Energy-Efficient Clustering System Model and

Reconfiguration Schemes for Wireless Sensor

Networks
Hang Su and Xi Zhang
Networking and Information Systems Laboratory
Department of Electrical Engineering
Texas A&M University, College Station, TX, 77843, USA
Email: {hangsu, xizhang}@ee.tamu.edu

AbstractBy taking the data-receiving energy consumption techniques [3] which works as follows. The sensors send the
into account, we propose an extended analytical model to derive collected raw data to the cluster-head first. Then, after
the optimal parameters the number of clusters for wireless
sensor networks. Our analyses show that the previous existing processing the data locally in the cluster-head, the cluster-head
model underestimates the optimal number of clusters within a transmits the processed data to the base station or data
sensor network. The simulation results verify our analyses and processing center.
demonstrate the modified model is more accurate in deriving the The clustering algorithms have received significant research
optimal number of clusters to maximize the network lifetime of attention recently. The early Linked Clustering Algorithm
the wireless sensor networks. Using our extended model, we also
develop the energy-threshold-driven based clustering reconfigu- (LCA) [4] is a simple clustering scheme where each node is
ration schemes with the optimal number of clusters to further assigned with the unique identification number (ID) for
improve the wireless sensor-network lifetime. The simulation cluster-head selection. To better control the number of cluster
experiments indicate that our clustering reconfiguration schemes members in each cluster, authors in [5] proposed Max-Min d-
outperform the existing ones in terms of sensor-network lifetime Hop Clustering algorithm, where the maximum number of
while imposing low overhead due to clustering reconfigurations.
hops between the cluster members and its cluster-head is less
Index Terms Wireless sensor networks, energy control, clus- than d. The authors of [7] developed a clustering algorithm that
tering reconfiguration, energy-threshold driven, Voronoi cells. can choose single-hop or multi-hop method adaptively based
on the analysis of the energy consumed. This system employed
I. INTRODUCTION
two types of nodes where one type is ordinary sensor node and
A S THE DEVELOPMENT of the VLSI device, commu-
nication and battery technologies, the tiny sensor nodes
have the capacities of sensing, wireless communicating, and
the other type is complex/high-energy node specially used as
the cluster-head. This scheme is not too robust because the
system can quit functioning if one of the special cluster-heads
data processing. This type of sensors not only collect the dies.
environment data by detecting the surrounding objects, but In [8], authors proposed a distributed clustering algorithm
also process the collected data and then transfer it to the base called Low-Energy Adaptive Clustering Hierarchy (LEACH)
station or data process center via wireless communication that includes two parts: clustering algorithm (LEACH-CA) and
channels. Compared with their wired counterpart, the wireless cluster-head rotation scheme (LEACH-CHRS). Their simula-
sensor networks are less costly and easier to deploy because tions show that there exists the optimal number of clusters
of its infrastructureless and ad hoc construction. Wireless which minimizes the energy consumption. However, they did
sensor networks have a wide range of applications in the fields not give the solution on how to derive the optimal number of
of military, environmental monitoring, and health and home- clusters for LEACH. The authors of [9] proposed a distributed
networks, etc. clustering algorithm, which we thus refer to as BC Clustering
The limited size of the sensor nodes in wireless sensor Algorithm (BCCA) in this paper. The BCCA is similar to
networks confines the energy they can carry [1] . As a result, LEACH-CA, but differs from LEACH-CA in that BCCAs
the transmission power of the sensor is significantly lower transmission power is fixed while LEACH-CAs transmission
than wired-line networks. The sensors transmit the collected power varies. One of the important contributions of [9] is the
data via multi-hop wireless communications channels. Authors development of an analytical model used to derive the optimal
in [2] showed that the energy consumed for transmission is number of clusters for BCCA. However, their model only
much higher than that for data processing. Thus, it will be considered the energy consumption used for the data
more energy-efficient to employ the clustering and data-fusion transmission without taking the data-receiving energy con-
The research reported in this paper was supported in part by the U.S. sumption into account. Unfortunately, if the data-receiving
National Science Foundation CAREER Award under Grant ECS-0348694. energy is ignored, the important fact that the cluster-head
uses more energy tnan the cluster member, except for the part tem or any location awareness device. Henee, the nodes are not
consumed for data-fusion, will mistakenly be neglected, aware of their localities. The energy in each node is limited.
significantly degrading the accuracy of modeling results. To All nodes transmit at the same power level and henee have the
remedy this deficiency, we propose an extended model which same radio range denoted by r. The energy at processing
takes the energy consumption used by not only the data center is unlimited since it has the wired power supply, and
