Upload
yacine
View
213
Download
1
Embed Size (px)
Citation preview
Energy-aware network configuration for Wireless Sensor Networks
Ada Gogu, Dritan Nace and Yacine Challal
Laboratoire Heudiasyc UMR CNRS 6599
Universite de Technologie de Compiegne
60205 Compiegne Cedex, France
{ada.gogu, dnace, ychallal}@hds.utc.fr
Abstract—This work addresses the problem of designing theoptimal network configuration for Wireless Sensor Network(WSNs). Specifically, given a surveying area our goal is todetermine the parameters of network configuration such thatsome objectives are met. We begin by studying the config-uration for a linear network and then extend our work tothe two dimensional network area. For the linear networkcase we target two objectives: i) minimizing the overall energyconsumption ii) and guarantying the node energy consumptionfairness. Then we discuss the overall energy consumptionminimization problem where the transmission energy takesonly some given discrete values. Next, we study the networkarea configuration problem and show how the computationmethod given for the linear case can be extended to handle it.The solution method for these problems is based on dynamicprogramming algorithm. This method has a low complexitycompared to other methods proposed in the literature andit can be a great help for network designers thanks to itssimplicity in implementation.
Keywords-WSN, Dynamic programming, network configura-tion;
I. INTRODUCTION
WSN technology needs to be carefully managed in order
to meet applications’requirements due to the stringent con-
straints such as energy, bandwidth, memory, etc. Optimiza-
tion techniques and strategies are applied at physical, access
control, network and application layer of this technology
to improve its performance. However a primary concern of
wireless sensor networks is the energy constraint. A careful
design of node deployment scheme can be a very effective
optimization mean for achieving energy conservation and
therefore the network lifetime extension.
In this study, we focus on the design of the optimal
network configuration strategy to meet a given objective.
We will examine a many-to-one wireless sensor network
where information is collected periodically from the sensors
to the base station. Such applications can be data monitoring
for agriculture, weather forecast, etc. Firstly, we consider
a linear network where nodes are placed uniformly. The
whole network will be divided in linear segments that we
will call cells. The configuration problem needs to determine
the number of cells and the size of each of them such
that: i) the total energy consumption is minimized, and ii)
every cell consumes the same energy. Moreover, the system
configuration problem for energy minimization is adapted
for the case where the sensor power transmission takes
discrete values. Next, we study the same problem for a two
dimensional network area.
The rest of the paper is organized as follows. Section II
describes briefly a state of the art for the problem in hand
and revisits some important questions of our previous work.
In section III, we present the linear network model and the
problem definition. Section IV is dedicated to the solution
method for the linear network configuration problem, the
respective algorithms for different optimization criteria and
their results. The two dimensional network configuration
problem, the solution method and results are discussed in
section V. We conclude the paper in section VI.
II. BACKGROUND
A. State of the art
Some similar problems discussed in the literature are
known as node sensor placement and network division for
data gathering sensor networks. The node sensor placement
problem intends to deterministically place the sensors in
order to meet some requirements such as minimization of
energy consummation [3], [4], fault tolerance [5], lifetime
maximization for event-driven networks [2], throughput sta-
bility under attacks [10]. While the network division problem
asks to divide the network in grids or coronas (if the surface
of the network is assumed circular) such that it meets certain
objectives.
In [3] the problem of node placement for minimizing
the total power consumption is formulated as a constrained
multi-variable nonlinear programming problem, which is
solved at optimality. However the considered energy con-
sumption model is very reduced and the authors do not take
in consideration the receiving or idle energy consumption. A
more realistic energy model is taken in account in [2] where
lifetime maximization is achieved through the minimization
of power consumption for each node. Hence, given distance
d from the Base Station (BS) the problem asks to define the
positions and the total number N of nodes which maximize
the network lifetime. They have formulated a multi-variable
nonlinear optimization problem to get the node position for
any given N . Then, they solve numerically this problem
and use the obtained solution to optimize the number Nof nodes. Whereas [5] discusses the problem of relay node
2011 International Conference on Network-Based Information Systems
978-0-7695-4458-8/11 $26.00 © 2011 IEEE
DOI 10.1109/NBiS.2011.14
22
B a s e S t a t i o n B a s e S t a t i o n
a ) b )
S o u r c e S o u r c e
1
2
3
n
2
3
n
1x
x
x
x
S c e n a r i o S c e n a r i o
Πni=1
ki
knd d
Figure 1: The multihop transmission for linear network
placement to ensure the k connectivity between the nodes
in the single and two-tiered network with one or more base
stations. The authors conclude that several (4) variants of the
problem are NP-hard and propose approximation algorithms.
