You are on page 1of 22

Page 1 of 22 IET Wireless Sensor Systems

This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

An Efficient Scalable Sensor Node Placement Algorithm for Fixed


Target Coverage Applications of Wireless Sensor Networks

Arouna Ndam Njoya1,* , Christopher Thron2 , Jordan Barry2 , Wahabou Abdou3 , Emmanuel Tonye4 ,
Siri Konje Lawrencia Nukenine1 , Albert Dipanda3
1
University Institute of Technology, University of Ngaoundéré, Ngaoundéré, Cameroon
2
Department of Mathematics, Texas A&M University-Central Texas, Killeen, USA
3
Laboratoire Electronique, Informatique et Image, Univ. Bourgogne Franche-Comté, Dijon, France
4
National Advanced School of Engineering, University of Yaoundé I, Yaoundé, Cameroon
5
University Institute of Technology, University of Ngaoundéré, Ngaoundéré, Cameroon
*
ndanjoa@gmail.com

Abstract: Large applications of sensor networks, such as environmental risk monitoring, require
the deployment of hundreds or even thousands of nodes. This paper proposes and implements a
novel stochastic physics-based optimization algorithm that is both efficient (guarantees full target
coverage with a reduced number of sensors) and scalable (meaning that it can be executed for very
large-scale problems in a reasonable computation time). The algorithm employs “virtual sensors”
which move, merge, recombine, and “explode” during the course of the algorithm, where the
process of merging and recombining virtual sensors reduces the number of actual sensors while
maintaining full coverage. The parameters which control sensor merging and explosion are varied
during the algorithm to perform the same function as an annealing schedule in simulated annealing.
Simulation results illustrate the rapidity and the effectiveness of the proposed method.

1. Introduction

In recent decades, technological progress in transmission of information and in embedded systems


have given rise to a proliferation of applications of wireless sensor networks. Once almost exclu-
sively reserved for military use, sensor networks are now used in health monitoring [34], traffic
monitoring, environmental risks monitoring, prevention of disasters, and so on [35].
Despite the declining price of sensors, for applications requiring hundreds thousands of sen-
sor nodes the cost of sensor networks remains high [36, 1, 2, 3]. In such applications sensor
node location is a critical issue, because it impacts both the cost and performance of the sensor
network [36, 4]. This indicates a growing need for efficient algorithms with reasonable execu-
tion times which can compute sensor deployments for very large networks that satisfy coverage
requirements while minimizing the number of sensors.
Generally, sensor nodes are deployed either randomly or deterministically according to the
nature of the monitoring area. When the region is not well known and inaccessible, random de-
ployment is used [5]. In this case, a systematic placement of sensors that minimizes the number
of sensors is not possible. On the other hand, deterministic deployment of sensors in specific
positions may be used when the region is accessible and its characteristics are known [5, 6, 7, 4].

1
IET Review Copy Only
IET Wireless Sensor Systems Page 2 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Although coverage requirements may vary considerably from application to application, the
requirements typically fall into one of three categories [5, 6, 7, 4]:
• Area coverage, which focuses on complete coverage of the entire monitoring area [8, 9, 10,
11].
• Target coverage, whose objective is to monitor specific points of the monitoring area. Target
coverage can be further divided into three categories: (1) Simple coverage, where each target
should be covered by at least one sensor; (2) k-coverage, which requires that at least k sensors
to cover each target; (3) Q-coverage, where the number of sensors required to cover each
target is specified individually for each target [7, 12, 13, 14].
• Barrier coverage, that seeks to limit intrusion into a bounded area. In this case, sensors are
required to cover the boundary of the restricted region [15, 16, 17].
This paper focuses on the deterministic placement of sensor nodes for simple target coverage.
The objective is to minimize the number of deployed sensors while maintaining coverage of all
targets. For this purpose, an algorithm that is based on an analogy with the physical motion of
astronomical bodies is introduced: an efficient sensor placement is attained through a process in
which potential sensor locations move and combine under the influence of “forces” exerted by
other potential sensor locations.
Our algorithm is noteworthy in three very important respects. First, unlike previous algorithms
it does not fix the number of sensors, but rather reduces the number of sensors during the course of
the algorithm’s execution by means of “sensor combine” operations. Second, the algorithm guar-
antees full target coverage, unlike many algorithms which attempt to obtain the “best” coverage
with a fixed number of sensors. Third, execution time scales linearly with the size of the problem,
which means that in can applied to extremely large systems.
The rest of the paper is organized as follows. Section 2 reviews previous work on the sensor
placement problem. In Section 3 the problem is formally defined. The proposed method is given in
Section 4. Section 5 presents the effectiveness of the proposed method through simulations results.
Section 6 summarizes our conclusions.

2. Related Work

The sensor deployment problem for target coverage has been shown to be a NP-hard optimization
problem [18, 5]. Existing approaches fall into three categories [5]: exhaustive search, discrete
optimization based approaches, and approximation algorithms.
Exhaustive search consists of enumerating all possible sensor placements and selecting the best
one(s). This technique is only practical for very small problems, because the computation time of
the exhaustive search grows exponentially with the number of targets [5, 19].
Discrete optimization-based approaches reformulate the problem as an integer-valued con-
strained optimization problem [19, 20, 21, 22]. It is possible to formulate the sensor placement
problem as an integer linear program [20, 21, 23], which can then be solved with conventional
solvers. Once again, a high computational effort is required to compute the solution in larger sys-
tems [6, 5, 20, 23, 37]. Computation time can be decreased by decomposing the original problem
into sub-problems that are solved independently and then merged to find an overall solution. How-
ever, these methods cannot guarantee optimality of the merged solution [5]. Some researchers have
proposed adaptive heuristics to approximate the solution of the problem [20, 21, 23, 14].

2
IET Review Copy Only
Page 3 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Approximation algorithms are heuristics that try to find a nearly optimal solution within a rea-
sonable execution time. Such algorithms are typically used when exact methods are impractical.
Many approximation algorithms based on different mathematical techniques have been developed
for various cases of the sensor node placement problem. Techniques employed include greedy
algorithms [24], simulated annealing [25], genetic algorithms [9, 26, 27, 28, 29, 30, 31], ant
colony optimization [4], artificial bee colony and particle swarm optimization [7]. Many of these
techniques are inspired by analogies with physical or biological processes.
In keeping with previous work, we take a heuristic approach due to the enormous computational
complexity of exact solutions for simple target coverage involving very large numbers of targets.
We also draw inspiration from a physical process, namely the motion of astronomical bodies under
the influence of mutual gravitational force.