transmission, but also the data-receiving into consideration. thus the system energy consumption does not consider the
Our analyses show that our extended model can correct the energy dissipation at the processing center.
underestimation on the optimal number of clusters, which is To make the analysis tractable, we also use the following
caused by the previous model proposed in [9]. Furthermore, three assumptions. First, the nodes at the wireless sensor
by using the optimal number of clusters, we also develop the network are distributed following the homogeneous spatial
energy-efficient clustering reconfiguration schemes, called Poisson process with intensity A in a 2-dimensional space.
Energy-Threshold-Driven Based Reconfiguration (ET-Driven), Second, the communication environment is contention-free
which are showed to be more energy-efficient as compared and error-free, and thus, nodes do not have to retransmit any
with the LEACH-CHRS scheme. data packets. Third, the compression ability is strong so that
The rest of the paper is organized as follows. Section II the cluster-head can compress several data packets, which are
proposes our modified system model for BCCA-based algo- from the member nodes, into a fixed-length packet.
rithm to derive the optimal cluster number by considering both As per our assumptions, the nodes are distributed according
data-transmission and data-receiving energy consumption. to the homogeneous spatial Poisson process and henee, the
Section III proposes our enhanced clustering algorithms. number of nodes in a square rea of side 2a is a Poisson
Section IV presents the simulation results to evalate the random variable N with mean equal to XA, where A = 4a?.
performance of proposed schemes. The paper concludes with Assume that for a particular realization of the process, there
Section V. are n nodes in this rea. Also, assume that the probability of
becoming a cluster-head is p; henee, on average, np nodes can
II. THE SYSTEM MODEL become cluster-heads. Let D be the random variable that
The BCCA algorithm is a distributed, randomized clustering denotes the length of the distance from a node located at (x,
algorithm with a run-time of O(k) rounds, where k is the y), i = 1,2,..., n to the processing center. Without loss of
mximum number of hops from the member nodes to their generality, we assume that the processing center is located at
cluster-head. Each of the nodes equipped with the same the center of the square rea. Then,
hardware has the unique ID. The BCCA algorithm works as
follows. In the BCCA, each sensor node becomes cluster-head
with the probability of p and advertises itself as a cluster-head
to the sensors within its radio range with the help of
broadcasting Cluster Jieadjmsg packets. These cluster-heads Now, since a node becomes a cluster-head with probability
formed this way are called the Volunteer Cluster-Heads (VCH). p, the cluster-head nodes and the non-cluster-head nodes are
The lifetime of the broadcasting packet is k hops. The non- distributed according to independent homogeneous spatial
cluster-head nodes that receive the broadcasting packets join Poisson processes PP1 and PPO with the intensity Ai = pX and
the cluster of the closest (measured by hops or Euclidean Ao = (1 -p)X, respectively.
distance) cluster-head. If a node neither is a cluster-head or Denote
If denoting
that L
the
v as
energy
the random
consumedvariable
for transmitting
for the total
a unit
length
of
has joined any cluster, it claims itself as the cluster-head. between
data by Pthe
t and
PP0theprocess
energypoints
consumed
to thefornucleus
receiving
in aaVoronoi
unit of
These cluster-heads formed in this way are called the Forced cell, then
databy Pr.its
Without
averageloss
canofbegenerality,
written aswe[10]:
let Pr = f3Pt, where
Cluster-Heads (FCH). Since the mximum number of hops of
the Cluster Jieadjmsg packet away from the cluster-head is A;, If we do not limit the mximum number of hops, according
if a node cannot receive any Cluster Jieadjmsg packet in the to the results from stochastic geometry, the square rea can be
time of t/. (the time in which the Cluster Jieadjmsg is partitioned into a number of Voronoi cells, each of which
forwarded), the node becomes a forced cluster-head. corresponding to a PP1 process point, called its nucleus [10].
The authors of [9] proposed the analytical model of the Let Nv be the random variable denoting the number of PPO
BCCA and showed that the energy are mainly determined by process points in each Voronoi cell and we get the expected
two parameters, the probability, denoted by p, of becoming a valu of Nv conditioned on N as follows [10]:
cluster-head and the mximum number, denoted by k, of hops.
However, the work in [9] did not consider the data-receiving
energy consumed at each cluster-head. We will develop a
modified analytical model that takes data-receiving energy into
account.
In our data gathering sensor networks, every node has the
same functions (processing and communicating). The nodes
are not equipped with GPS (Global Positioning System) sys-
Fig. 1. The valu of optimal probability p* against data-receiving energy-cost Fig. 2. Energy consumption against probability of becoming a cluster-head with
parameter fi with different vales of modifying parameter m when A = 15 and a different node densities. The line marked with x in xy-plane indicates the
= 5. optimal p achieving the lowest energy-cost performance for a given node
density .