The time complexity is O(n4) where n is the number
of sensor nodes. Despite the heaviness of the proposed
methods, it is shown that allocating nodes properly brings
to performance improvements of both network lifetime and
total power consumption.
Another similar problem concerns the network configura-
tion one. Zhang and Shen in [12] propose a method to divide
the network in coronas, then the coronas in subcoronas and
finally the subcoronas in zones such that the load for every
zone is balanced. The problem of finding the optimal number
of coronas is modeled as an optimization problem and a
simulated annealing algorithm is proposed. Then the division
of coronas in subcoronas and zones is solved iteratively.
This network division scheme assumes that the coronas and
the subcoronas have the same width and the number of
zones is the same for each of them. While in [11], Tsai
et al. propose an algorithm for node deployment strategy to
determine the number of nodes in every zone to handle the
overall generated and relayed traffic for energy balancing.
This method provides a non-uniform random deployment,
with a density which increases closer to BS.
B. Previous work
In our previous work [4], we discussed the problem of
deployment/transmission range assignment scheme in order
to maximize the lifespan of the whole network through total
energy savings. Given a source sensor in a distance d from
the BS, the problem asks to determine the number of hops
and the sensor placement between the sensor and the BS
such that the total energy consumption is minimized. This
model is called a linear network. The energy model that
we use to estimate the energy consumed by the sensors
is proposed in [6]. We note ETX the energy used for
transmitting, ERX for receiving and Eidle for the idle state
as in equation (1).
ETX = (Eelec + Eampdp) · β
ERX = Erec · βEIdle = c · Erec · β
(1)
In these equations β gives the bit rate of the radio and p is
the path loss exponent which varies between 2 and 4. Eelec
is the energy/bit consumed by the transmitter electronics,
Eamp is the energy/bit consumed by the amplifier, Erec
is the energy/bit consumption of the receiving circuitry
and c is a constant. We take the value of c equal to 1
because the energy consumption in idle state is very close
to energy consumption for receiving. For this problem we
examine two scenarios: a) the source sensor node does the
transmitting and all the intermediate sensor nodes act simply
as relaying nodes and b) the intermediate sensors add their
own information before relaying. Figure 1 illustrates the
two cases of the linear network. While the first problem is
quite simple, the second one becomes more complex. Based
on dynamical programming method we found an analytical
formula to solve the second problem which gives the optimal
number of nodes and their respective distances from BS. As
will be shown there is a property which permits us to obtain
this formula.
Let’s suppose that xn is the optimal location of the last
relay node in the linear network where n relay nodes are
employed. Given the energy formula (1), it can be shown
that the value kn = xn/d does not depend on distance d.
Going further, this can be generalized for all intermediate
sensors xp|1 ≤ p ≤ n. The following proposition holds:
Proposition 1. For a given n, the ratios kp = xp/d, 1 ≤p ≤ n, are the same for any distance d.
Proof: We can show this result by mathematical in-
duction on the number of n. Let’s take n = 1. Clearly,
the energy function associated with the transmission through
an intermediate sensor placed at x1 takes its minimum for
derivative set to 0. The energy function is of form
α1(x1)p + α2(d− x1)
p + α3 (2)
where αi are constants. Then, the derivative gives
α1p(x1)p−1−α2p(d−x1)
p−1. It is straightforward that this
function takes 0 for some k1 = x1
dthat does not depend
on d. The reasoning can be followed for some higher nassuming that the result holds for n − 1. Indeed, from the
recurrence assumption we can deduce that also for n the
energy function is similar to the energy function given in
(2) except of using other constants. For this, we decompose
the energy in two parts, the first captures the optimal energy
needed for the transmission through n − 1 sensors from 0to xn and the second from the last sensor xn to d. For
the first part we consider the optimized energy, that is the
23
−60 −40 −20 0 20 40 60−60
−40
−20
0
20
40
60
dR
Base Station
Y0
Sensor node
l
Figure 2: A sensor network with perfect traffic aggregation
first n− 1 sensors are optimally placed for the distance xn.