3. Sensor deployment for target coverage

3.1. Sensor coverage model


Let us consider static sensors with boolean disk coverage model and identical sensing radius rs .
This model corresponds to the coverage function defined in (1):


1 if d(s, t) ≤ rs
f (s, t) = (1)
0 otherwise,

where d(s, t) is the Euclidean distance between a sensor s and a target t. With this model, a target
is considered to be covered by the sensor if it is located inside the disk. Otherwise the target is not
covered.

3.2. Problem Definition


We formulate sensor deployment for target coverage in the following manner. There are m targets
T = {Tj , j ∈ {1, ..., m} | Tj is the target indexed by j} where each target has a definite location
within a given rectangular region. The objective is to find the locations of n (n ≤ m) sensor nodes
S = {Si , i ∈ {1, ..., n} | Si is the sensor indexed by i} in such a way that, the number of sensors
can be minimized and each target is monitored by at least one sensor node.

4. Proposed Method

4.1. Overview
The proposed method is based on the use of virtual sensors. These can be moved, merged, recom-
bined or unmerged in order to provide the best deployment. Initially, each target is covered by one
virtual sensor, so that at the beginning of the algorithm there are as many targets as virtual sensors.
It is evident that many sensors should cover multiple targets if we want to obtain a covering that
minimizes the number of sensors. To achieve this, the proposed algorithm allows the fusion of
close virtual sensors if this operation guarantees that all targets will be covered. This operation is
called “Merge”. Besides merging, the algorithm include other operations to dynamically change
the cover in order to explore efficient coverage options. Two operations denoted by “Grab” and

3
IET Review Copy Only
IET Wireless Sensor Systems Page 4 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

“Release” shuffle targets between existing sensors. In order to guarantee that the algorithm does
not get stuck in a suboptimal configuration, an “Explode” operation is also included that restores
a combined sensor back to the original state where each target is covered by only one sensor. In
addition to these operations that reconfigure the sensor coverage, a quasi-physical motion is intro-
duced in order to place sensors in favorable positions for merging or other operations. This motion
is determined by velocity and acceleration parameters which are specified for each active sensor.

4.2. Core algorithm


The algorithm consists of four modules: an initialization phase, an optimization engine, a result
analyzer and a returning phase. These four components collaborate to determine the minimum
number of sensors required to cover all targets in a given region. The flowchart in Fig. 1 presents
these modules and their interactions. The algorithm starts with the random generation of the set
of targets, the assignment of the value to rs , the initialization of the simulation parameters and the
creation of sensor data structure. Once entering the loop, some of these parameters are passed to
the optimization engine. The latter determines the merge distance of valid sensors, computes the
acceleration (change in velocity) and identifies candidate merges (Sensor Recompute), examines
sensors one by one, to see if there are any others close enough to merge (Sensor Merge and Re-
combine). Thereafter, the Sensor Update procedure is executed to check whether the sensor moved
too far and its targets are no longer covered. If so, it moves the sensor back along the velocity vec-
tor towards the previous position. The list of valid sensors performed by the optimization module
is transmitted to the result analyzer that computes the current number of sensors and examines if
there is an improvement of the minimum number of sensors. The loop terminates when there is no
more improvement after wait period or the stop condition is met. Before returning the minimum
number of sensor and the sensor locations, the last module checks for double coverage and remove
redundant sensors. In Fig. 1, the “Recombine?” and “Explode?” conditions are used to recombine
or explode valid sensors according to their respective probabilities. The parameters in Table 1 are
used to set up the proposed approach.
Table 1 Parameters and their definitions
Parameter Description
numIterM ax The maximum number of total iterations.
waitP eriod The period after which the algorithm terminates if no improvement is made.
nP lanet The number of consecutive iterations with “planetary” dynamics.
In “planetary” dynamics, sensors move as if they were “planets” and combine
when they collide. The default value is fixed to 10.
K The “force constant” in the sensor attraction. When set equal to rs2 /8(where rs
is the sensor radius), two stationary sensors that are a distance of rs from each
other will move 1/8 of a sensor radius towards each other in 1 iteration.
drag This variable keeps the sensors from building up too much speed.
If set equal to 0.5, the sensor will slow down by 1/2 at each time step.
recombineP robability This variable gives the probability of recombining(exchange targets).
the default value is fixed to 0.25.
grabP robability The “grab” probability. The default value is fixed to 0.5. The release probability is
equal to 1 - grabP robability.
explodeP robability The maximum probability that a given sensor will “explode” into separate
sensors (one for each covered target). Explosions give the algorithm a chance
to re-match sensors to targets.

4
IET Review Copy Only
Page 5 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Fig. 1. Flowchart of the proposed method

4.3. Sensor data structure


The sensor data structure contains sensor and target information. Targets are indexed, and asso-
ciated with each target is a virtual sensor which covers that target. Note that only valid virtual
sensors are retained. We designate eight fields for the sensor data structure as summarized in Table
2.
For the rest of the paper, we use Si , s~i , Ti , and ~ti to denote the ith sensor, the position of the ith
sensor, the ith target, and the position of the ith target respectively. The notation d(Si , Sj ) refers to
the distance between sensors i and j, and is equal to ||~si − ~sj ||.
Fig. 2a illustrates an example case of the sensor data structure, in which there are 6 targets
and four valid sensors S1 , S3 , S4 and S6 associated with targets 1, 3, 4 and 6 respectively. The
mayMerge lists of sensors S1 , S3 , S4 and S6 are {2}, ∅, {5, 6}, and ∅ respectively. The isMerged
lists of sensors S1 , S3 , S4 and S6 are {1, 2}, {3}, {4, 5}, and {6} respectively. The tryMerge list for
each sensor depends on the parameter nPlanet. If the merge distance is equal to 2rs as illustrated
in Fig. 2a, the tryMerge lists of valid sensors are ∅ , ∅, {6} and ∅ for sensors S1 , S3 , S4 and S6
respectively.