3 (0 < 3 < 1) is defined as the data-receiving energy-cost

parameter. Taking the IEEE 802.11 2Mbps wireless network Taking derivative at both sides of Eq. (6) with respect to p, we
card for an example, the valu of 3 is about 0.73 [11]. can obtain:
The distance of d units between two nodes is equivalent to
where c' = 3.060^ - [(4mr/3)/(l + /3)]A/A. Notice that c'
\d/(mr)~\ hops, where r is the sensor radio range and m (0 < m
reduces to c defined in [9] when 3 = 0, implying that the
< 1) is the modified parameter which characterizes the sensor-
model does not consider the energy consumed by receiving
nodes density in wireless sensor network and is determined by
packets.
m = 1 e~A/100. In the wireless sensor networks, there are np
Solving the Eq. (7), we obtain three roots, including two
Voronoi cells. Therefore, the total energy V consumed by the
complex ones and a real-valued one. We take the only real
nodes in all Voronoi cells to communicate a unit of data to
root of Eq. (7) as the optimal probability p* for a sensor node
their cluster-heads can be written as:
to become the cluster-head, which is determined by:

Define p to be the energy consumed by the cluster-heads

to transfer the collected and processed data to the precessing
center. Then, the expected valu of p conditioned on J\ is
determined as follows: Using Eq. (8), Fig. 1 plots the optimal probability p*(/3, m)
as a function of the data-receiving energy-cost parameter 3
and the sensor-nodes density parameter m. From Fig. 1, we
observe that p* (/?, m) is a monotonically increasing function
of both 3 and m, which reveis the insight that the optimal p*
needs to be increased when taking the data-receiving energy-
where npPr refers to the receiving energy consumed by the cost and sensor-node density into account. Thus, p* obtained
processing center. Because the processing center is supplied by in the original model proposed in [9] is underestimated without
unlimited power, we do not consider the energy consumption considering the data-receiving energy consumption and sensor-
in processing center. node density. Thus, the modified model we proposed is more
We define to be the total energy spent in the sensor accurate in deriving the optimal p* to maximize the lifetime of
network system. Therefore, the average total energy spent in wireless sensor networks. Furthermore, since the average
the system can be written as: number of sensor nodes in each cluster equals E[AFV]
E[NV\N = n} = A0/A1 = (1 -p*)\/{p*\) = {l-p*)/p\ Fig. 1
implies that the existence of data-receiving energy con-
sumption requests the number of sensor members in each clus-
ter be reduced, or equivalently the number of the total cluster-
heads in the entire wireless sensor network be increased, to
maximize the lifetime of the wireless sensor network. The
 TABLE I THEORETICAL OPTIMAL
consumption for transmitting and receiving, respectively, in
PROBABILITY AND MAXIMUM HOPS
one round as shown in Table II. We find that the data-
Node Density 5 7.5 10 15 20 30 receiving energy consumption takes a great portion of the total
Optimal Prob. .1685 .1469 .1321 .1151 .1047 .0921 energy consumption, and the percentage of data receiving
Max. Hops 4 3 3 3 2 2 energy ulereases with the increase of nodes density. That also
TABLE II COMMUNICATION ENERGY
motivates us to enhance the analytical model developed in [9].
By Eq. (2), we know that each cluster-head on average has (1
CONSUMPTION IN A CLUSTER-HEAD
p)/p cluster members and then receives (1 p)/p packets
Density of nodes from its cluster members per round. Then, the cluster-head
5 10 15 20 30 spends P+Pr(l p)/p = [1+(3(1 p)/p]Pt unit energy for
Tx Energy 1 1 1 1 1 communication in each round.
Rx Energy 4.92 6.57 7.69 8.55 9.86
Rx/(Tx+Rx) 83.1% 86.8% 88.5% 90.0% 90.8%
III. THE CLUSTERING RECONFIGURATION SCHEMES
We observe that the cluster-head receives the packets from
above observations are expected since the higher the sensor all the member nodes of its cluster, but the member nodes do
data-receiving energy-cost for each sensor member in a cluster, not need to receive the packets from other peer nodes in the
the fewer the number of sensors in each cluster is desired to same cluster. As a result, the cluster-head spends more energy
reduce the data-receiving/-gathering energy consumption at all than the member nodes. Therefore, the entire network lifetime