The recurrence hypothesis allows then to obtain a simplified
formula for the minimum energy used for the transmission
from 0 to xn, which is of form α′
1(xn)p+ c. Then, the total
energy can be expressed as α′
1(xn)p + α
′
2(d − xn)p + α
′
3,
where α′
i are constants. Then, we can apply the same schema
as for n = 1.
The linear case is extended for a two dimensional network,
where we assume to have N nodes placed at a distance dfrom BS, as in Figure 2. Also we assume a perfect aggre-
gation mechanism where each intermediate node transmits
only a certain amount of information despite what it receives.
This problem asks to compute the placement and the number
of relaying sensor nodes such that all the information sensed
by the N sensor nodes may be transmitted with a minimum
energy.
Our method gives optimal solution for the problem, how-
ever the above model remains mainly theoretical because
in real scenarios a node may receive and transmit a certain
traffic basing on network traffic pattern. Nevertheless, it can
be useful to calculate lower bounds of energy consumption.
Our contributions in this paper are twofold. First, we adapt
the mathematical model used for the relay sensor placement
problem to solve the sensor network configuration one which
has certain practical interests. Second, we propose a method
to solve optimally the problem of network configuration
for different cases of traffic load and to reduce its time
complexity.
III. PROBLEM FORMULATION AND LINEAR NETWORK
MODEL
A. Problem statement
We consider a linear network with a length d from the
BS, where the nodes are uniformly distributed, as in Figure
3. The network will be divided in cells. For any cell, there
will be only one node, called cluster head, which receives the
information from the other nodes of the same cell and from
the cluster head of the upstream cell. Then, the received
data will be transmitted towards the cluster head of the
Figure 3: Network model
downstream cell. We employ in this work a aggregation
model proposed by [8] which has the form y = mx + cwhere x and y are the input and output information quantity
respectively, m varies in [0, 1] and c is a constant. More
specifically, in this model we have assumed that the cluster
head will not aggregate its information, and thus m will be
equal to 1. The network model considered here is proposed
in [9]. We assume that each node generate α Erlang of data,
and the radio data rate is β bps. The nodes are uniformly
distributed with a density nd. The ith cell in the network
(Figure 3) will consume an energy ETX for transmitting
the information which is computed according to the equation
(3).
ETX(i) = (Eelec +Eamp(xi − xi−1)p) · xi · nd · α · β (3)
Whereas the energy used by the ith cell for receiving is
expressed by the equation (4).
ERX(i) = Erecxi−1 · nd · α · β (4)
Before a node goes to sleep mode, it can be either in
transmitting, receiving or idle state. Therefore, the energy
consumed in the idle state for a node of the ith cell is:
EIdle(i) = Erec
(
1− 2xi−1 · nd · α)
· β (5)
The network configuration problem consists in determining
the optimal number of cells and the respective length of each
cell such that i) the total energy consumption is minimized
(we will call this Strategy 1) and ii) every cell in the network
consumes the same energy or in other words we need to
ensure the energy fairness between the cells (Strategy 2).
So far we have assumed that the node transmission energy
depends on its communication distance. Since real sensors
have discrete levels of power transmission, we consider the
case of minimizing the total energy consumption of nodes
in the linear network using only some given discrete energy
values.
IV. SOLUTION METHOD AND NUMERICAL RESULTS FOR
THE LINEAR NETWORK MODEL
Our goal is to determine the parameters of network
configuration such that a given objective is met. Specifically,
24
given the linear network size, we need to find the optimal
cells number and their respective distances which guarantee
a lower bound of energy consumption (in IV-A1), the energy
balancing between cells (in IV-A2) or energy consumption
minimization for discrete energy levels (in IV-A3). As it can
be seen, the number of possible combinations is exponential
for any of these problems, therefore we propose a method
based on dynamic programming. Dynamic programming is
a sequential approach [1] for optimizing a given objective
function. The problem is thus broken down into stages and
the aim at every stage is to select the optimal decision so that
the objective is optimized over the total number of stages.
