You are on page 1of 15

Accepted Manuscript

Novel Efficient Deployment Schemes for Sensor Coverage in Mobile


Wireless Sensor Networks

Wei Fang, Xinhong Song, Xiaojun Wu, Jun Sun, Mengqi Hu

PII: S1566-2535(16)30254-8
DOI: 10.1016/j.inffus.2017.08.001
Reference: INFFUS 888

To appear in: Information Fusion

Received date: 27 December 2016


Revised date: 12 May 2017
Accepted date: 3 August 2017

Please cite this article as: Wei Fang, Xinhong Song, Xiaojun Wu, Jun Sun, Mengqi Hu, Novel Efficient
Deployment Schemes for Sensor Coverage in Mobile Wireless Sensor Networks, Information Fusion
(2017), doi: 10.1016/j.inffus.2017.08.001

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT

Highlights
• The Voronoi blindzone polygon is studied for finding
coverage holes efficiently.
• Two schemes are proposed based on Voronoi blindzone
polygon and local operators.
• Latest metrics are used to evaluate the performance of
proposed deployment schemes.
• The proposed two deployment schemes show effective-
ness by the simulation results.

T
IP
CR
US
AN
M
ED
PT
CE
AC
ACCEPTED MANUSCRIPT

2 JOURNAL OF LATEX CLASS FILES, VOL. 0, NO. 0, MAY 2017

Novel Efficient Deployment Schemes for Sensor


Coverage in Mobile Wireless Sensor Networks
Wei Fang, Member, IEEE, Xinhong Song, Xiaojun Wu, Jun Sun, Mengqi Hu

Abstract—In the study of improving the efficiency of mobile but random deployment of mobile sensors at initial phase
wireless sensor networks(MWSNs), an important issue is to can not guarantee the required coverage. Therefore, how to

T
maximize the sensor coverage in a given sensing field by proper design efficient deployment algorithms to move the sensors
deployment of sensors. In this paper, two novel sensor deployment
for maximizing the sensor coverage is one of the fundamental

IP
schemes are proposed to address the coverage issue in MWSNs.
The first scheme is blind-zone centroid-based scheme (BCBS) and design issues in mobile wireless sensor networks (MWSNs)
the second one is disturbed centroid-based scheme (DCBS). The [14] and is the problem to be addressed in this paper. Extensive
main ideas both in BCBS and DCBS are the proposed schemes research efforts have been conducted in the field of MWSNs

CR
to find the target locations for sensors to heal the coverage to develop deployment algorithms for increasing the network
holes efficiently. The definition of Voronoi blind-zone polygon
is given for introducing the proposed two schemes. In order to coverage [15]. Among the literature, virtual force based and
clearly illustrating the effect of Voronoi blind-zone polygon, three computational geometry based deployment approaches have
possible cases of it are studied in detail. In BCBS, Voronoi blind- been studied as the main approaches [10].
zone polygon of each sensor is firstly determined by considering
the positions with its neighbors. And then the centroid of the
Voronoi blind-zone polygon is regarded as the target location
of each sensor if the coverage can increase. In DCBS, sensors
find coverage holes according to the centroid-based scheme at
US In the virtual force based deployment algorithms, the sen-
sors are regarded as virtual particles subject to repulsive
or attractive virtual forces among them. The sensors move
according to the resultant forces on them when the distance
AN
first in each round. They then move to the target locations between sensors is less than a threshold. Originated from
under the local perturbation and local reconstruction operators. the concept of potential field used in mobile robotic route
These two operators are designed by studying two forms of local [16], Howard [17] et al. studied a potential field-based de-
convergence. Under the guideline of testbed-based multi-metric
quality measurement of sensor deployment schemes, simulation ployment approach and proved the convergence property. In
M