5
IET Review Copy Only
IET Wireless Sensor Systems Page 6 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Table 2 Sensor data structure fields


Field Description
loc 2 × m matrix containing (x, y) coordinates of sensors.
vel 2 × m matrix containing (x, y) coordinates of sensor velocities.
accel 2 × m matrix containing (x, y) coordinates of sensor accelerations.
valid logical vector of size m. sensor.valid[i] is true if sensor i is valid, which indicates
that sensor i is in fact included in the cover. If sensor i is not part of the actual
cover, the value of sensor.valid[i] is f alse.
pending logical vector of size m. sensor.pending[i] is true if the status of sensor i will be
changed from invalid to valid. Status changes are made at the end of
each iteration to avoid conflicts.
mayMerge cell array of size m. sensor.mayMerge[i] is a list of the indices of candidate sensors
which may possibly cover target i. Attempts to merge are limited to these sensors.
Note that only indices > i are included in the list of possible merges. This is to avoid
attempting the same merge twice. This array is generated during initialization and never changes.
isMerged cell array of size m. sensor.isMerged[i] is a logical vector which indicates which
of the targets in sensor.mayMerge[i] are actually covered by sensor i.
This cell array is recomputed at each iteration.
tryMerge cell array of size m. sensor.tryMerge[i] is a logical-valued vector
which indicates which of the indices of sensors in sensor.mayMerge[i] are close
enough to attempt a merge. This cell array is recomputed at each iteration.

4.4. Sensor Recompute


Algorithm 1 goes through all valid sensors that are within possible merging range of the current
sensor, and tests to see which are close enough to merge. Sensors which pass the test are put on
the tryMerge list. Sensors which do not pass the test are subject to physical motion, and their
accelerations are computed. In other words, Algorithm 1 computes sensor tryMerge lists and
accelerations for valid sensors. The acceleration for sensor i is calculated as
X ~sj − ~si
~ai = −drag · ~vi + K (2)
d(Si , Sj )2
j∈mayMerge(i)∧j is valid

where the drag term is included to ensure that the velocity does not become too large.

4.5. Sensor Merge and Recombine


Algorithm 2 (Fig. 4a) performs the function of merging and recombination of virtual sensors. It
examines sensors one by one, to see if there are any others close enough to merge. The algorithm
also performs “Sensor Explode” and “Sensor Grab/Release” functions, as described below.

4.5.1. Sensor Explode: Fig. 2b illustrates an example of the effects of the Sensor Explode
operation. In this case, the sensor S1 which covers targets T1 , T2 , T3 , and T4 explodes and produces
four sensor nodes Sj located at positions t~j , j = 1, 2, 3, 4. The velocities of these sensors are
computed as follows:

r   t~j − s~1


− s P rob
vj = + (3)
2 (d(S1 , Tj ) + 0.00001) 2
where P rob is a bivariate normal random variable. (3) causes the sensors resulting from the explo-
sion to move away from the location of sensor 1, in order to increase the chances of their combining

6
IET Review Copy Only
Page 7 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

S2
d(S1, S3)>2rs t3
t1

t2 S3
S1
t2 t2
d(S 1, S 6)> s

Explode S1

t1 t1
2r

t4 t3 t4 t3
t5
t4 S4 S1
t6 S4
S3
S6

a b

S25 S25 S12


S2 S6 S2
t25 t25 t6 t12 t6 t12
t34 t34 t2 t8 t13 t2 t8 t13
Release(S2, S12 ) t18 t18
t2 t2 t34 t34
t6 t6 t25 t25
t8 t12 t8 t12
t13 S2 t13 Grab(S2, S6 )
t18 t18 S25 S25
S2 S12 t25 S6 S2
t34

Grab(S2, S25 )
Release(S2, S25 )
t2 S2
t6 S12
t12 t6 t12
t8
t13 t2 t8 t13
S6 t18 S8 t18
t34
t25

c d

Fig. 2. Structure and operators


a Notation used for targets and sensors
b Effect of Sensor Explode operation on a sensor that covers multiple targets
c Effect of Sensor Release operation applied to sensor S2
d Effect of Sensor Grab operation applied to sensor S2

with other sensors.

4.5.2. Sensor Recombine: Two operators are performed in sensor recombine: Sensor Grab,
and Sensor Release. Note that only one of these operations is executed when sensor recombine is
called. The choice of which operation to execute is determined by the grab probability.

Sensor Release: An example case of the Sensor Release concept is shown in Fig. 2c in which
there are 3 sensors indexed by 2, 12 and 25 (each sensor is indexed with the smallest index of those
targets covered by the sensor). The Sensor Release operation eliminates all possible targets within
sensor 2 and gives them to sensors 12 and 25, where 12 and 25 are in sensor 2’s tryMerge list.
After reassigning, each remaining sensor is indexed by the minimum index of all targets covered

7
IET Review Copy Only
IET Wireless Sensor Systems Page 8 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Algorithm 1 Sensor Recompute


Input: sensor (sensor structure) , merge distance, drag
Output: Update of tryM erge lists and acceleration matrix
1: Set acceleration matrix to zeros
2: Compute vector of valid sensors validSensors
3: for i = 1 to length (validSensors) do
4: Set sensor.tryM erge[i] of Si to f alse
5: for j = 1 to length (sensor.mayM erge[i]) do
6: Sj ← sensor.mayM erge[i, j]
7: if Sj is valid then
8: if d (Si , Sj ) < merge distance then
9: Add Sj to Si tryM erge list
10: else
11: compute sensor acceleration using (2)
12: end if
13: end if
14: end for
15: end for

by that sensor. In the example in Fig. 2c, the remaining sensors are 2, 6, and 8. After execution of
this operation, the sensor.valid fields of remaining sensors are set to false and their sensor.pending
fields are set to true.

Sensor Grab: Fig. 2d describes an example case in which there are 3 sensors indexed by 2, 6
and 25. In this case, sensors 6 and 25 are both on the tryMerge list for sensor 2. The Sensor Grab
operation reassigns to sensor 2 all targets covered by 6 and 25 which can possibly be covered by
sensor 2. After reassigning, each remaining sensor is indexed by the minimum index of all targets
covered by that sensor. For the example in Fig. 2d, the sensor.valid fields for sensors 2, 12 and
25 are set to false, and the sensor.pending fields for 2 and 12 take the value true. Observe that the
number of valid sensors has decreased by one.