cluster-heads. is short because of the unbalance of energy load.
Let a be the probability for a sensor node to become a We develop two types of clustering reconfiguration schemes
forced cluster-head and k be the number of hops allowed based on the BCCA clustering algorithms. The clustering
between a node and its cluster-head. Intuitively, the greater the reconfiguration schemes try to balance the depletion of battery-
valu of k, the less the valu of a. Authors in [9] derive the power at the different nodes. The key point of the clustering
relationship between k and a as follows: reconfiguration schemes is the dynamic rotation of the cluster-
head. That is, the cluster-head will be replaced by the member
nodes of its cluster under the certain specified conditions.
In our clustering reconfiguration schemes, the role of
cluster-head just switch from a node to another node within
In the n-station networks, the number of the forced cluster- the same cluster. Henee, the optimal parameters p and k
heads is na if the number of hops between a node and its obtained in Section II remain the same because the total
cluster-head is limited to the valu obtained from the Eq. (9). number of cluster-heads do not change. We note that the
In order to reduce the energy consumption, the number of clustering reconfiguration schemes can work even when the
forced cluster-heads should be small. For example, the valu forced cluster-heads exist1 because there is no other member
of a can be set to 0.001, i.e., there is a forced cluster-head node in the cluster of forced cluster-heads and forced cluster-
among 1000 nodes on average. heads do not need to spend energy on receiving packets of
We conduct simulation experiments to verify that the param- member nodes.
eters p* of our algorithms derived through Eq. (8) minimizes We classify our clustering reconfiguration schemes into two
the energy dissipated in the system. We use Matlab to simlate types: Time Driven (T-Driven) algorithm and Energy
the energy consumption in one round by using the BCCA Threshold Driven (ET-Driven) algorithm.
algorithm. The nodes are distributed according a homogeneous T-Driven algorithm: Every node re-eleets cluster heads
spatial Poisson process in a square rea whose side is 5 units. every specified rounds no matter how much energy it has. In
The processing center is located in the center of the square other words, the cluster-head selection algorithm will be
rea, i.e., (2.5,2.5). The transmission range is assumed to one executed periodically, for instance for each 10 rounds. If we
unit. Let (3=1 and both the transmission and receive power set the reconfiguration period longer than the lifetime, the T-
are in a unit energy. We use Eq. (9) to get the valu of the Driven algorithm reduces to static clustering algorithm because
mximum hops k with r = 1 and a = 0.001. the clustering reconfiguration will never take place.
We consider a range of vales for the probability p of The energy overhead of T-Driven algorithm consists of two
becoming a cluster-head in the algorithm with different node parts, cluster-head re-electing and nodes joining new cluster.
densities. We conduct 1000 experiments for each parameter p In fact, T-Driven algorithm is similar to LEACH-CHRS. In
separately and the results provided in the Fig. 2 are the average LEACH-CHRS, the probability of becoming a cluster-head for
energy consumption over 1000 experiments. We can observe each period is chosen to ensure that every node becomes a
from Fig. 2 that the energy dissipated in the networks is indeed cluster-head at least once within 1/p rounds [8]. Both of the
minimum at the theoretically optimal vales of the parameter LEACH-CHRS and T-Driven re-elect the cluster-heads every
p* calculated using Eq. (8) which are given in Table I with the
node density being 5, 7.5, 10, 15, 20 and 30. 'in practice, the probability (a) of forced cluster-heads is quite small, e.g., a
= 0.001.
The simulation also provides the cluster-heads energy
 The rotation of cluster-head is executed and the member
node of maximum residual energy becomes the new cluster-
head if the condition, Emax/ECH > Thcond2, is satisfied, where
Thcond2 is a predetermined parameter. Intuitively, the greater
the value of Thcond2, the less frequently the change