A. Optimization Criteria
1) Total energy minimization: Here the objective is to
minimize the total energy consumption for the linear net-
work presented in III-A. Our problem is separated into stages
determined by the number of cells. At some stage n we
determine the minimum energy necessary to transmit the
data by using the energy values obtained from the previous
stage. Hence, the optimal solution for an instance of a
problem with n cells and a given distance d can be seen as
equivalent to the optimal solution for some distance x < dand n − 1 hops, plus a final hop from this point (x) to the
BS. Algorithm 1 describes these steps. The parameter ∆
Algorithm 1: Strategy 1
Input: Distance d; node density nd
Output: Optimal number of network division n;
Optimal cell sizes x; Minimal energy E.
Initialization : {Eny |y ≤ d}, n = 1;
do {For all 1 ≤ y ≤ d calculate:
En+1y = min0<x<y
{
En−1(x) + ∆}
n → n+ 1} while(En−1
d ≤ End )
display n,E
given in Algorithm 1 is calculated according to formula (6).
∆ = (Eelec + Eamp(x− y)p) · y · nd · α · β+Erec · nd · α · β · x+ Erec · (1− 2 · x · α) · β
(6)
Note that Eny represents the total energy (optimized) con-
sumption where the network length is y and it is divided
into n cells.
2) Energy fairness: Nodes in WSN are often prone to
failures due to energy depletion. This phenomena happens
more frequently between the nodes close to the base station,
which is known as the hole problem [7]. The failure of
such nodes may culminate into disconnections, network
partitioning, and eventually stopping the whole network
operation. In order to avoid this, we want to define the
network configuration parameters such that each node in the
network consumes the same energy despite the amount of
received and generated data. Assuming that the network will
be divided into k cells, the total energy consumption for cell
i is given by (7).
E = (Eelec + Eamp(xi − xi−1)p) · xi · nd · α · β
+ Erec · nd · α · β · xi−1
+ Erec · (1− 2 · xi−1 · α) · β(7)
It is then possible to formulate the problem through a system
of non linear equations. Instead, we propose the Algorithm
2 also based on dynamic programming method to solve this
problem.
Algorithm 2: Strategy 2
Input: Distance d; node density nd.
Output: Optimal number of network division n;
optimal cell sizes x; Minimal energy E.
Initialization : {Eny |y ≤ d}, n = 1;
do {For all 1 ≤ y ≤ d calculate:
Calculate
En+1y |y ≤ d = min0<x<y
{
En(x)− E(y − x)}
n → n+ 1
} while(En−1
d ≤ End )
display n,E
In this case, Eny represents the energy consumed by one
cell1 when the network of a length y is divided in n cells.
E(y − x) = (Eelec + Eamp(y − x)p) · y · nd · α · β+Erec · nd · α · β · x+Erec · (1− 2 · x · α) · β)
(8)
The initialization phase of the algorithm requires O(d/δy)computation to construct the vector En
y , where δy is the step
of distance discretization. The calculation procedure will be
repeated n times and costs at most O(d/δy)2. Finally, the
computation complexity of the algorithm is O(n · (d/δy)2).3) Energy minimization with discrete transmission levels:
In the above models, the transmission sensor energy is
computed according to the exact distance of transmission.
This model does not reflect the reality as node transmission
energy takes values in a given discrete interval. In this
section, we show how to modify Algorithm 1 to handle this
case.
B. Numerical results
We have tested our algorithms for the optimal strategy
1 and 2 and compared their results with the best case of
uniform cell division. The uniform cell division divides the
1every cell will consume the same energy
25
Algorithm 3: Energy minimization with discrete trans-
mission levels
Input: Distance d, node density nd, the list of energy
power levels E = {e1, e2, ...en}, the
corresponding list of the respective distances
L = {l1, l2, ..lm}.
Output: Optimal number of network division n,
optimal cell sizes x, Minimal energy EInitialization : {En
y |y ≤ d}, n = 1
do {For all 1 ≤ y ≤ d calculate:
En+1y = min 0<x<y
⌈y−x⌉∈L
{
Enx + x · Erec · nd · α · β +
Erec · (1− 2 ·x ·α ·nd) · β+E(⌈y − x⌉) · y ·α · β ·nd
}
n → n+ 1} while(En−1
d ≤ End )
display n,E
network in number n of equal cell size and the best case
corresponds to the n that minimizes the energy value. The
results of total energy consumption for the three cases with
respect to the distance are shown in fig 4. For this simulation
we set the simulation parameters as detailed in table I.