4.6. Sensor Update


The Sensor Update function first updates the velocities and positions of sensors which have under-
gone physics motion. The update equations are:
(u)
~vi = ~vi + ~ai (4)
(u) 1
~si = ~si + ~vi + ~ai (5)
2
(u) (u)
where ~vi and ~si are the updated velocity and position respectively for sensor Si . These cor-
respond to discretized physics motion equations with time step equal to 1. Note that merged and
recombined sensor locations are not updated.
We must also check whether any targets have gone out of coverage by their assigned sensors.
(u)
Fig. 3a shows a case where the sensor position has been updated to a position ~si such that the
assigned target ~t is no longer covered. In such cases, the updated position is pulled back along

8
IET Review Copy Only
Page 9 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

(u)
the segment joining ~si and ~si to a position where ~t remains within coverage. This segment can
(u)
be parametrized as ~si + β(~si − ~si ), where 0 ≤ β ≤ 1. For ~t to remain covered, we require that
f (β) < rs2 , where
   2
(u)
f (β) = ~t − ~si + β ~si − ~si , 0 ≤ β ≤ 1 (6)

We may re-express f (β) as follows:

f (β) = aβ 2 + bβ + c,

where
2
(u)
a = (~si − ~si )

(u)
b = 2(~t − ~si ) · (~si − ~si )
2
c = ~t − ~si

Since f 00 > 0 it follows that f is convex and

f (β) ≤ f (0) + β(f (1) − f (0)) for 0 ≤ β ≤ 1

If we let
rs2 − f (0)
βr =
f (1) − f (0)
it follows that

f (βr ) ≤ rs2

which implies from (6) that

0 2

~ 2
(t − ~
s i ≤ rs
)

where
0 (u)
~si = ~si + βr (~si − ~si ). (7)
0
Thus ~si can be chosen as the corrected updated target position.

5. Experiments

5.1. Simulation parameters


The main parameters used in the simulation are:
• Maximum number of iterations (numIterMax): 5000
• Wait Period (waitPeriod): 1000

9
IET Review Copy Only
IET Wireless Sensor Systems Page 10 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

250

200

Number of sensors
150

100

50

0
0 500 1000 1500 2000 2500

Iterations

a b

Fig. 3. Process of distance update, followed by an illustration of search space exploration


a Explanation of notation used in distance adjustment equations (6), (7)
b Number of sensors in coverage solution versus iteration number for a single instance of 500
targets in a 750 m × 750 m region, with wait period equal to 1000 iterations. There are 5 bursts
every 1000 iterations

• nPlanet: 10
• K: rs2 /8
• Drag: 0.5
• Recombine probability: 0.25
• Grab probability: 0.5
• Explode probability: see (8)
Note that parameters used are not fully optimized, because the present paper is merely intended
to provide initial proof of concept. The explode probability is computed as follows:

1 waitP eriod 2
) × 500
× 10(−mod(numIter, 5 waitP eriod2
)
(8)
ncur
where ncur is the current number of sensors, and numIter the iteration number. The explosion
probability is designed so that short bursts of high explosion rate occur at regular intervals, which
gives the algorithm the opportunity to explore the solution spaces more thoroughly. The current
code is arranged so that at least 5 bursts occur during the wait period as shown in Fig. 3b.

10
IET Review Copy Only
Page 11 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Fig. 4. Pseudocodes for algorithm subroutines


a Algorithm for Sensor Merge and Recombine
b Algorithm for Sensor Update

5.2. Test data sets


In our first set of simulations a square region is considered as sensor field, and 36 random test
scenarios were generated. The rs was set to 75 m. The number of targets was varied from 250 to

11
IET Review Copy Only
IET Wireless Sensor Systems Page 12 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

8000 in geometric progression with ratio 2. The region size was varied from 750 m to 24000 m in
geometric progression with ratio 2. For each test scenario, 10 random target location configurations
were generated using the given parameters, and the algorithm was run 10 times on each target
location configuration. For each run the final number of sensors, number of iterations and run time
were recorded.
In our second set of simulations, we used the GPS coordinates of 106 targets that represent
administrative centers, public centers, and crossroads of the city of Yaounde, Cameroon [33]. For
this test instance rs varies from 250 m to 1000 m in arithmetic sequence with increment 250.

5.3. Simulation results


For the first set of simulations, typical final sensor configurations are shown in Fig. 5. Numerical
results are summarized in Tables 3 and 4. Fig. 5a and Fig. 5b illustrate the deployment of sensors
in low and high density region respectively. In the high-density case, the target coverage problem
is practically equivalent to the area coverage problem. Fig. 5c and Fig. 5d present another case of
sensor deployment in 3000 m × 3000 m region for 250, and 2000 targets. The number of targets
covered by one sensor increases with the density of targets.
Table 5, and Fig. 6 present the number of sensors computed when varying rs for the second set
of simulations.

5.4. Statistical analysis of random target coverage


In this section, we present a statistical analysis of the results obtained from the first set of simula-
tions (random sensor configurations). In order to evaluate the efficiency of our algorithm (in terms
of producing covers with relatively few sensors), we compare its performance to exact results
which are available for randomly-distributed targets of large and small target densities. Exact
results are not available for intermediate target densities because the optimal-cover problem is
NP-complete, so it is not practical to compute solutions in the general case.
When the target density is very large, the best target cover is identical to total area coverage.
It is known that the circumscribed circles for a regular hexagonal lattice (honeycomb) give the
minimum size cover for total area coverage of large areas in which boundary effects are unimpor-
tant [38]. It is therefore appropriate to compare covers at large density with hexagonal coverage.
Accordingly, for random target configurations at large densities we define the cover size ratio for
a sensor network as the number of sensors in a hexagonal full-area cover of the region divided by
the number of sensors in the particular network.
Fig.7a shows the mean cover size ratio achieved by our algorithm as a function of the parameter
γ, where

γ = (number of targets) * (area of a single sensor) / (area of the region) (9)

The parameter γ can be interpreted as the mean number of targets per single sensor area for a
randomly-placed sensor in the region. From Fig.7a, we observe that the relative improvement
achieved by the algorithm depends on region size, target radius, and number of targets only through
the parameter γ. Thus for instance, if we quadruple the region area and number of targets while
holding sensor radius fixed, then the number of sensors in the cover will also roughly quadruple.
The fact that the relative improvement of our algorithm is independent of problem size shows that
the efficiency does not degrade for larger problems. We may also note from Fig.7a that the cover
size ratio is less than 1 when γ is less than 16: this shows that the algorithm is more efficient at

12
IET Review Copy Only
Page 13 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

1500

700

600

500 1000
Height(m)

Height(m)
400

300

500
200

100

0
0 100 200 300 400 500 600 700 0
0 500 1000 1500
Width(m) Width(m)

a b
3000 3000

2500 2500

2000 2000
Height(m)

Height(m)

1500 1500

1000 1000

500 500

0 0
0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000 2500 3000
Width(m) Width(m)

c d

Fig. 5. Sensor deployment found by the proposed method


a 250 targets (27 sensors), 750 m × 750 m square region
b 4000 targets (149 sensors), 1500 m × 1500 m square region
c 250 targets (131 sensors), 3000 m × 3000 m square region
d 2000 targets (353 sensors), 3000 m × 3000 m square region

covering randomly-distributed targets than hexagonal area coverage as long as the parameter γ is
16 or less. Thus for even for rather high target densities, we can still obtain improvements over
full-area coverage by using our algorithm.
For random target configurations with very low target densities, it is possible to compute the

