In this section, we will describe the details of these cluster heads can save energy since that they only broad
three phases. cast information within their own clustering area whereas
the cluster heads in LEACH broadcast the information for
4.1 Clustering the whole network. Similarly, the common nodes only re
ceive the broadcast information from its own cluster head so
Firstly, we tag the near or far nodes according to their that their energy consumption and operation complicacy are
distances to the BS. If one node's distance d is less than the reduced. The flow chart of electing cluster head is shown in
threshold do , the node is tagged as near; otherwise, the node Figure 2.
is tagged as far. All near nodes belong to one near cluster.
Then we combine the Greedy method of choosing initial 4.3 Two-hop transmission
centroids with the K-means algorithm to build the clusters
in the far area. The hybrid algorithm is called as Greedy During the data transmission phase, the common nodes
K-means algorithm whose steps are shown in Figure 1. in one cluster send data by the TDMA schedule to the clus
K-means algorithm is one of the most clustering algo ter head which is similar in LEACH. The transmission be
rithms because of its effi c iency in c lu stering large data sets. tween cluster heads and the BS has two types: one-hop
It aims to partition n observations into k clusters in which transmission and two-hop transmission. W hen the near
each observation belongs to the cluster with the nearest cen cluster is alive, the far area's cluster heads may trans
troids [4] . A better choice of initial centroids leads to a mit their aggregated data to the BS via the near cluster
better result of clustering. We utilize the greedy method to head. Otherwise, they directly communicate with the BS.
determine the initial centroids, allowing each node to stay The flow chart of two-hop transmission between the cluster
as far away from another as possible. heads and BS is shown in Figure3.

4.2 Cluster head election 5 Simulation Analysis

After setting up the clusters, the cluster head will be ro We develop a simulator in VC to evaluate the perfor
tated by computing the residual energy of nodes and select mance of routing protocols. In this section, we simu-

Proceedings of the Ninth International Conference on Machine Learning and Cybernetics, Qingdao, 11-14 July 2010

Common node

Receive the broadcast

Figure 2. The flow chart of cluster head elec


late ACHTII-LEACH and further compare it with other

three protocols: Direct Transmission, Static Clustering, and Figure 3. The flow chart of two-hop transmis
LEACH in terms of network lifespan. We continue to use sion
the simulation parameters in Subsection 2.2.
Figure 4 presents the improvements gained by ACHTII always below 0.01 whereas LEACH is more than 0.1.
LEACH under 6 different settings of nodes' distribution. Table 3: 'T/ under different settings
The values on the y axis are obtained by averaging 500 sam
'T/\Set 1 2 3 4 5 6
ples of the network lifespan under the same node setting
LEACH 0.168 0.152 0.142 0.152 0.145 0.151
and the given protocol. These average values of ACHTII
LEACH are whining the confidence interval with the rate ACHTCH 0.004 0.004 0.008 0.010 0.008 0.008
. .

95%.LEACH achieves approximately 4.5 times and 2.8 AddItionally, we adopt another looser definition of the
times longer for the lifespan compared with Direct Trans network's lifespan, which is the number of rouoos when a
mission and Static Clustering. In Direct Transmission and certain percentage of nodes run out of power [11] . For ex
Static Clustering, the network terminates quickly because ample, the percentage is set as a =25% in [8] and a =0%
the far nodes or the fixed cluster heoos will consume lots of in the previous sections. Figure 5 clearly illustrates that
energy to communicate with the BS. In contrast, LEACH ACHTII-LEACH also outperforms its counterparts under
performs localized algorithms, data aggregation along with the new definition. For example, when a =25%, ACHTH
randomly self-elected cluster heads, which decreases redun LEACH can increase the lifespan of network by factors of
dancy data and prolongs the lifespan. Moreover, ACHTII 10.2 times over Direct Transmission, 10.7 times over Static
LEACH achieves approximately 2 times longer network Clustering,and 2 times over LEACH respectively. More
lifespan than LEACH. The main reason is that the random over, the advantage of ACHTII-LEACH is more distinct
ized rotation of cluster heads in LEACH makes some clus when the lifespan of network is defined with a smaller a.
ter heads concentrate in a certain area ofWSN. Then many The curve of ACHTII-LEACH is smoother than the one
common nodes which in "cluster head deprived" will dis of LEACH. When a=25%, the network using ACHTH
sipate amount of energy to transmit data to their faraway LEACH terminates at the 204th round while the network us
cluster head. However, ACHTII-LEACH uses the balanced ing LEACH terminates at the 148th round. When a=50%,
clustering approach and allows two-hop communication to the lifespan of network is about 250 rounds using ACHTH
distribute energy dissipation evenly among nodes. And then LEACH and LEACH.
ACHTII-LEACH increases the lifespan of network.
Next, we define 'T/ as the ratio of the standard deviation 6 Conclusion
of the lifespan samples to their average. The smaller
means the more stability of routing protocol. Table3 then We have proposed ACHTII-LEACH to improve LEACH
demonstrates that ACHTII-LEACH is better in stabil and remedy its defects. The clusters are set up based on the
ity than LEACH since that the of ACHTII-LEACH is Greedy k-means algorithm. The cluster heads are elected

