Ad Hoc & Sensor Wireless Networks Vol. 8, pp. 5–20
Reprints available directly from the publisher
Photocopying permitted by license only
©2009 Old City Publishing, Inc.
Published by license under the OCP Science imprint,
a member of the Old City Publishing Group
Algorithms for Relay Node Selection
in Randomly Deployed Homogenous
Cluster-based Wireless Sensor Networks
Nauman Aslam, William Robertson and William Phillips
Department of Engineering Mathematics & Internetworking, Dalhousie University,
Halifax, NS, Canada, B3J-2X4
E-mail: {nauman.aslam,bill.robertson,william.phillips}@dal.ca
Received: December 15, 2007. Accepted: December 24, 2008.
Extended network life is one of the fundamental objectives in the design of
Wireless Sensor Network systems. Clustering protocols help in achieving
this objective in an efficient and scalable manner by organizing nodes into
small hierarchical groups. In single hop clustering protocols, assumptions
are made about direct communication from the cluster heads to the base
station. Such assumptions seem too strong for realistic situations as the
cluster head may have limited radio transmission range in some cases.
Relay nodes could be used to leverage additional energy saving through
multi-hop transmissions. Moreover, for a randomly deployed homogeneous network, placement of either relays or cluster heads at predetermined
locations poses challenges including mobility, location awareness and
energy provisioning. In this paper we propose a three-tier architecture for
randomly deployed wireless sensor networks without making any assumptions about mobility and placement of nodes at desired locations. We also
propose and evaluate heuristic based algorithms for relay node selection.
Simulation results demonstrate significant gains in network life time by
using the proposed algorithms.
Keywords: Wireless sensor networks, relay, clustering, energy efficiency, topology
control.
1 INTRODUCTION
Wireless Sensor Networks (WSNs) have emerged as the-state-of-the-art technology in data gathering from remote locations by interacting with physical
phenomena and relying on collaborative efforts by large number of low cost
5
“aswin117” — 2009/5/14 — 14:18 — page 5 — #1
6
N. Aslam et al.
devices [1,2]. Typically, a WSN is comprised of hundreds or thousands of
low cost sensor nodes. Each sensor node has an embedded processor, a wireless interface for communication, a non replenish-able source of energy, and
one or more onboard sensors such as temperature, humidity, motion, speed,
photo, and motion detectors. Once deployed, sensor nodes collect the information of interest from their on board sensors, perform local processing of
these data including quantization and compression, and forward the data to a
base station (BS) directly or through a neighboring relay node [3].
Energy conservation is one of the fundamental challenges in WSNs. In
a sensor node, energy is consumed in sensing, computation and processing,
however the wireless transceiver consumes a significant amount of energy as
compared to the others. Most of the WSN deployments are based on application specific data gathering i.e. a large number of sensor nodes send their
data to the BS. Clustering is employed as a standard approach for achieving energy efficiency and scalable performance in WSNs [4]. Sensor nodes
can be organized hierarchically by grouping them into clusters, where the
data is collected and processed locally at the cluster head. Single hop clustering protocols follow a two-tier architecture, consisting of sensor nodes and
cluster heads. In tier one communication, each sensor node sends its data
to its local cluster head. The cluster head performs data aggregation, compression and data fusion. In tier two the cluster heads send the processed
data to the BS. In two-tier architectures the relay nodes are often synonymous to the cluster head which relays the data collected from sensor nodes
to the BS. In protocols where the cluster head selection is performed randomly, the distribution of cluster heads is not well controlled. Most protocols
based on two-tier architecture use direct communication for sending cluster
head data to the BS. If a cluster head is located far away from the BS, its
energy consumption would be greater than that of the one located closer to
the BS for similar traffic patterns from cluster members. Moreover, the cluster
head job requires receiving data messages from member nodes, in-network
processing of the data, and transmission to the BS using high power levels. Intuitively, a relay node could off set some of the higher transmission
costs incurred by a cluster head in data transmission to the BS. We present a
simple three-tier architecture that allows further energy savings by employing relay nodes. Relay nodes act as forwarding agents to transport the data
collected by the cluster heads. Two novel algorithms are presented for selection of relay nodes thus providing a viable and energy efficient topology for
cluster-based WSNs.