results are sufficiently presented to demonstrate the effectiveness [18], a deployment optimization strategy based on Target
of the proposed two deployment schemes. Involved Virtual Force Algorithm (TIVFA) is proposed, where
Index Terms—Coverage, centroid, mobile wireless sensor net- the configuration of WSN can be adjusted dynamically for
works, Voronoi diagram, sensor deployment. improving coverage and detection probability. Kribi et al.
ED

improved Virtual Force Algorithm (VFA) in [19] from four


aspects, which are coverage, connectivity, fault-tolerance and
I. I NTRODUCTION
energy [20]. A distributed deployment strategy based on VFA
Wireless sensor networks (WSNs) are composed of a set of is presented in [21]. In [22], VFA is worked with Delaunay
PT

distributed tiny low-power sensors that can sense or monitor triangulation by considering the adjacent relationship among
physical or environmental conditions. Over the past decades, sensors. In [23], a centroid-directed virtual force deployment
WSNs have been used in a wide range of applications, strategy is designed which incorporates the coverage hole
including target tracking and detection [1] [2] [3], disaster fixing ability with the spreading ability. The specific approach
CE

intervention [4], environmental monitoring [5] [6] [7], health to obtain the weights of virtual force on the sensors by the
monitoring [8] [9], and so on. The problems in WSNs, such centroid of Voronoi polygon is not given in [23] and relies on
as deployment [10], data collection [11], network routing numerous attempts. A distributed deployment algorithm based
[12], energy consumption [13], have arisen and been studied. on virtual forces for three dimensional sensing area is provided
AC

However, deployment of sensors is a fundamental aspect in in [24].


WSNs as it greatly influences the performance of network, Voronoi diagram and Delaunay triangulation are two fa-
which includes the coverage, connectivity, uniformity, etc [10]. mous computational geometry structures with a wide range
In many situations, WSNs work in unknown, hostile, remote of applications [25] and have been widely used for sensor
harsh field, toxic region or disaster area which makes it diffi- deployment problem [23]. Wang et al. successfully proposed
cult or impossible for human to deploy the sensors. A possible three deployment algorithms based on Voronoi diagram, which
way is to use mobile sensors which can move autonomously, are VEC, VOR, and Minimax, to discover coverage holes
[26]. However, the performance of VOR and Minimax are
W. Fang, X. Wu, J.Sun is with Department of Computer Science and
Technology, School of Internet of Things Engineering, Jiangnan University, greatly affected by initial position of sensors; VEC may suffer
Wuxi, China, UK. Email: fangwei@jiangnan.edu.cn. performance decreasing with the increasing of the number of
X. Song is with School of Internet of Things Engineering, Jiangnan sensors [27]. In [28], Centroid and Dual-Centroid schemes are
University
M. Hu is with the University of Illinois at Chicago, USA proposed by using Voronoi diagram and centroid to maximize
Manuscript received - the sensing coverage. But the location of neighbours is not
ACCEPTED MANUSCRIPT

SHELL et al.: BARE DEMO OF IEEETRAN.CLS FOR JOURNALS 3

consider in [28], the centroid of Voronoi polygon cannot ith sensor is denoted as si = (xi , yi ). Boolean disk coverage
be guaranteed to be a good location to improve coverage. sensor model, which is the most widely used sensor coverage
An energy-efficient deployment algorithm based on Voronoi model [36] [10] in the literature for its simplicity, is used in
diagram is presented in [29]. Boukerche and Fei presented a this paper.
deployment algorithm to detect coverage holes locally based
Definition 1. Any sensor who can communicate with the
on the localized Voronoi diagram construction [30]. Based on
sensor si through single hop or multiple hops is called the
the distances of each sensor and the points inside its coordinate
communicable node of the sensor si .
Voronoi polygon from the edges or vertices of the polygon,
Mahboubi et al. proposed a set of distributed deployment Definition 2. A pair of communicable nodes whose Voronoi
algorithms for efficient field coverage [27] [31]. In [32], a polygons share an edge are the neighbors.
secure Voronoi-based deployment algorithm is studied which The following assumptions are taken into account for the
can guarantee termination and is tolerant of an attack at low coverage problem in this paper [27], [28].