specified rounds. The difference between them is that LEACH-

CHRS ensure every node becomes cluster-head within 1/p
rounds, but T-Driven cannot ensure this. Thus the LEACH-
CHRS algorithm has the same overhead as T-Driven algorithm.
ET-Driven algorithm: We set a threshold value as the
criterion. If the energy of a specified node match the criterion,
the cluster to which the specified node belongs, re-elect the
cluster-head. Compared with the T-Driven algorithm, the
number of clusters which are involved in changing is small in
ET-Driven algorithm intuitively.
We develop two types of clustering reconfiguration criterion
for the ET-Driver algorithm below.
1. Clustering Reconfiguration Criterion 1 (ET-Driven 1):
Let Einit and Eres be the initial energy and residual energy of the
cluster-head, respectively, and Ememi be the residual energy
of member node i. Let C be the set of all nodes in a cluster.
Every round the member node i adds its residual energy
information Ememi to the packet which contains the collected
data and transmits the modified packet to the cluster-head.
Therefore, the cluster-head knows the residual energy of each
member node. The cluster-head compares the value of Einit and
Eres. Then, the cluster-head notifies the member nodes that the
member node, which has the most residual energy, will
replace itself to become the new cluster-head if Eres/Einit <
Thcond1 is satisfied, where Thcond1 is a predetermined parameter.
The ET-Driven 1 pseudo algorithm executed in the cluster-
head is described in Fig. 4.

Algorithm 1 ET-Driven 1
1: New Round begins
2: repeat
3: Receiving the message including raw data plus energy information (Ememi)
from Node i 4: until get messages from all
member nodes 5: Processing the raw data
6: Transmitting the aggregated data to processing center 7: if
Eres/Einit < Thcond1 then 8: Rotating the role with Node [arg
maxiC Ememi] 9: end if

Fig. 4. The ET-Driven 1 algorithm

2. Clustering Reconfiguration Criterion 2 (ET-Driven 2):

We use the same assumptions as the Clustering Reconfigu-
ration Criterion 1. Let C be the set of all nodes in a cluster and
ECH be the residual energy at cluster-head.
Every round the member node i adds its residual energy in-
formation Ememi to the data packet. The maximum residual
energy Emax among the member nodes can be calculated by the
cluster-head as:
times of reconfiguration. The ET-Driven 2 pseudo code ex-
ecuted in the cluster-head is shown in Fig. 5. The energy
consumed by cluster rotation, nodes joining new cluster and
transmission of energy information between cluster-head and
members impose the energy overhead of the ET-Driven
algorithm.