The rest of the paper is organized as follows. Section 2 discusses the related
work. Section 3 presents an overview of the proposed three tier architecture.
Assumptions about the network are outlined in Section 4. The relay node selection algorithms are presented in Section 5. Simulation results are discussed
in Section 6. Finally, our main conclusions and future research directions are
highlighted in Section 7.
“aswin117” — 2009/5/14 — 14:18 — page 6 — #2
Relay Node Selection
7
2 RELATED WORK
In recent years several research efforts investigated relay node placement
strategies for distributed WSNs. Most of the research efforts were mainly
related to two-tiered architectures. The work in [5], among the early research
efforts in this area, considers a two-tiered wireless sensor network consisting
of sensor clusters deployed around strategic locations and Base Station (BS)
whose locations are relatively flexible. In addition to sensor nodes and cluster
heads, the authors also consider presence of application nodes (ANs) that is
similar to relay nodes. The authors proposed an approach to extend the network
lifetime by arranging the location of BSs and ANs. In [6], a polynomial time
heuristic is proposed for a joint problem of energy provisioning (EP) and relay
node placement (RNP) for a two tier architecture. In the two-tier architecture
considered, the aggregation and forwarding nodes (AFNs) are synonymous
to cluster heads that also act as relay nodes if required. The authors have proposed an iterative algorithm that is based on linear programming to find the
optimal location of relay nodes. They developed a polynomial–time heuristic
algorithm, smart pairing and Intelligent Disc Search (SPINDS), that solves
the EP–RNP problem. SPINDS consists of iterations to find the optimal
the location of relay nodes. Xu et al. investigated relay node deployment
strategies in heterogeneous WSNs for single hop [7] and multi-hop [8] cases
from a connectivity and lifetime perspective. Three deployment strategies
are presented for relay nodes, namely, connectivity oriented, life-time oriented, and hybrid oriented. Similar to [6], cluster heads act as relay nodes
in transporting the data collected from member nodes to the BS. The proposed deployment strategies are evaluated with respect to network lifetime
and energy utilization metrics. The results suggest that the hybrid deployment strategy outperforms the other two. In another related work [9] by the
same group, the relay node placement problem is addressed using heuristic
based Far Near Max Min principle. The authors in [10] investigate relay node
placement for large scale WSNs considering scalability and extended lifetime
as their prime objectives. Similar to the assumption made by [6], their network model also assumes a hierarchical architecture where cluster heads are
considered as relay nodes. They formulate the relay node placement as two
optimization problems: (i) Connected Relay Node Single Cover (CRNSC)
problem and (ii) 2-Connected Relay Node Double Cover (2CRNDC) problem. Two polynomial time approximation algorithms are proposed to solve
the CRNSC problem.
Research in [11] considers deployment of additional relay nodes for constructing a fault tolerant network. The problem of maintaining connectivity is
tackled by adding a small number of nodes (relays) ensuring that the resulting communication graph is 2-connected. Two approximation algorithms are
proposed for 2-edge and e-vertex connectivity. In [12], both single-tiered
and two-tiered relay node placement problems are investigated with the
“aswin117” — 2009/5/14 — 14:18 — page 7 — #3
8
N. Aslam et al.
objective that the network remains 2-connected. For the single tiered problem, a 14-approximation algorithm is presented. For the two-tiered problem, a
(10 + e)-approximation algorithm is presented. In [13], the authors have proposed a slightly different approach using mobile relay nodes that can move
around the network to maintain connectivity. In addition to mobility, these
mobile nodes are resource rich in communication and processing capabilities,
which allow them to act both as mobile relay and sink.
Although the work proposed in this paper shares the common objective of
extending the network lifetime, it is different in the following respects to the
related techniques noted above. Our work assumes a homogeneous network
where all nodes have the same computation and communication capabilities.