T
expense of performance. In order to find the shortest node Assumption 1: There is no obstacle in the sensor field. The
movement path to heal the coverage holes, a Delaunay-based

IP
boundary of the sensor field is regarded as a wall-like obstacle.
coordinate-free mechanism for full coverage is proposed in Assumption 2: All sensors have the same sensing range Rs
[33], where local shortest paths for healing holes can be and communication range Rc .
found without knowing exact location of sensors. Sung and

CR
Assumption 3: Each sensor knows its own location informa-
Yang proposed a distributed greedy deployment approach for tion and the other communicable sensors’ location information
improving the coverage of directional sensor networks [34]. through connected communication routes.
For better trade-off between the coverage and the energy The coverage problem studied in this paper can be stated
consumption, an multi-objective immune algorithm using the as: Given the 2D sensor filed and the set of sensors with the
Voronoi diagram properties was proposed in MWSNs [35].
In this paper, to address the coverage problem in MWSNs,
we design two novel efficient centroid-based deployment
schemes for sensor coverage in MWSNs by the iterative pro-
US sensing range Rs and communication range Rc , it is desired to
deploy the sensors by proper deployment algorithms in order
to maximize the sensor coverage or minimize the coverage
holes.
AN
cedures. The first scheme is blind-zone centroid-based scheme
(BCBS). In BCBS, the candidate destination location of each B. Commonly Used Performance Metrics
sensor is the centroid of Voronoi blind-zone polygon instead In this paper, we firstly select five commonly used perfor-
of the centroid of Voronoi polygon. The Voronoi blind-zone mance metrics [23], [27], [29], which are coverage rate, mov-
M

polygon is helped to find the coverage hole by considering ing distance, deployment time, uniformity, and connectivity, to
the coverage of neighbors to the related Voronoi polygon. The compare the proposed algorithms with the other deployment
second scheme is disturbed centroid-based scheme (DCBS) algorithms.
with the proposed local perturbation and local reconstruction 1) Coverage rate: Coverage rate (CR) is one of the most
ED

operators and is very effective to increase the coverage when important performance metrics for evaluating the deployment
the sensors are stagnated. A distinguishing characteristic of algorithms in WSNs. In the boolean disk coverage sensor
BCBS and DCBS that different from the deployment schemes model, CR can be defined as the ratio of the union of areas
[26] [27] [28] is that there are no manual parameters need to covered by each sensor to the area of the 2D sensor field [23],
PT

be set. The performance of BCBS and DCBS are evaluated that is,
using two sets of simulations and compared with four other ∪N Cri
CR(%) = i=1 , (1)
deployment algorithms in terms of coverage, moving distance, A
deployment time, uniformity, and connectivity based on the where Cri is the area covered by the sensor si and N is
CE

simulation results. the number of sensors, A is the area of the 2D sensor field.
The rest of this paper is organised as follows. Section II From (1), it is clear that coverage rate is closely related to the
presents the problem formulation and performance metrics. sensing range of each sensor.
Section III introduces the preliminaries concerning the Voronoi 2) Moving distance: The moving distance (MD) of a sensor
AC

diagram, Voronoi polygon and its centroid, the basic idea of during the deployment represents the cumulative distance
centroid-based deployment scheme, and ideal distribution of traveled until it reaches its final location after the initial
sensors. In Section IV, two proposed deployment schemes deployment. In this paper, we consider the impact of network
are described. Simulation results and concluding remarks are connectivity on the moving distance. After the initial random
drawn in Section V and Section VI, respectively. deployment, each sensor records its traveled cumulative dis-
tance, denoted as LTi , until the network is connected. And then
II. P ROBLEM FORMULATION AND PERFORMANCE METRICS the sensors do not move until the deployment algorithms work
out their final locations. After that, the sensors move to their
A. Problem Formulation final locations directly. The linear distance for the ith sensor
We consider the coverage problem in the context of a two- is denoted as LVi . Therefore, for the set of N sensors, the
dimensional (2D) flat sensor field with length L and width average moving distance is calculated as,
W. A set of mobile wireless sensors S = {s1 , s2 , · · · , sN } PN
(LT + LVi )
are distributed in the 2D sensor field. The location of the M D = i=1 i . (2)
N
ACCEPTED MANUSCRIPT