Fig. 5. The ET-Driven 2 algorithm

IV. SlMULATION RESULTS

We conduct the simulation experiments to get the network
lifetime of the static BCCA, LEACH, T-Driven and ET-Driven
algorithms. In [13], The network lifetime is defined as the
period from the time when network starts working to the time
when the first node stops functioning due to lack of energy. We
also use this network lifetime definition in this paper.
For comparison convenience, the LEACH used for our
testing below is modified based on the original LEACH. In our
experiment versin of LEACH, the node transmits in a fixed
power and connects to the cluster-head via multi-hop method.
In other words, our experiment versin of LEACH is a
combination of BCCA and LEACH-CHRS. The optimal pa-
rameters given in Section II are applicable for the experiment
versin of LEACH.
For the same system as the one described in the Section II.
Our experiments compare the network lifetime when static
BCCA, LEACH, T-Driven with different reconfiguration peri-
ods, and ET-Driven with different clustering reconfiguration
conditions are adopted. The reconfiguration periods for T-
Driven are set to 1, 10. We cali them as T-Driven 1 and T-
Driven 10, respectively. The valu of Thcon(1 for ET-Driven 1 is
set to 20A, which is the optimal valu obtained through a large
number of simulation experiments. Using the same method, we
know that the optimal valu of Thcond2 for ET-Driven 2 is 0.4.
Fig. 5 (a) compares the network lifetime among the different
schemes. We see that the network lifetime of MTE and the
Static BCCA are shorter because of the unbalanced energy load
and decreases rapidly as the node density increases. The MTE
gives the worst performance because it consumes a great deal
of energy every round as discussed in Section II. Comparison
with the Static BCCA, the LEACH, T-Driven and ET-Driven
algorithms achieve the relatively stable network-lifetime
performance regardless of the node density. The lifetime of ET-
Driven is also longer than that of LEACH and T-Driven for
different node densities.
 Fig. 5. Comparisons among algorithms with various node densities. (a) Network lifetime. (b) The average number nodes subjected to change cluster every round. (c) Energy overhead
consumed by clustering reconfiguration.