The work in [7–9] assume a heterogeneous network that includes a few powerful resource-rich nodes. Our work considers a three-tier architecture where
the communication between the cluster head and BS is performed through
relay nodes. Moreover the algorithms proposed in this paper are based on
local message exchange and do not include any significant processing and
computational overheads.
3 THREE-TIER ARCHITECTURE
Our architecture employs relay nodes at the third tier which act as forwarding
agents between the cluster head and the sink. Once the cluster head selection is
completed, relay node selection is performed. Each cluster head selects a relay
node to which it transmits the data. The use of relay nodes is motivated by
minimizing the energy consumption in the transmissions between the cluster
heads and the BS by introducing relay nodes at the intermediate level. Such
an architecture also provides benefits of light weight data forwarding and
greatly reduces the overheads associated with routing protocols in multi-hop
communication. Figure 1 shows an example of the three tier architecture used
in our framework. A sensor node can take one of the following roles in our
model:
Plain Nodes: have onboard sensors to monitor the physical phenomena and
produce data relating to an event. Each node also has a unique hard-coded ID
to identify itself to its neighbors. During the cluster set up phase each node of
this type associates with a cluster head for sending and receiving data.
Cluster Heads: gather data from their associated member sensing nodes and
forward it to the nearest relay node. The cluster head is also responsible for
the TDMA slot allocation in the data transmission phase.
Relay Nodes: forward the data collected from the cluster head to the sink.
Relay nodes do not perform in-network processing on the data received from
the cluster head. Data is forwarded in the same shape as received from the
cluster head.
“aswin117” — 2009/5/14 — 14:18 — page 8 — #4
Relay Node Selection
9
FIGURE 1
Overview of three-tier architecture.
4 NETWORK MODEL AND ASSUMPTIONS
The following assumptions are made for the sensor network under consideration:
1. Nodes are dispersed randomly following a Uniform distribution in a
2-dimensional space.
2. The location of the BS is known to all sensors. It can be assumed that
the BS is equipped with GPS which is used to transmit its location
information to all sensor nodes in the deployment area.
3. The BS is considered a powerful node with enhanced communication
and computation capabilities and no energy constraints.
4. The nodes are capable of transmitting at variable power levels depending on the distance to the receiver as in [4]. For instance, MICA
Motes use the MSP430 [14,15] series micro controller which can be
programmed to 31 different power levels.
5. The nodes can estimate the approximate distance by the received signal strength, given that the transmit power level is known, and the
communication between nodes in not subject to multi-path fading.
6. We assume that the sensor nodes are capable of switching between
sleep and active modes. Most commercially available platform such
as MICAz [16], TelosB [17], TMote Sky [18] support features such
“aswin117” — 2009/5/14 — 14:18 — page 9 — #5
10
N. Aslam et al.
Set of sensor nodes
ith node in S
Set of Cluster Head
Set of Relay Candidates
ith node in CH
Set of Relay Nodes
Probability to be relay candidate
Communication cost from i th cluster head to
j th relay node
S
si
CH
R
chi
RN
P
cij
TABLE 1
Notations used
as auto suspend, wake, and sleep mode that are used to minimize the
sensor node’s energy consumption.
7. A network operation model consists of set up (cluster head election,
relay selection, cluster formation) and data transmission phase.
8. The energy model presented in [4] is adopted here. The amount of
energy consumed for transmission ETx , of an l-bit message over a
distance d is given by,
for 0 ≤ d ≤ dcrossover
l.Eelect + l.εfs .d 2
ETx =
(1)
l.Eelect + l.εmp .d 4 for d ≥ dcrossover
where Eelect is the amount of energy consumed in electronics, εfs is the energy
consumed in amplifier when transmitting at a distance shorter than dcrossover ,
and εmp is the energy consumed in amplifier when transmitting at a distance
greater than dcrossover .
The energy expended in receiving an l-bit message is given by,
ERx = lE elect
(2)
These assumptions are very common among sensor networks and are cited in
most of the clustering protocols for data gathering applications including a few
well known ones, such as [4,19]. Although assumption 5 limits the environments where systems can be implemented, use of accurate radio propagation
models could be used to get the accurate estimates of the signal strength.