13
IET Review Copy Only
IET Wireless Sensor Systems Page 14 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Table 3 Results of the first set of simulations


region No. of Min. Max. Average Standard deviation
size(m × m) targets n n n Iteration T ime n Iteration. T ime
750 250 27 34 30.49 802.48 29.48 1.24 527.96 19.61
1500 250 67 77 71.43 821.45 49.21 2.50 507.45 23.14
3000 250 131 144 139.52 440.97 24.95 3.45 399.82 18.45
6000 250 195 206 200.70 52.12 18.02 3.54 144.17 11.52
12000 250 227 239 235.20 10.40 15.64 3.36 2.43 9.93
24000 250 244 249 246.70 9.71 9.92 1.68 3.06 5.22
750 500 33 39 35.68 1036.31 60.44 1.38 598.81 33.65
1500 500 87 99 92.16 1355.18 88.68 2.08 649.41 44.48
3000 500 194 212 203.09 1123.09 84.51 3.68 591.27 49.13
6000 500 332 361 348.13 269.80 38.75 7.57 284.66 23.45
12000 500 430 447 440.30 25.42 41.85 4.80 50.77 21.69
24000 500 475 487 483.50 10.35 28.32 3.85 1.75 19.39
750 1000 38 45 40.99 1174.64 128.35 1.49 640.05 59.09
1500 1000 108 120 114.64 1947.12 142.39 2.38 963.89 97.19
3000 1000 267 288 276.26 2123.35 103.16 4.57 1019.98 43.13
6000 1000 524 561 546.41 876.15 65.96 10.47 599.80 31.53
12000 1000 776 827 807.36 109.76 35.60 13.13 146.99 8.48
24000 1000 932 954 942.40 10.40 34.16 6.91 1.97 3.92
750 2000 42 48 44.78 1296.60 293.82 1.25 567.55 72.20
1500 2000 132 145 137.29 2212.02 307.37 2.57 949.51 89.19
3000 2000 353 374 363.20 3156.92 273.91 4.30 1049.89 60.31
6000 2000 774 812 790.47 2438.64 226.80 8.40 932.47 59.81
12000 2000 1364 1393 1381.94 509.41 99.88 9.06 379.25 30.22
24000 2000 1773 1802 1788.31 31.44 66.55 8.80 49.12 10.38
750 4000 45 52 48.31 1423.10 768.87 1.46 661.46 249.64
1500 4000 149 163 156.57 2498.29 1588.30 2.96 899.76 8218.70
3000 4000 440 468 451.03 4013.02 696.58 5.15 928.53 131.78
6000 4000 1074 1110 1092.33 4246.79 642.22 8.46 742.85 105.81
12000 4000 2129 2208 2172.28 2130.64 451.24 17.88 881.54 159.15
24000 4000 3205 3280 3240.74 310.81 160.54 22.66 292.58 36.85
750 8000 - - - - - - - -
1500 8000 165 180 171.88 2728.51 2430.44 2.63 917.28 1232.12
3000 8000 526 551 536.28 4240.88 3117.74 5.40 780.00 1580.15
6000 8000 1422 1459 1438.93 4762.58 1544.59 7.76 283.56 42.55
12000 8000 3107 3176 3154.83 4507.33 1507.72 15.80 624.03 109.53
24000 8000 5472 5588 5525.91 1389.34 872.44 28.10 631.14 328.28

smallest cover within a reasonable computation time by first identifying all pairs of targets that
are within 2rs of each other, and then finding (by exhaustion) all clusters of 3, 4, 5, 6, . . . sensors.
This method was used to estimate the mean cover size for random target configurations for several
different region sizes and target densities.
In Table 6, these results are compared with those obtained from the proposed algorithm. We
find that there is extremely close agreement between the theoretical best mean sensor per target
values and the sensor per target values obtained by our algorithm. This shows that our algorithm is
very close to optimal for random target distributions with low target densities.
The results we have described so far show that for both small and large densities, the number
of sensors required to cover a randomly-distributed set of targets is proportional to the number
of targets, and the constant of proportionality depends only on the parameter γ. The constant of

14
IET Review Copy Only
Page 15 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

Table 4 Variation between different random configurations


region No. of Variation between different random target Variation between different runs of the
size(m × m) targets configurations for the same scenario algorithm for the same random configuration
std.(avg.(n)) std.(avg.(It.1 )) std.(avg.(T.2 )) avg.(std.(n)) avg.(std.(It1 )) avg.(std.(T.2 ))
750 250 0.86 227.66 17.13 0.95 486.58 9.95
1500 250 2.31 139.74 19.03 1.23 499.73 13.87
3000 250 3.56 218.08 17.60 0.53 329.12 6.38
6000 250 3.71 93.00 10.88 0.00 49.73 3.32
12000 250 3.52 0.70 9.92 0.00 1.26 1.97
24000 250 1.77 2.12 5.08 0.00 1.33 0.92
750 500 0.87 168.14 28.27 1.13 571.81 19.47
1500 500 0.98 181.25 33.58 1.89 644.33 30.27
3000 500 3.55 196.84 32.55 1.47 580.42 37.28
6000 500 7.93 111.16 21.78 0.17 257.21 9.96
12000 500 5.03 30.20 22.60 0.00 22.94 1.72
24000 500 4.03 0.46 20.14 0.00 1.11 1.35
750 1000 0.50 204.17 19.10 1.47 624.18 52.64
1500 1000 0.90 224.41 77.17 2.30 968.92 54.65
3000 1000 3.56 340.52 19.35 3.15 994.18 35.50
6000 1000 10.94 257.92 19.45 0.90 557.16 22.71
12000 1000 13.77 72.83 5.29 0.09 112.00 5.01
24000 1000 7.24 0.70 1.44 0.00 1.05 3.26
750 2000 0.39 228.13 26.71 1.22 539.82 68.98
1500 2000 1.13 349.33 29.13 2.36 904.71 87.52
3000 2000 2.65 401.23 27.06 3.50 996.18 54.92
6000 2000 8.21 386.68 24.65 3.08 870.81 55.34
12000 2000 9.49 130.95 12.32 0.46 350.89 26.49
24000 2000 9.23 33.86 6.02 0.03 21.92 4.71
750 4000 0.42 151.53 81.33 1.45 658.66 237.00
1500 4000 0.72 222.09 2620.68 2.90 886.98 2786.36
3000 4000 2.55 241.67 54.41 4.60 921.31 118.97
6000 4000 6.89 228.13 43.37 5.50 714.97 83.20
12000 4000 18.59 417.40 85.47 2.38 807.23 131.42
24000 4000 23.76 171.74 21.94 0.32 224.63 28.47
750 8000 - - - - - -
1500 8000 0.69 283.61 644.91 2.58 900.25 851.40
3000 8000 3.28 232.83 1192.36 4.47 761.66 1029.95
6000 8000 5.61 108.94 31.44 5.84 253.73 29.43
12000 8000 15.70 194.89 35.78 5.19 582.07 89.99
24000 8000 29.43 293.28 209.42 1.48 572.83 264.53
avg.: Average, std.: Standard deviation
1 Iteration
2 Time