4 JOURNAL OF LATEX CLASS FILES, VOL. 0, NO. 0, MAY 2017

3) Deployment time: The deployment time used in this are not greater than their distance to the other nodes cj , where
paper is defined as the elapsed time until all the sensors move j is the index different from i, that is [27],
to their final locations from the initial deployment. The value
ci = {x ∈ <2 |d(x, si ) ≤ d(x, sj ), j ∈ {1, · · · , N }, j 6= i}
of deployment time can be used to compare the computational
(4)
complexity of the deployment schemes.
where d(x, si ) is the Euclidean distance between the points x
4) Uniformity: Uniformity of deployment refers to uni-
and si . Another characteristic of Voronoi diagram is that the
formly distributed sensor nodes in the sensor field, which can
edges of Voronoi polygon ci are the perpendicular bisectors
decrease interference between sensors and consequently con-
of the line segment between the node si and its Voronoi
sume communication energy more efficiently. The network’s
neighbor nodes. This means any point on the Voronoi edges
lifetime can be increased with the uniformity of deployment
has the same distance to its generating node and corresponding
compared to the sensors with an irregular topology. Uniformity
neighbor node. From the characteristic of Voronoi diagram,
(U) is defined as the the average local standard deviation of

T
if any point in a Voronoi polygon cannot be detected by the
the distances between sensors,
v sensor associated with that polygon, no other sensor can detect

IP
N u ki ki it. That is, each sensor only needs to check the coverage of
1 Xu t1
X 1 X
U= ( Di,z − Di,j )2 , (3) its own Voronoi polygon and then the ‘coverage holes’ can
N i=1 ki j=1 ki z=1 be found. Based on the Voronoi diagram, the sensor coverage

CR
problem is turned into the coverage problem of each Voronoi
where Di,j is the distance between ith and jth sensors, ki
polygon, which can reduce the complexity of the problem [28]
stands for the number of neighbors of the ith sensor. The
[30]. In this paper, we consider the influence of communication
smaller value of U represents more uniformly distribution of
range Rc to the partition of Voronoi polygon. Given a convex
the sensors in the 2D sensor field.
5) Connectivity: Connectivity is also an important issue
to be considered during the deployment of sensors, which
influences data routing and information transferring. The net-
work is said to be connected when there is always a network
US sensor region, the condition of Rc ≥ 2Rs is both necessary
and sufficient which can guarantee that fully coverage of
the sensor field implies connectivity [39]. At the beginning
of a centroid-based deployment algorithm, the sensors are
AN
distributed with random initial locations and the network is
connection between two sensors through single hop or multiple
not connected in general. The network is thus divided into
communication hops [37]. However, the network is not always
several subnetworks according to the connectivity. Therefore,
connected even if the implicit assumption exists, such as high
in this paper the Voronoi diagram is created independently for
sensor density or large communication range [38]. In this
the different subnetworks and the sensors in these subnetworks
M

paper, the connectivity is considered during the deployment


are deployed respectively. When there is only one sensor or
of the proposed algorithms.
two sensors in a subnetwork, the sensors must be in a big
converge hole and the Voronoi polygon can not be drawn for
ED

the sensor. Under such situations, the sensors stay where they
C. Testbed-Based Multi-Metric Quality Measurement (T- are for saving energy and do not carry out the deployment
MQM) algorithm until the subnetwork consists at least three sensors.
In [7], seven metrics, which are correlated with each other
PT