Table 1 outlines the notations used in the following sections.
5 RELAY NODE SELECTION ALGORITHMS
We present two algorithms in this section for relay node selection. Both of the
algorithms are based on our motivation on optimizing the energy consumption
by finding a unique relay for each cluster head.
“aswin117” — 2009/5/14 — 14:18 — page 10 — #6
Relay Node Selection
11
FIGURE 2
Relay Selection using Minimum Weight Perfect Matching.
5.1 Minimum weight perfect matching (MWPM)
Finding a unique relay node for each cluster head can be modeled using a
well known weighted matching problem in graph theory. A graph G(V , E)
consists of a set V of vertices and a set E of edges. A graph is bipartite if
the vertex set V can be partitioned into two sets CH and R such that no edge
in E has endpoints in the same set of bipartition. A matching M ⊆ E is a
collection of edges such that every vertex of V is incident to at most one edge
of M. If no vertex is exposed there is a perfect match. In our case, as shown
in Figure 2, the vertex set V consists of a set of cluster heads CH and a set of
relay candidates R. By definition,
V = CH ∪ R
(3)
with CH ∩ R = φ and E ⊆ CH × R.
Each edge e(i, j ) ∈ E is associated with a cost cij , which is the communication cost to forward a unit of data to the BS. The weight of matching M is
the sum of the weights of edges in M. Our goal is to find a perfect matching
M such that the total cost is minimized i.e.
Minimize c(M)
Where the cost of a matching M is given by,
c(M) =
cij
(4)
(i,j )∈M
The minimum weighted cost perfect matching problem in bipartite graphs
is also referred as the Assignment Problem. The best known algorithm for
“aswin117” — 2009/5/14 — 14:18 — page 11 — #7
12
N. Aslam et al.
Algorithm 1. Minimum Weight Perfect Matching
∀si ∈ S
1. u = rand(1)
2. if u ≤ p
3.
State = ‘Relay_Candidate’
4.
Broadcast Relay_Candidate_Msg
5.
Wait for Relay_Select Msg from cluster head
6.
If (msg received)
7.
State = ‘Relay_Node’
8.
Else
9.
State = ‘Plain_Node’
10.
End if
11. end if
∀chi ∈ CH
12. Receive Relay_Candidate_Msg
13. For each of the messages received, add a vertex rj and an edge e(i, j ).
14. Determine the corresponding cost cij for each edge e(i, j ).
15. Solve the resulting graph for assignment problem using the Hungarian
method.
16. Update the graph matching and find minimum cost relay node.
solving the Assignment Problem in weighted bipartite graphs is the Kuhn
Munkres technique [20], known as the Hungarian algorithm [21]. We apply
the technique in [20] to find the best choice of relay node for each cluster head
such that the over all communication cost to the BS is minimized.
5.2 Minimum distance relay selection (MDRS)
The second algorithm is based on the simple idea that a cluster head would
select a unique relay, in its radio transmission range, which has the minimum
distance to the BS. Considering that candidates for the relay nodes are selected
randomly, some cases may occur where a cluster head cannot find a relay candidate whose communication cost to the BS is lower than that of its own. In
such a scenario finding a unique relay may violate our energy conservation
objective. An intuitive solution is to relax the condition of finding the unique
relay in the case where a relay node with lesser cost does not exist. In other
words if a cluster head cannot find a relay that has communication cost lower
than the cluster head’s own cost to the BS, it will choose to become it own relay
node. The pseudo code for minimum distance relay selection is outlined in
Algorithm 2.
In the first part (line 1 to 11) for both of the above mentioned algorithms,
relay candidates are elected randomly, based on a predefined probability p,
from sensor node set S. It is also worth mentioning here that the relay nodes
“aswin117” — 2009/5/14 — 14:18 — page 12 — #8
Relay Node Selection
13
Algorithm 2. Minimum Distance Relay Selection
∀si ∈ S
1. u = rand(1)
2. if u ≤ p
3.
State =‘Relay_Candidate’
4.