Table 5 Results of the second set of simulations


Sensing Number of sensors in the solution Iteration number Run time(s)
radius (m) min. max. avg. std. avg. std. avg. std.
1000 23 26 24.60 0.82 558.82 431.70 10.71 2.88
700 33 35 34.07 0.62 463.52 429.72 8.59 4.06
500 50 51 50.01 0.10 130.83 225.36 5.83 1.64
250 73 73 73 0.00 32.11 24.32 6.16 2.85
avg.: Average, std.: Standard deviation

15
IET Review Copy Only
IET Wireless Sensor Systems Page 16 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

11 11

10 10

9 9

8 8

7 7
Height(km)

Height(km)
6 6

5 5

4 4

3 3

2 2

1 1

0 0
0 5 10 15 0 5 10 15
Width(km) Width(km)

a b
11 11

10 10

9 9

8 8

7 7
Height(km)

Height(km)

6 6

5 5

4 4

3 3

2 2

1 1

0 0
0 5 10 15 0 5 10 15
Width(km) Width(km)

c d

Fig. 6. Sensor deployment of administrative centers, public centers, and crossroads of the city of
Yaounde, Cameroon with different sensor radii
a rs = 1000 m (23 sensor nodes)
b rs = 750 m (33 sensor node)
c rs = 500 m (50 sensor nodes)
d rs = 250 m (73 sensor nodes)

proportionality is graphed as a function of γ (in log scale) in Fig.7d. The curve resembles a logistics
curve, so we attempted a fit with the following functional form:

number of targets
number of sensors = (10)
(1 + (a × γ)b )
The least-squares best fit parameters were found to be a = 1.5 and b = 0.855. As shown in
the figure, the logistics curve with these parameters fits the data extremely well. This formula is
of considerable mathematical interest, because it estimates the expected number of circles of fixed
size that are required to cover a set of random points in the plane with a given density.
Besides efficiency, we are also interested in the computational complexity of the algorithm.

16
IET Review Copy Only
Page 17 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

a b

c d

Fig. 7. Performance evaluation of the proposed approach


a Relative efficiency versus hexagonal closest packing
b Iterations per target as a function of number of targets
c Execution time per target as a function of number of targets
d Sensors per target versus γ
Table 6 Coverage efficiency of algorithm at low densities compared to
theoretical optimum
region number γ Obtained True optimum mean
size of targets sensors/target sensors/target
24000 250 0.01 0.99 ± 4.48e − 03 0.98 ± 7.55e − 04
24000 500 0.02 0.97 ± 5.10e − 03 0.97 ± 7.28e − 04
12000 250 0.03 0.94 ± 1.27e − 02 0.94 ± 6.93e − 04
24000 1000 0.03 0.94 ± 4.58e − 03 0.94 ± 6.93e − 04
12000 500 0.06 0.88 ± 6.36e − 03 0.89 ± 9.77e − 04
24000 2000 0.06 0.89 ± 2.92e − 03 0.89 ± 9.77e − 04
6000 250 0.12 0.80 ± 9.39e − 03 0.81 ± 6.32e − 03
12000 1000 0.12 0.81 ± 8.71e − 03 0.81 ± 6.32e − 03
24000 4000 0.12 0.81 ± 3.76e − 03 0.81 ± 6.32e − 03
6000 500 0.25 0.70 ± 1.00e − 02 0.72 ± 7.27e − 02
12000 2000 0.25 0.69 ± 3.00e − 03 0.72 ± 7.27e − 02
24000 8000 0.25 0.69 ± 2.33e − 03 0.72 ± 7.27e − 02

Fig.7b shows the mean number of iterations per target as a function of problem size (number of
targets), for different values of the parameter γ. The graphs are downward sloping, which implies
that the number of iterations per target decreases as the number of targets increase, for fixed γ. This
shows that the number of iterations scales sub-linearly with the number of targets: so for example,
if the region size and the number of targets are doubled, the number of iterations will be less than
double. Thus our algorithm scales well. Note also that the number of iterations per target is largest
at intermediate densities, and is smaller at very low and very high densities.
Fig.7c shows the mean execution time as a function of problem size(represented by the number
of targets), for different values of the parameter γ. The irregularity of the graphics illustrates the

17
IET Review Copy Only
IET Wireless Sensor Systems Page 18 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

high variability in execution time. (If less variability in execution time is required, the user may
impose a maximum execution time with very little degradation in performance.) The curves are
more or less flat. This means that the execution time per target depends only on the target density
and not on the region size. For instance, if the region size and the number of targets are doubled
while keeping the sensor radius fixed, then the execution time roughly doubles. The figure also
indicates that lower target densities correspond to lower execution: this result is reasonable since
it should be easier to find the solution at low density. Note finally that in contrast to iteration num-
ber, execution time is highest for higher values of γ, indicating that large densities require more
computation time when the number of targets is fixed.
Table 3 summarizes the number of iterations and the running time of the proposed method for
the first set of simulations. The maximum running time returned by our method is 3117.74 seconds
(about an hour) for 8000 targets in a 3000 m × 3000 m region. This shows the capability of our
method to solve very large-scale problems in a comparatively short time. One factor that con-
tributes significantly to the reduction in computational effort is the construction of the mayMerge
list field. At the beginning of the algorithm, the mayMerge list is generated for each target i. This
list contains the indices of sensors > i that are sufficiently close to i to have some possibility of
merging. When the algorithm searches for possible merges for each sensor i, the search is re-
stricted to only those virtual sensors that are included in the mayMerge list of i and not the set of
all targets (virtual sensors). As a result, the search computation time is greatly reduced.
Table 4 shows the variation in results between configurations, and between different runs on the
same configuration. Columns 3 and 6 show the variation in mean sensor number between config-
uration and the variation between different runs on the same configuration, respectively. In both
cases, the variations are relatively small compared to the number of targets. For a fixed number of
targets, variation in sensor number between configurations tends to be larger for smaller densities,
while variations between runs for the same configuration tend to be larger for larger densities. On
the other hand, Table 4 shows that the variation in number of iterations is larger for larger densi-
ties both between configurations and among different runs on the same configuration. In general,
the variation in iteration number among runs on the same configuration dominates the variation
between random configurations of the same size and density. As far as execution times, variations
between configurations at the same size and density are of comparable magnitude to variations
among different runs on the same configuration.