Table I: Simulation parameters for the linear network
Type Parameter Value
NetworkNetwork Length 107 ∼ 400m
Node distribution Uniformly
Node density 0.16
ApplicationData generation rate β 485 bps
Data traffic per node α 0.003 Erlang
Compression ratio m = 1 | c = α · β
Energy
Eelec 50nJ/bit
Eamp 100pJ/bit/m2
Erec 50nJ/bit
p 2
L = {l1, l2, ·lm} [10, 20, . . . 80] in meters
E = {e1, e2, ·em} [0.06, 0.09 . . . 0.54] · 10−6J/bit
Method Distance discretization δy = 1
We notice that strategies 1 and 2 outperform the uniform
one saving 31% of the energy or in other words increasing
130% the network lifetime. On the other part, we notice
that Strategy 1 gives indeed a lower bound for total energy
consummation but it is quite near to this obtained from
the strategy 2. So, for e.x. in 400m it saves 3% in energy
compared with strategy 2.
For a range of distances we studied the optimal number
of cells provided by the two algorithms and the results are
100 150 200 250 300 350 4000
1
2
3
x 10−4
distance D from the Base St.
Glo
ba
l E
ne
rgy E
(J)
Energy consummation
Strategy 2
Strategy 1
Uniform energy
Figure 4: Energy consumption
0 50 100 150 200 250 300 350 4001
2
3
4
5
6
7
Network length (/m)
Nu
mb
er
of
ce
lls
Optimal number of cells versus distance
Strategy−2
Strategy−1
Figure 5: Optimal number of cell division versus distance
shown in Figure 5. In fact, we observe that the number of
cells for the two strategies changes only for some small
interval of distances, however the cells division is not
necessarily the same even for the same number of cells.
The table II gives the cells division beginning from the
one closest to the Base Station. We notice that the cells
division for the distance d = 200m is quite different even
when the optimal number of cells is 3 for both. Table III
presents the energy consumption for the two surveyed cases
of d = 200m and d = 400m from which we can distinguish
the energy differences between the two strategies.
As expected, for the energy discretization problem the
sensors spent more energy. For this simulation we assume
that each node can transmit to distances L = l1, l2, . . . lmusing the respective energies E = e1, e2, . . . em. Figure
6 shows that the continuous energy value still remains an
acceptable approximation of energy behavior.
26
cell sizesStrategy 1 Strategy 2
d=200m d=400m d=200m d=400m
cell 1 57 52 42 39
cell 2 65 53 48 41
cell 3 78 59 110 44
cell 4 66 47
cell 5 77 52
cell 6 93 60
cell 7 117
Table II: The optimal configuration for the linear network
Energy (J)
Strategy 1d=200m 9.2842 · 10−5
d=400m 2.6981 · 10−4
Strategy 2d=200m 1.0058 · 10−4
d=400m 2.7893e · 10−4
Table III: Energy
V. TWO-DIMENSIONAL SENSOR NETWORK
A. Problem statement
The second problem considered here is the two-
dimensional sensor network configuration one. We assume
that the nodes are uniformly distributed in a circular moni-
toring area. There is only one sink positioned in the center
of the zone.
The sensors will transmit their information according
to a many-to-one traffic pattern using a multihop scheme.
Minimizing total energy consumption can be achieved by op-
timally dividing the area in concentric coronas C1, C2, ...Cn
centered at the sink and further, dividing every corona in
a given number of zones. Figure 7 gives an example of
network division with 3 coronas containing respectively in
100 150 200 250 300 350 4000
0.5
1
1.5
2
2.5
3
3.5x 10
−4
Distance (/m)
En
erg
y (
/J)
Energy Discretization
Discrete Overall Energy
Overall Energy
Figure 6: Energy disretization
−60 −40 −20 0 20 40 60−50
−40
−30
−20
−10
0
10
20
30
40
50
Figure 7: Two dimensional network
1, 4 and 8 zones. Moreover, the length of every corona
can be varied. For each zone it will be only one sensor
that will receive the information from other sensors of the
zone, aggregate and transmit it to the next zone closer to
the BS. This sensor will be called a cluster head of the zone
and every node of the zone may play this role. Notice that
we can easily shift from this model to the sensor specific
sensor placement, which asks to determine the cluster head
position, by adapting the radio transmission range affected
to each zone. According to the aggregation model proposed
in the section III-A we will analyze the case2 with m = 0.