Broadcast Relay_Candidate_Msg
5.
Wait for Relay_Select Msg from cluster head
6.
If (msg received)
7.
State =‘Relay_Node’
8.
Else
9.
State =‘Plain_Node’
10.
End if
11. end if
∀chi ∈ CH
12. Receive Relay_Candidate_Msg
13. Extract id, cost to BS, store information in the Relay_Candidate_List
(RCL)
14. Sort the RCL in increasing order by cost to BS
15. Search the RCL from top and find the relay node rj not already selected
by any other cluster head
16. If rj .cost_to_BS < self.cost_to_BS
17.
chj .relayID ← rj
18.
Send Msg Relay_Select
19. else
20.
chj .relayID ← chj .ID
21. end if
are selected after a set of cluster heads is elected. However, the selection of
relay nodes is performed before the cluster formation step i.e. when sensor
nodes associate themselves with cluster heads. This is useful because, some
clustering algorithms such as [22,23] include inter-cluster (cluster head to BS)
communication cost in their cluster formation step. Hence cluster heads are
able to advertise a correct inter-cluster communication cost eliminating any
chances of errors. The probability p is chosen to ensure that, for a given round,
the number of relay candidates is greater than the number of cluster heads.
Intuitively, a large p value would result in more overheads as more messages
are exchanged between relay candidates and the cluster heads. On the other
hand if p is very small then we may encounter cases where each cluster head
is not able to find a suitable relay node. Finding an optimal p that satisfies
both objectives of minimum overheads and finding a best relay for each cluster
head can be very complicated. We investigate the tradeoffs related to different
values of p with respect to different network sizes in Section 6.
“aswin117” — 2009/5/14 — 14:18 — page 13 — #9
14
N. Aslam et al.
6 SIMULATION RESULTS
In this section, we evaluate performance of our algorithms via simulations.
We assume ideal conditions at the physical level such that the probability of
wireless interference is negligible. The simulator was developed in MATLAB
which allows efficient and realistic modeling of sensor nodes by using object
oriented programming support and integrated technical computing environment. As described in Section 3 we use a model where network operation
progresses in rounds. Each round in turn consists of a setup phase and data
transmission phase. The setup phase further consists of clustering phase, relay
selection and cluster formation. The clustering phase refers to electing a set
of cluster heads from the active nodes and the cluster formation refers to the
process where sensor nodes become cluster members. In the data transmission
phase, each sensor node sends a fixed amount of data to its cluster head which
forwards it to its relay node. The relay node ultimately forwards this data to
the BS. Modeling the network operation in this manner also allows network
life to be measured as the number of rounds until the very first node runs out
of its energy or until a percentage of nodes completely exhaust their battery
and die. The lifetime on an individual sensor node is measured in the number
of rounds before its energy is depleted. We implement both relay selection
algorithms using the clustering algorithm proposed in [22]. In simulation of
the MWMP algorithm, only rounds where a perfect matching is found between
CHs and RNs is considered for the analysis. We use a number of performance
metrics related to network lifetime and energy conservation to evaluate both
algorithms. We also compare both algorithms with the direct transmission case
(cluster head to BS) to analyze the gain in network life time. In addition to
network life, the parameters critical to the performance of both algorithms are
also investigated. Table 2 provides the common simulation parameters used
in the experiments.
Sensor Deployment Area
BS Location
Number of Nodes
Data Packet Size
Control Packet Size
p
Initial Energy
Eelect
εfs
εmp
100 × 100 m
(50, 200) m
100–400
1024 bits
128 bits
0.15
0.1 J
50 nJ/bit
10 pJ/bit/m2
0.0013 pJ/bit/m4
TABLE 2
Simulation Parameters
“aswin117” — 2009/5/14 — 14:18 — page 14 — #10
15
100
200
90
180
80
160
Number of Active Nodes
Number of Active Nodes
Relay Node Selection
70
60
50
40
30
Direct Transmission
MWPM
MDRS
20
10
0
200
400
600
800
Number of Rounds
140
120
100
80
60
Direct Transmission
MWPM
MDRS
40
20
1000
0
1200
200
(a)
1000
1200
1000
1200
(b)
300
400
350
250
Number of Active Nodes
Number of Active Nodes
400
600
800
Number of Rounds
200
150
100
Direct Transmission
MWPM
MDRS
50
0
200
400
600
800
Number of Rounds
(c)
300
250
200
150
Direct Transmission
MWPM
MDRS
100
50
1000
1200
0
200
400
600
800
Number of Rounds
(d)
FIGURE 3
Network life in rounds Vs. number of active nodes in network size of: (a) 100 Nodes, (b) 200
Nodes, (c) 300 Nodes, (d) 400 Nodes.