5.5. Analysis of Yaounde test instance


In this section we analyze the results obtained using real-world data from the city of Yaounde,
Cameroon. The results, shown in Table 5, indicate that for larger values of rs , the number of
sensors is inversely proportional to rs ; for smaller values of rs , the number of sensors no longer
follows this relationship. This result can be interpreted in terms of the mathematical notion of
“fractal dimension” [39]. A set of points in the plane has fractal dimension D if the number of
circles of radius R required to cover the set is roughly proportional to R−D . The results in Table
5 indicate that when the sensor radius is large, the effective fractal dimension of the set of targets
is 1. However, when the sensor radius becomes smaller, the effective fractal dimension decreases.
When the sensor diameter is smaller than the minimum distance between targets, then the effective
fractal dimension is 0 because at that point the number of sensors must be equal to the number of
targets and does not change if the sensor radius is further decreased.
We may use Equation (10) to provide a rough preliminary estimate of the number of sensors
required for a cover in the Yaounde test instance,, based only on region size and number of targets.

18
IET Review Copy Only
Page 19 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

The smallest rectangle that contains the targets is 15 km by 10 km: and this information, together
with the number of targets and the sensor radius, is sufficient to determine the value of γ. Table 7
gives preliminary estimates of the cover size using Equation (10) and these computed values of γ.
The theoretical formula overestimates the actual number of sensors from 14 to 22 percent. An
overestimate is to be expected, because the targets are more clustered than would be obtained by a
uniform distribution, so fewer sensors are required to cover them. The fact that the overestimation
is less than 22 percent means that Equation (10) can provide a conservative but fairly close pre-
liminary estimate for the number of sensors required, even before the actual target distribution is
known. In practice, this can be useful for planning purposes.
Table 7 Predicted versus actual number of sensors, Yaounde
Sensor radius(km) 1 0.75 0.5 0.25
γ parameter 2.22 1.25 0.56 0.14
Predicted number of sensors 27.92 39.11 57.14 84.04
Actual number of sensors 23 33 50 73
Difference (%) 21 19 14 15

6. Conclusion

In this paper, we have presented a physics-based heuristic for the efficient placement of sensor
nodes for target coverage problem in wireless sensor networks. We analyzed the characteristics of
the problem and tried the best way to move and merge sensors in order to provide into a reasonable
time, a minimum number of sensors that covers all targets. To achieve this, we have introduced the
concept of virtual sensor and have designed three operators, namely the explosion, the grab and
the release. The explosion allowed exploring the solution space while grab and release contributed
to its exploitation. The performance of the proposed method has been proved through intensive
simulations whose statistical analyzes have shown the capacity and the efficiency of the proposed
approach to solve the problem raised. For future work, we will focus on the integration into the
present method a model of connectivity between sensors deployed.

7. References

[1] Aguiar, E.F.K., Roig, H.L., Mancini, L.H., et al.: ‘Low-cost sensors calibration for monitoring
air quality in the Federal District-Brazil’, Journal of Environmental Protection, 2015, 6, (2), pp
173–189.
[2] Sivaraman, V., Carrapetta, J., Hu, K., et al.: ‘Hazewatch: A participatory sensor system for
monitoring air pollution in Sydney’. Proc. 38th Annual IEEE Conference on Local Computer
Networks-Workshops, Sydney, Australia, 2013, pp. 56–64.
[3] Carrapetta, J.: ‘Haze watch: Design of a wireless sensor board for measuring air pollution’
(University of New South Wales, Sydney, Australia, 2010), pp. 1-67.
[4] Liu, X.: ‘A deployment strategy for multiple types of requirements in wireless sensor net-
works’, IEEE Trans. Cybernet., 2015, 45, (10), pp. 2364–2376.
[5] Wang, B.: ‘Coverage problems in sensor networks: Survey’, ACM Comput. Surveys, 2011,
43, (4), pp. 32:1–32:53.