The network lifetimes achieved by T-Driven algorithms are show its correctness by simulations. Our analytical analyses
almost the same although the clustering reconfiguration period is reveal the insight that the original analytical model under-
diferent. In every round the total energy consumption by all estimates the optimal number of clusters and thus needs to be
nodes are nearly the same statistically. The most important factor modified. The simulation results verified our analyses and show
that limits the network lifetime is the energy draining of cluster- that the modified model we proposed is more accurate in deriving
head. If one node is not allowed to become a cluster-head for a the optimal number of clusters to maximize the lifetime of
long time, no matter what period of the clustering wireless sensor networks. Based on our modified model, we also
reconfiguration, T-Driven achieve almost the same performance. proposed and analyzed the clustering reconfiguration schemes: T-
Fig. 5 (b) shows that how many nodes are subject to joining the Driven and ET-Driven to improve the network lifetime of the
new cluster on average per round when employing different sensor networks. The simulation results show that our proposed
algorithms. Clearly, in MTE and Static BCCA, none of the node ET-Driven can significantly prolong the wireless sensor network
is required to change cluster. T-Driven 1 algorithm has the largest lifetime as compared to that using the T-Driven and LEACH-
number of nodes that join new clusters and the number is linear to CHRS scheme.
the density of nodes. The number of nodes that join new clusters
in ET-Driven 2 is the minimum among all the clustering REFERENCES
reconfiguration schemes. Compared with the T-Driven and [1] I. F. Akyildiz, W. Su, Y. Sankarsubramaniam, and E. Cayirci, Wireless sensor
networks: a survey, Computer Networks, 38 ,2002, pp. 393-422. [2] G. J. Pottie and
LEACH, ET-Driven have advantage on the less nodes which are W. J. Kaiser, Wireless Integrated Network Sensors, Communi-
forced to join the new clusters. From Fig. 5 (b), we learn that cations ofthe ACM, Vol. 43, No 5, May 2000, pp 51-58. [3] A. Boulis, S. Ganeriwal,
and M. B. Srivastava, Aggregation In Sensor Networks:
there are nearly no nodes that are subject to changing cluster per An Energy-Accuracy Trade-off" Elsevier Ad Hoc Networks Journal, Vol. 1, 2003,
round. pp. 317-331. [4] D. J. Baker and A. Ephremides, The Architectural Organization of A
Mobile
The results of energy overhead of LEACH, T-Driven and ET- Radio Network Via A Distributed Algorithm, IEEE Trans. Comm., November
Driven algorithms are shown in Fig. 5 (c). We find that the T- 1981, COM-29(ll), pp. 1695-1701. [5] A. D. Amis, R. Prakash, T. H. P. Vuong and D.
T. Huynh, Max-Mm D-Cluster
Driven 1 spends the most energy on the clustering Formation in Wireless Ad Hoc Networks, Proc. of INFOCOM, March 2000. [6] M.
reconfiguration. This is expected, because when employing T- Gerla, T. J. Kwon, and G. Pei, On Demand Routing in Large Ad Hoc Wireless
Driven 1, the nodes in network are required to reconfigure or re- Networks With Passive Clustering, Proc. of WCNC, 2000. [7] V. Mhatre and C.
Rosenberg, Design Guidelines for Wireless Sensor Networks:
cluster in each round. The more frequently the clusters change, Communication, Clustering and Aggregation, Elsevier Ad Hoc Networks Journal
the more the energy overhead is imposed. In addition, we find that Vol. 2, 2004, pp. 45-63. [8] W. R. Heinzelman, A. Chandrakasan and H.
Balakrishman, Energy-Efficient
the energy overhead caused by T-Driven and LEACH is much Communication Protocol for Wireless Microsensor Networks, Proc. Of IEEE
higher than that by ET-Driven algorithms. Although the ET-Driven HICSS, January 2000. [9] S. Bandyopahyay and E. J. Coyle, An Energy Efficient
Hierarchical Clustering
spend extra energy in messages transfer between cluster-head and Algorithm for Wireless Sensor Networks, Porc. of INFOCOM, March 2003. [10] S.
members, it reduces the number of nodes forced to join new G. Foss and S. A. Zuyev, On a Voronoi Aggregative Process Related to a
bivariate Poisson Process, Advances in Applied Probability, vol. 28, no. 4, 1996,
clusters caused by clustering reconfiguration. The ET-Driven pp. 965-981. [11] Laura Marie Feeney and Martin Nilsson, Investigating the energy
algorithms, especially ET-Driven2 algorithm achieve the better consumption
energy-saving performance than T-Driven and LEACH of a wireless network interface in an ad hoc networking environment, Proc. of
INFOCOM, Anchorage AK, USA, Apnl 2001. [12] T. Shepard, Decentralized
algorithms. Channel Management in Scalable Multihop Spread
Spectrum Packet Radio Networks, Laboratory for Computer Science, Mas-
V. CONCLUSIN sachusetts Institute of Technology, Cambridge, Tech. Rep. MIT/LCS/TR-670, July
1995. [13] J. H. Chang and L. Tassiulas, Energy Conserving Routing in Wireless Ad
We extended the existing analytical model to derive the Hoc
optimized parameter - the number of clusters - for BCCA and Networks, Proc. of INFOCOM, Tel Aviv, Israel, March 2000, pp. 22-31.