We first evaluate the performance of the proposed algorithms with respect
to the network life time metric. Scalability with respect to network size is an
extremely important design goal in WSNs. In order to evaluate the scalability
of the proposed algorithms, we used different network sizes of 100 to 400
nodes. All simulation results are averaged over 20 randomly generated network topologies. The value of p is taken as 0.15. Figure 3 shows the network
life in data collection rounds plotted against the number of active nodes. We
compare both algorithms to the direct communication case where cluster heads
send their data directly to the sink. It can be observed in Figure 3 that both
algorithms extend network life significantly as compared to the direct transmission case. For the network size of 100 and 200 nodes the gain in network
life for MDRS and MWPM is approximately 25% and 15%. For the network
size of 300 and 400 nodes this gain amount to approximately 15% and 10%
for MDRS and MWPM respectively.
By using relay nodes at the third tier the distance between the cluster head
and the BS, which is the most expensive part of communication, is divided into
smaller portions rendering significant energy savings. In minimum distance
relay selection, energy is saved by introducing the relaxation for cluster head to
become their own relay if a relay with lower communication cost is not found.
“aswin117” — 2009/5/14 — 14:18 — page 15 — #11
16
N. Aslam et al.
Direct Transmission
MWPM
MDRS
15
10
5
0
0
200
400
600
800
Number of Rounds
1000
40
Average Residual Energy (J)
Average Residual Energy (J)
20
30
25
20
15
10
5
0
1200
Direct Transmission
MWPM
MDRS
35
0
200
60
1200
80
Direct Transmission
MWPM
MDRS
50
40
30
20
10
0
200
400
600
800
Number of Rounds
(c)
1000
1200
Average Residual Energy (J)
Average Residual Energy (J)
1000
(b)
(a)
0
400
600
800
Number of Rounds
Direct Transmission
MWPM
MDRS
70
60
50
40
30
20
10
0
0
200
400
600
800
Number of Rounds
1000
1200
(d)
FIGURE 4
Average Residual Energy of System Vs. Number of Rounds in network size of: (a) 100 Nodes,
(b) 200 Nodes, (c) 300 Nodes, (d) 400 Nodes.
We also investigate the average residual energy of the system because it
also provides an estimate of the network life. Figure 4 illustrates results for
the four different network sizes used here. The y-axis indicates the average
residual energy of the system and the x-axis denotes the number of rounds.
It can be observed that the mean residual energy of the system in the case
of the MDRS is higher than that of both the MWPM and direct transmission
cases. In other words the system energy in MDRS dissipates at a much slower
rate as compared to the MWPM and direct transmission cases. Hence, the
network life under the MDRS algorithm is enhanced compared to MWPM
and direct transmission. These results corroborate the results presented in
Figure 3.
The average energy consumed per round is also analyzed for all network
sizes. Figure 5 depicts the results for average energy consumed per round
for different network sizes using a random topology. These statistics were
collected using 20 randomly generated topologies with no dead nodes in the
network. Again we observe a similar behavior as seen previously in Figures 3 and 4. The use of relay nodes results in an energy efficient topology
compared to the direct transmission case by exploiting the most expensive
part of communication. In MDRS, the relaxation for the cluster head to find a
“aswin117” — 2009/5/14 — 14:18 — page 16 — #12
17
Average Energy Consumed per Round (J)
Relay Node Selection
0.1
MDRS
MWPM
Direct Transmission
0.08
0.06
0.04
0.02
0
100
200
300
Network Size
400
FIGURE 5
Average energy consumed per round vs. network size.