and called T-MQM, were proposed to effectively measure B. Centroid and Centroid-Based Scheme
the performance of sensor deployment schemes with high The centroid or geometric center of a polygon is the
precision. T-MQM is designed by concerning the quality arithmetic mean position of all the points in the shape. The
of coverage and connectivity and considering the interde- lines through the centroid can divide the polygon into two parts
CE

pendency among the metrics. Seven metrics in T-MQM are with equal area. The equation for calculating the centroid of
summarised in Table I. The definitions used in Table I are a polygon can be found in [28].
listed in Table II. All the seven correlated metrics will be Centroid-Based Scheme (CBS) [28] is an iterative deploy-
used to evaluate the proposed BCBS and DCBS and reported ment scheme based on Voronoi diagram. At the initial phase,
AC

in Section V. the sensors are deployed randomly in the sensor field. And
then the iterative procedure begins. At the beginning of each
III. P RELIMINARIES iterative, each sensor constructs its local Voronoi polygon
regarding its neighbor’s and own location information. The
A. Voronoi diagram and The Partition of Voronoi Polygon sensor field is then partitioned into several Voronoi polygons
Given the set of sensors S and the sensor field mentioned and each sensor only needs to sense and detect in its local
in Section II, the network is represented as a graph and Voronoi polygon. In the next phase, each sensor checks
each node in the graph is regarded as a sensor. The Voronoi whether all the vertices of its local Voronoi polygon are in
diagram partitions the sensor field into N convex polygons. the sensor’s sensing range. If yes, the Voronoi polygon is
Each polygon ci contains only one sensor si which is called considered fully covered by the sensor and then the sensor
the generating node of polygon ci [27]. The Voronoi polygon, stays in its current location without any movement. Otherwise,
or Voronoi region, ci , coordinated of its generating node si is the sensor calculates the centroid of its local Voronoi polygon
the set of all points in the sensor field whose distances to si and moves to the centroid as its next location if the local
ACCEPTED MANUSCRIPT

SHELL et al.: BARE DEMO OF IEEETRAN.CLS FOR JOURNALS 5

TABLE I
S EVEN CORRELATED METRICS IN T-MQM

Performance Metrics Formulations


Total Coverage Area (CaT ) CaT = CaT O + CaT N O
T TO T NO
Ca Ca +Ca
Effective Coverage Area (CaE ) CaE = 1
N Ca
= 2
N πRs
IO
Ca
Net Effective Coverage Area (CaN E ) CaN E = Ca1 − CaIO = πRs2 (1 − πRs2 )
CaIO
2
NE πRs (1− 2 ) IO
Ca πRs Ca
Net Effective Coverage Area Ratio (CaN ER ) CaN ER = Ca1 = 2
πRs
=1− πRs2
IO
Ca
Total Non-Overlapped Coverage Area (CaT N O ) CaT N O = N CaN E = N (πRs2 − CaIO ) = N πRs2 (1 − πRs2 )
IO
Ca
Total Overlapped Coverage Area (CaT O ) CaT O = CaT − CaT N O = CaT − {N πRs2 (1 − 2
πRs
)}
TO 2 IO

T
KCa N πRs N Ca
Quality of Connectivity (Qc ) Qc = E
Ca
= KN πRs2 {1 − CaT + CaT }

IP
TABLE II
D EFINITIONS USED IN TABLE I

CR
Definitions
CaT O The sum of the total overlapped coverage area in a sensing field.
CaT N O The sum of the total non-overlapped coverage area in a sensing field.
Ca1 The coverage area by an individual sensor.
CIO The overlapped coverage area within an individual sensor’s coverage area.
K The quality of connectivity conversion constant. K = 1 means ideal case.

coverage by the sensor can be increased. The total coverage


by all the sensors is much greater and the sensors is more
US
AN
evenly distributed by the iteratively procedure. The iterations
terminate when pre-defined maximum number of iteration
is reached or the total coverage can not be increased by
a certain amount. CBS is easy to be implemented without
any parameters and helps to increase the coverage. The main
M

characteristic of CBS is the approach to decide the new


location of a sensor based on the centroid of its local Voronoi
(a) E = 1 (b) E = 1
polygon.
ED