Indeed, if we suppose that the sensor information will not
be aggregated by the cluster heads the problem would be
reduced to the linear network case because the sensor will
select the shortest path to transmit the information. Finally,
for the network model as described above, the questions are:
• What is the optimal number of coronas that achieves
the energy minimization of the whole network? What
are their respective radii?
• What is the optimal number of zones that a corona is
divided in?
Here we assume that it will be a cluster head for each
zone that will aggregate the traffic and will transmit
toward the cluster head of the zone closer to the BS.
B. Minimum energy consumption
Unlike the linear network case, the problem for a two
dimensional network with traffic aggregation becomes more
complex to handle. We opt for a method based on parameter
discretization of the radius and the number of zones for each
corona respectively, and apply the dynamic programming
principle. Hence, for a given sensor network radius R with
uniform node distribution nd and a constant traffic α · βgenerated by each sensor, we apply the Algorithm 4.
2A node will totally aggregate the information and transmit only a fixedamount of information
27
Algorithm 4: Network area configuration
Input: Node density nd, Network area radius R,
Generated traffic per node α · βOutput: Minimum energy E; Optimal number of
coronas n; Coronas’ radius x; Number of
zones for each corona P .
Initialization : {End |d ≤ R}, n = 1
do {Calculate En+1
d,N matrix
En+1
d,N = minx≤d,P≤N
{
Enx,P + P · f1(d, x,N, P )
+f2(d, x,N, P )}
(9)
set Ond,N = x, P
n → n+ 1
} while(En−1
R,N0≤ En
R,N0)
display n,E
According to this algorithm we begin by computing the
first vector E1 containing values of E1d |d ≤ R using the
formula (10).
E1d |d ≤ R = (Eelec + Eamp · d
p) · nd · π · d2 · α · β (10)
This corresponds to the case when nodes transmit directly
to the BS. In the next step, the number of coronas is two
and we calculate all {E2d,N |d ≤ R,N ≤ N0} values. The
parameters used in equation (9) are computed as follows:
f1(d, x,N, P ) = (Eelec+Eamp·Rp(d, x,N, P )+Erec)·α·B
(11)
where R(d, x,N, P ) is such that:
R2(d, x,N, P ) = d2 + x2 − 2 · d · x · cos((NP
− 1) · π
N)
This is the energy used by the cluster head of each zone of
the nth corona to transmit the information towards the next
cluster head closer to the BS.
f2(d, x, n, p) = ((Eelec + Eamp ·Rp(d, x,N) + Erec) · nd
·π · (d2 − x2) · α · β(12)
with R(d, x,N) such that:
R2(d, x,N) = d2 + x2 − 2 · d · x · cos(π
N)
Equation (12) gives the energy consumption spent by all the
sensors in a corona due to the intra zone transmissions3.
We continue like this with {End,N |d ≤ R,N ≤ N0} for
subsequent values of parameter n. We stop the calculations
when we reach some n such that EnR,N0
≤ En+1
R,N0. The
3Intra zone transmissions include the receiving and transmitting opera-tions inside a zone between the cluster head and the other sensors in thezone.
solution can be easily built by backtracking the stored
values Ond,N . At the end of the algorithm we obtain the
optimal number of coronas to achieve a minimum energy
consumption, as well as the number of zones for each
intermediate corona and the corresponding radius.
C. Numerical results
For the simulations, the nodes are uniformly deployed in
the circular area and generate a constant traffic. The cluster
head of each zone aggregates all the information that it
receives and transmits a constant traffic to the closest cluster
head. Table IV lists the system configuration parameters in
details. For these parameters, some experimental results of
the Algorithm 4 are presented in table V. As we can observe,
for each corona the algorithm determines its respective
radius and number of zones, which both decrease uniformly.