Proceedings of the Ninth International Conference on Machine Learning and Cybernetics, Qingdao, 11-14 July 2010

250 ------
Secondly,the performance of ACHTH-LEACH can be fur
Direct Static Gus LEACH ACHTH-LEACH ther improved if some parameters and threshold values are


This paper is supported by Scientific Research Founda

tion for Talents (063Ox12107),Xiamen University.


o ------
[1] W. R. Heinzelman, A. Chandrakasan and H. Balakr
Setting ishnan, "Energy Efficient Communication Protocol
for Wireless Micro sensor Networks", Proceedings of
Figure 4. Lifespan of network under different the 33rd Hawaii International Conference on System
setting Sciences,2000.
[2] W. Heinzelman, A. Chandrakasan and A. Balakrish
nan, "An Application-specific Protocol Archiecture
_Direct _StaticClu< ...... LEACH -ACHTH.LEACH
for Wireless Microsensor Networks", IEEE Transac
:c300 tions on Wireless Communications, Vol 1, No.4, pp.
e 250
660-670, Oct. 2002.
-:;;: [3] D. Xu and H. Zhang, "Improved algorithm of LEACH
0200 protocol introducing residual energy", Computer En
gineering and Applications, :II 45, No.28, pp. 115-
.; 150 1 48
o 119,2009.
1ij 100 [4] P. Berkhin, "A Survey of Clustering Data Min
'" ing Techniques", http://www . springerlink,
.l!! 50
:::; com/content/x321256p66512121/
[5] M. Younis,M. Youssef and K. Arisha, "Energy-aware
routing in cluster-based sensor networks", Proceed
0% 5% 10"11> 15% 20% 25% 30% 35% 40% 45% 50%
ings of the10th IEEE MASCOTS'02, 2002.
[6] X. Zhou, B. Qin and F. Xu,"Wireless sensor networks
Figure 5. Relationships between network and security", National defense industry Press,2007.
lifespan and a
[7] K. Akkaya and M. Younis, ''A survey on routing pro
tocols for wireless sensor networks",Elsevier Ad Hoc
Networks, Vol 3, Issue 3, Pages 325-349, pp. 325-349,
by considering the residual energy of sensor nodes. And the May 2005.
cluster heads may adopt two-top transmission to reduce the [8] S. D. Muruganathan, D. C. Ma, RI. Bhasin and A.
energy spent on sending data to the BS. The simulation re O. Fapojuwo, "A Centralized Energy-Efficient Rout
sults have shown that ACHTH-LEACH effectively prolong ing Protocol for Wireless Sensor Networks", IEEE Ra
the lifespan of network by the balanced clustering approach dio Communications, March 2005.
and the two-hop communication to the BS. For example, [9] M. C. M. Thein and T. Thein, ''An Energy Effi
ACHTH-LEACH increases more than 2 times longer net cient Cluster-Head Selection for Wireless Sensor N et
work lifespan than LEACH. ACHTH-LEACH can evenly works", Proceedings ofISMS201O, 2010.
distribute the energy load among all sensor nodes and build [10] Y. Xu, J. Heidemarm and D. Estrin, "Geography
a stable routing environment. Even when the definition of informed energy conservation for ad hoc routing",
lifespan is loosed to allow a certain percentage of nodes Proceedings of the 7th Annual ACMIIEEE MobiCom,
to run out of power, ACHTH-LEACH still outperforms July 2001.
LEACH in terms of the network lifespan. [11] K. Sha and W. Shi, "Revisiting the Lifetime of Wire
We will keep on ACHTH -LEACH and make it more flex less Sensor Networks", Proceedings of the ACM con
ibe and energy-efficient Firstly, the re-clustering mecha ference on Embedded Networked Sensor Systems,
nism can be introduced when many new sensor nodes join 2004.
in and the percent of nodes alive is less than a thresmld.