Number of Relay Candidates / Relays
45
40
35
30
Min. distance relay selection (Candidates)
Min. distance relay selection (Relays)
25
Minimum weight matching (Candidates)
Min. weight matching (Relays)
20
15
10
5
0
0
2
4
6
8
10
12
14
Number of Cluster head
FIGURE 6
No. of Relays and Relay Candidates vs. No. of Cluster Heads.
unique relay with lower cost than itself assists in saving more energy compared
to MWPM.
Figure 6 shows the number of relay nodes and relay candidates for both
algorithms. A p value of 0.15 is used. It can be observed that both techniques
render the same number of candidates; however, the MDRS produces a slightly
less number of relay nodes compared with the other techniques. Such behavior
discounts the cases when a cluster head cannot find a minimum cost relay and
chooses to become a relay for itself.
As discussed in Section 5, the probability to become relay candidate, p,
is an important parameter in both algorithms. We investigate the effect of p
“aswin117” — 2009/5/14 — 14:18 — page 17 — #13
18
N. Aslam et al.
Energy Consumed (J)
0.04
MWPM
0.03
0.02
0.01
MDRS
0
0.25
0.2
400
0.15
300
0.1
200
0.05
P
100
Network Size
FIGURE 7
Energy consumed in setup phase per round with respect to p and network size.
MWPM
Energy Consumed (J)
0.12
0.1
0.08
0.06
MDRS
0.04
0.02
0.25
0.2
0.15
0.1
P
0.05
100
150
200
250
300
350
400
Network Size
FIGURE 8
Energy consumed in data transmission phase per round with respect to p and network size.
on the energy consumed in setup and data transmission phase for the network
sizes considered here. Figure 7 analyzes the energy consumed in the setup
phase per round for the value of p ranging form 0.05 to .25 and network sizes
of 100 to 400 nodes. Similarly, Figure 7 analyzes the energy consumed in data
transmission phase per round for the value of p ranging form 0.05 to .25 and
network sizes of 100 to 400 nodes.
“aswin117” — 2009/5/14 — 14:18 — page 18 — #14
Relay Node Selection
19
From Figure 7, it can be clearly seen that there is no difference in the energy
consumed by both algorithm. For a given probability p and a network size,
the number of relay candidates are the same. Hence the amount of energy consumed due to exchange of control message remains the same in both cases. It is
also worth noticing here the impact of node density (network size). For a high
density network the energy consumed rises much more sharply as compared to
one of low density. Figure 8 illustrates the energy consumed per round for the
data transmission phase under similar settings. Two points are evident here.
First, for a given network size, p does not effect energy consumption. This is
because, after the setup phase, role of p becomes insignificant. At this point
each cluster head has already selected a relay node. Second, the difference of
energy consumed between MDRS and MWPM. Again, this difference is due
to the energy efficient topology achieved by MDRS.
7 CONCLUSIONS AND FUTUREWORK
In this paper we have presented a three-tier architecture for randomly deployed
homogeneous WSNs. We have also presented two simple algorithms for relay
node selection. The MWPM algorithm uses the Hungarian method to find
a unique relay for each cluster head. The MDRS algorithm will only select
a relay node when the communication cost of transporting data to the BS
through the relay node is less than direct communication. We have shown
that by introducing relay nodes at the third tier, the network life time can be
enhanced significantly. We intend to investigate the problem of relay node
placement for fault tolerance and reliability in our future work.
REFERENCES
[1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam and E. Cayirci, A Survey on Sensor Networks.
IEEE Communications Magazine 40 (2002), 102–114.
[2] H. Karl and A. Willig, Protocols and Architectures for Wireless Sensor Networks, Wiley,
2005.
[3] F. Zhao and L. J. Guibas, Wireless Sensor Networks: An Information Processing Approach,
Morgan Kaufmann, 2004.