Table IV: Simulation parameters
Type Parameter Value
NetworkArea radius 100m
Node distribution Uniformly
Node density 0.02
ApplicationData generation rate β 100 bps
Data traffic per node α 0.003 Erlang
Compression ratio m = 0 | c = α · β
Energy
Eelec 50nJ/bit
Eamp 100pJ/bit/m2
Eelec 50nJ/bit
p 2
Cell sizesAlgorithm results
Radius of each Number of zonescorona from the BS
corona 1 100 27
corona 2 90 24
corona 3 80 21
corona 4 70 18
corona 5 60 15
corona 6 50 13
corona 7 40 10
corona 8 30 7
corona 9 20
Table V: Optimal configuration for the network area
The energy consumption of the whole network for differ-
ent network radius and node density changes is presented
in Figure 8. We notice that small changes in node density
28
50 100 150 200 250 3000
0.5
1
1.5
2
2.5
3
3.5
4
4.5x 10
−4
Area Radius (/m)
En
erg
y (
/J)
Node density = 0.04 m2
= 0.028 m2
= 0.02m2
Figure 8: Energy versus network area radius
may greatly influence the energy consumption of the net-
work. Moreover, we observe that the network configuration
solution, as the one depicted in table V, is quite sensitive
toward these changes.
VI. CONCLUSIONS
In this paper we have analyzed the network configuration
problem for many-to-one WSN. For a given network where
the nodes are uniformly distributed, the problem requires
to determine the optimal number of cell division, and their
respective cell sizes such that some energetic objectives are
met. We show that appropriate configuration of WSN, which
addresses the cell organization and the respective power
assignments, leads to considerable savings in energy. Our
algorithms which intend to i) minimize the total energy
consumption and ii) assure the energy node consumption
fairness are based on dynamic programming approach. We
also show how this method can be adapted even for the
discrete case of energy consumption minimization problem.
Next, the linear network case is extended to the two dimen-
sional network one. In general, we observe that dynamic
programming is an effective method to handle the trade-offs
between the parameters for some variants of the network
configuration problem. It provides optimal solution of the
problem for a given objective function and reduce the com-
putation complexity compared with other methods proposed
in literature. Moreover, the method can be adapted for other
similar energy models consumption without increasing the
computational complexity of the solution. In the future, we
will extend our work to adapt and implement this method
for the two dimensional sensor network with randomly node
distribution and test it in real scenarios.
REFERENCES
[1] R. Bellman (1957). Dynamic Programming. PrincetonUniversity Press, ISBN 0486428095, 2003.
[2] Y. Chen, C.N. Chuah, and Q. Zhao. Sensor placement formaximizing lifetime for unit cost in wireless sensor networks.IEEE MILCOM, 2:1097–1102, 2005.
[3] P. Cheng, C.N. Chuah, and X. Liu. Energy-aware nodeplacement in wireless sensor networks. IEEE GLOBECOM,5:3210–3214, 2004.
[4] A. Gogu, D. Nace, and Y. Challal. A note on joint optimaltransmission range assignment and deployment for wirelesssensor networks. IEEE Networks, pages 1–6, 2010.
[5] B. Hao, H. Tang, and G. Xue. Fault-tolerant relay nodeplacement in wireless sensor networks. HPSR, pages 246–250, 2004.
[6] W. R. Heinzelman, A. P. Chandrakasan, and H.Balakrishnan.Energy efficient communication protocol for wireless mi-crosensor networks. Proceeding of the 33rd Hawaii Inter-national Conference on System Sciences, 2000.
[7] J. Li and P. Mohapatra. Analytical modeling and mitigationtechniques for the energy hole problem in sensor networks.Pervasive and Mobile Computing, pages 233–254, 2007.
[8] V.P. Mhatre, C. Rosenberg, D. Kofman R. Mazmudar, andN. Shrof. Design guidelignes for Wireless Sensor Networks:Communication, Clustering and Aggregation. Ad Hoc Net-works, 2:45–63, 2004.
[9] Q.Gao, K.J. Blow, D.J. Holding, I.W.Marshall, and X.H.Peng. Radio range adjustment for energy efficient wirelesssensor networks. Elsevier, Ad Hoc Networks, pages 75–82,2006.
[10] X. Wang, W. Gu, S. Chellappan, K. Schosek, and D. Xuan.Lifetime optimization of sensor networks under physicalattacks. IEEE ICC, 5:3295–3301, 2005.
[11] Y.Tsai, K. Yang, and S. Yeh. Non-uniform node deploymentfor lifetime extension in large-scale randomly distributedwireless sensor networks. AINA Proceedings, pages 517–524,2008.
[12] H. Zhang and H. Shen. Balancing energy consumption tomaximize network lifetime in data-gathering sensor networks.IEEE Trans. on Par. and Distrib. Sys., 20:1526–1539, 2009.
29