19
IET Review Copy Only
IET Wireless Sensor Systems Page 20 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

[6] Deif, D.S., Gadallah, Y.: ‘Classication of wireless sensor networks deployment techniques’,
IEEE Commun. Surveys Tuts., 2014, 16, (2), pp. 834–855.
[7] Mini, S., Udgata, S.K., Sabat, S.L.: ‘Sensor deployment and scheduling for target coverage
problem in wireless sensor networks’, IEEE Sens. J., 2014, 14, (3), pp. 636–644.
[8] Erdelj, M., Mitton, N., Razafindralambo, T.: ‘Robust wireless sensor network deployment’,
Discrete Mathematics & Theoretical Computer Science, 2016, 17, (3), pp 105–130.
[9] Yoon, Y., Kim, Y.-H.: ‘An efficient genetic algorithm for maximum coverage deployment in
wireless sensor networks’, IEEE Trans. Cybern., 2013, 43, (5), pp. 1473–1483.
[10] Bartolini, N., Calamoneri, T., Fusco, E.G., et al.: ‘Push & pull: autonomous deployment of
mobile sensors for a complete coverage’, Wirel. Netw., 2010, 16, (3), pp. 607–625.
[11] Chang, C.-Y., Chang, C.-T., Chen, Y.-C., et al.: ‘Obstacle-resistant deployment algorithms
for wireless sensor networks’, IEEE Trans. Veh. Technol., 2009, 58, (6), pp. 2925–2941.
[12] Chaudhary, M., Pujari, A.K.: ‘Q-coverage problem in wireless sensor networks’. Proc. Int.
Conf. Distrib. Comput. Netw., India, Jan. 2009, pp. 325–330.
[13] Gu, Y., Liu, H., Zhao, B.: ‘Target coverage with QoS requirements in wireless sensor net-
works’. Proc. Intell. Pervas. Comput., Oct. 2007, pp. 35–38.
[14] Cardei, M., Thai, M.T., Li, Y., et al.: ‘Energy-efficient target coverage in wireless sensor
networks’. Proc. 24th Annu. Joint Conf. IEEE INFOCOM., Mar. 2005, pp. 1976–1984.
[15] Wu, Y., Cardei, M.: ‘Distributed algorithms for barrier coverage via sen-
sor rotation in wireless sensor networks, J. Comb. Opt., 2016, pp. 1–22,
http://link.springer.com/article/10.1007/s10878-016-0055-3, accessed 20 July 2016.
[16] Chen, J., Wang, B., Liu, W.: ‘Constructing perimeter barrier coverage with bistatic
radar sensors’, Journal of Network and Computer Applications, 2015, 57, pp 129–141,
http://dx.doi.org/10.1016/j.jnca.2015.07.015, accessed 13 June 2016.
[17] Deng, X., Wang, B., Wang, C., et al.: ‘Barrier coverage in wireless sensor networks with
adjustable sensing ranges’, Int.J.Ad Hoc and Ubiquitous Computing, 2014, 15, (1/2/3), pp.
121–132.
[18] Ke, W., Liu, B., Tsai, M.: ‘Constructing a wireless sensor network to fully cover critical grids
by deploying minimum sensors on grid points is np-complete’, IEEE Trans. Comput., 2007,
56, (5), pp. 710–715.
[19] Xu, H., Zhu, J., Wang, B.: ‘On the deployment of a connected sensor network for condent
information coverage’, Sensors(Basel), 2015, 15, (5), pp 11277–11294.
[20] Altnel, K., Aras, N., Gney, E., et al.: ‘Binary integer programming formulation and heuristics
for differentiated coverage in heterogeneous sensor networks’, Computer Networks, 2008, 52,
(12), pp 2419–2431.
[21] Xu, X., Sahni, S.: ‘Approximation algorithms for sensor deployment’, IEEE Trans. Comput.,
2007, 56, (12), pp. 1681–1695.

20
IET Review Copy Only
Page 21 of 22 IET Wireless Sensor Systems
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

[22] Chakrabarty, K., Iyengar, S., Qi, H., et al.: ‘Grid coverage for surveillance and target location
in distributed sensor networks’, IEEE Trans. Comput., 2002, 51, (12), pp. 1448–1453.
[23] Wang, J., Zhong, N.: ‘Efficient point coverage in wireless sensor networks’, J. Comb. Opt.,
2006, 11, (3), pp. 291–304.
[24] Dhillon, S.S., Chakrabarty, K.: ‘Sensor placement for effective coverage and surveillance in
distributed sensor networks’. Proc. IEEE Conf. Wireless Commun. Netw., New Orleans, LA,
USA, Mar. 2003, pp. 1609–1614.
[25] Lin, F., Chiu, P.: ‘A near-optimal sensor placement algorithm to achieve complete coverage-
discrimination in sensor networks’, IEEE Commun. Lett., 2005, 9, (1), pp. 43–45.
[26] Banimelhem, O., Mowafi, M., Aljoby, W.: ‘Genetic algorithm based node deployment in
hybrid wireless sensor networks’, Communications and Network, 2013, 5, (4), pp 273–279.
[27] Kalayci, T.E., Ugur, A.: ‘Genetic algorithm-based sensor deployment with area priority’,
Cybernetics and Systems, 2011, 42, (8), pp 605–620.
[28] Seo, J.-H., Kim, Y.-H., Ryou, H.-B., et al.: ‘Optimal sensor deployment for wireless surveil-
lance sensor networks by a hybrid steady-state genetic algorithm’, IEICE Trans. Commun.,
2008, E91-B, (11), pp. 3534–3543.
[29] Wu, Q., Rao, N.S.V., Du, X., et al.: ‘On efficient deployment of sensors on planar grid’,
Comput. Commun., 2007, 30, (14-15), pp. 2721–2734.
[30] Zhao, C., Yu, Z., Chen, P.: ‘Optimal deployment of nodes based on genetic algorithm in
heterogeneous sensor networks’. Proc. IEEE International Conference on Wireless Communi-
cations, Networking and Mobile Computing (WiCom), Sep. 2007, pp. 2743–2746.
[31] Jia, J., Chen, J., Chang, G., et al.: ‘Energy efficient coverage control in wireless sensor
networks based on multi-objective genetic algorithm, Computers and Mathematics with Ap-
plications’, 2009, 57, (11-12), pp 1756–1766.
[32] Welzl, E.: ‘Smallest enclosing disks (balls and ellipsoids)’, in Maurer, H. (Ed.): ‘New Results
and New Trends in Computer Sciences’, 1991, 555, pp. 359–370.
[33] Tonye, E., Tsague, W., Bateki, R.: ‘Réalisation dd’un capteur de pollution et rapatriement des
donnes vers un SIG web’ (University of Yaounde I, National Advanced School of Engineering,
Yaounde, Cameroon, 2015), pp. 1-75.
[34] Contreras, W., Ziavras, S.: ‘Wireless sensor network-based pattern matching technique for the
circumvention of environmental and stimuli-related variability in structural health monitoring’,
IET Wirel. Sens. Syst., 2016, 6, (1), pp 26–33.
[35] Zonouz, E.A., Xing, L., Vokkarane, M.V., et al.: ‘Hybrid wireless sensor networks: a relia-
bility, cost and energy-aware approach’, IET Wirel. Sens. Syst., 2016, 6, (2), pp 42–48.
[36] Ali, A., Ikpehai, A., Adebisi, B., et al.: ‘Location prediction optimisation in WSNs using
Kriging interpolation’, IET Wirel. Sens. Syst., 2016, 6, (3), pp 74–81.
[37] Zou, Yi., Chakrabarty K.: ‘Sensor Deployment and Target Localization Based on Virtual
Forces’. Proc. IEEE Infocom Conference, Mar. 2007, pp. 1293–1303.

21
IET Review Copy Only
IET Wireless Sensor Systems Page 22 of 22
This article has been accepted for publication in a future issue of this journal, but has not been fully edited.
Content may change prior to final publication in an issue of the journal. To cite the paper please use the doi provided on the Digital Library page.

[38] Kershner, R.: ‘The Number of Circles Covering a Set’, American Journal of Mathematics,
Jul. 1939, 61, (3), pp 665–671.
[39] Weisstein, E. W. ‘Fractal Dimension’, http://mathworld.wolfram.com/FractalDimension.html,
accessed 05 December 2016

22
IET Review Copy Only

You might also like