[4] W. Heinzelman, A. Chandrakasan and H. Balakrishnan, An Application-Specific Protocol Architecture for Wireless Microsensor Networks. IEEE Transactions on Wireless
Communications 1 (2002), 660–670.
[5] J. Pan, Y. T. Hou, L. Cai, Y. Shi and S. X. Shen, Topology control for wireless sensor
networks, in Proceedings of the 9th annual international conference on Mobile computing
and networking, 2003, pp. 286–299.
[6] Y. T. Hou, Y. Shi, H. D. Sherali and S. F. Midkiff, On energy provisioning and relay node
placement for wireless sensor networks. Wireless Communications, IEEE Transactions on
4 (2005), 2579–2590.
“aswin117” — 2009/5/14 — 14:18 — page 19 — #15
20
N. Aslam et al.
[7] K. Xu, H. Hassanein, G. Takahara and Q. Wang, Relay node deployment strategies in heterogeneous wireless sensor networks: single-hop communication case, in IEEE GLOBECOM,
2005.
[8] K. Xu, H. Hassanein and G. Takahara, Relay node deployment strategies in heterogeneous wireless sensor networks: multiple-hop communication case, in Second Annual IEEE
Communications Society Conference on Sensor and Ad Hoc Communications and Networks,
IEEE SECON 05 2005, pp. 575–585.
[9] Q. Wang, K. Xu, G. Takahara and H. Hassanein, Locally Optimal Relay Node Placement in
Heterogeneous Wireless Sensor Networks, in IEEE Globecom 5, 2005.
[10] J. Tang, B. Hao and A. Sen, Relay node placement in large scale wireless sensor networks.
Computer Communications 29 (2006), 490–501.
[11] A. Kashyap, S. Khuller and M. Shayman, Relay placement for higher order connectivity in
wireless sensor networks, in IEEE INFOCOM, 2006.
[12] W. Zhang, G. Xue and S. Misra, Fault-Tolerant Relay Node Placement in Wireless Sensor
Networks: Problems and Algorithms, in Proceedings of the IEEE INFOCOM, vol. 7, 2007.
[13] W. Wang, V. Srinivasan and K. C. Chua, Using mobile relays to prolong the lifetime of wireless sensor networks, Proceedings of the 11th annual international conference on Mobile
computing and networking, 2005, pp. 270–283.
[14] A. Chipcon, CC1000 Single Chip Very Low Power RF Transceiver Data Sheet, 2004.
[15] I. Texas Instruments, MSP430x13x, MSP430x14x Mixed Signal Microcontroller.
Datasheet, 2001.
[16] Crossbow Technology Inc., http://www.xbow.com/Products/Product_pdf_files/Wireless
_pdf/6020-0060-01_A_MICAz.pdf.
[17] Crossbow Technology Inc., http://www.xbow.com/Products/Product_pdf_files/Wireless
_pdf/TelosB_Datasheet.pdf.
[18] Moteiv Corporation, http://www.moteiv.com/xcart/product.php?productid=1, in
[19] O. Younis and S. Fahmy, HEED: A Hybrid, Energy-Efficient, Distributed Clustering
Approach for Ad Hoc Sensor Networks. Transactions on Mobile Computing 3 (2004),
660–669.
[20] J. Munkres, Algorithms for the Assignment and Transportation Problems. Journal of the
Society for Industrial and Applied Mathematics 5 (1957), 32–38.
[21] H. W. Kuhn, The Hungarian method for the assignment problem. Naval Research Logistics
52 (2005), 7–21.
[22] N. Aslam, W. Robertson, S. C. Sivakumar and W. Phillips, Energy Efficient Cluster Formation Using Multi-Criterion Optimization for Wireless Sensor Networks, presented at 4th
IEEE Consumer Communications and Networking Conference (CCNC), 2007.
[23] M. Ye, C. F. Li, G. H. Chen and J. Wu, EECS: An Energy Efficient Clustering Scheme
in Wireless Sensor Networks, IEEE Int’l Performance Computing and Communications
Conference (IPCCC), 2005, pp. 535–540.
“aswin117” — 2009/5/14 — 14:18 — page 20 — #16