C. Ideal Distribution
PT

Consider the coverage problem stated in Section II, under


the ideal case in which node density is sufficiently high,
to fully cover the sensor field with minimal overlap (or the
minimal sensors), the three adjacent√sensors should form an
CE

equilateral triangle with side length 3Rs [39]. Based on this


result, we propose an approach to estimate the minimal sensors
NT for the coverage problem. In our approach, we consider
the X-direction and Y-direction of the sensor field respectively (c) E = 2
AC

and summarize the complex distribution cases on the margin Fig. 1. Three distribution cases on the margin
of the sensor field. Then three kinds of cases can be obtained
as shown in Fig. 1. The formulas to calculate the NT are given
by Eqs. (5) and (6), IV. T HE P ROPOSED S ENSOR D EPLOYMENT S CHEMES
l L m j W k A. Blind-zone CBS (BCBS)
NT = √ ·( · 2 + E), (5)
3Rs 3Rs Definition 3. An Voronoi blind-zone refers to the uncovered
 W
1, 0 ≤ rem( 1.5R ) ≤ Rs area if its coordinate sensor is removed.
E= W
s
, (6)
2, Rs < rem( 1.5Rs ) ≤ 1.5Rs
Definition 4. An Voronoi blind-zone polygon is the polygon
whose vertices are the interactions of its coordinate Voronoi
where rem(·) is the function to take the remaining. Three
blind-zone.
distribution cases shown in Fig. 1(a)∼(c) correspond to the
three cases of E in (6). There are three cases of the Voronoi blind-zone polygon
ACCEPTED MANUSCRIPT

6 JOURNAL OF LATEX CLASS FILES, VOL. 0, NO. 0, MAY 2017

regarding the vertices’ position of the Voronoi polygon. The centroid of the Voronoi blind-zone polygon is marked by a
Case 1: The Voronoi polygon vertices are all covered by black triangle and the black cross is the centroid of the Vonoroi
its coordinate sensor and the Voronoi blind-zone may emerge polygon v1 v2 v3 v4 v5 . From Fig. 3(b), it is more reasonable for
if the sensor is removed. If the Voronoi blind-zone exists, the the sensor O to move to the centroid of the Voronoi blind-zone
Voronoi blind-zone polygon can be constructed whose vertices polygon than the centroid of the Voronoi polygon.
are the intersections between any two neighbors’ sensing disks.
For instance, consider the sensor O in Fig. 2(a), four vertices
of its Voronoi polygon, which are marked by four little circles,
are all covered and the black cross is the centroid of the
Voronoi polygon. The sensor O has four neighbors, which
are A, B, C, and D. The coverage area by these five sensors
is marked gray. As shown in Fig. 2 (b), the sensor O is

T
removed and the Voronoi blind-zone is emerged, which is
shown in white. And then the Voronoi blind-zone polygon

IP
can be determined by connecting the intersections between
the sensing disks of A and B, B and C, C and D, and D and
A, which are marked by four white little circles. In Fig. 2(b),

CR
the black triangle is the centroid of the Voronoi blind-zone (a) (b)
polygon.
Fig. 3. Voronoi blind-zone polygon: Case 2

US Case 3: All the vertices of an Voronoi polygon are not


covered by its coordinate sensor. Then the Voronoi blind-zone
polygon is equivalent to the Voronoi polygon.
As illustrated above, the centroid of Voronoi polygon is
AN
sometimes unreasonable to be a candidate location for a
sensor to move in CBS. The reason is that there is overlap-
ping covered area by a sensor and its neighbors, while the
overlapping covered area is not taken into account in CBS.
Therefore, we propose to find the Voronoi blind-zone polygon
M

and take its centroid as the candidate location for a sensor to


(a) (b) move. The new deployment scheme is called BCBS which can
improve the coverage and the uniformity of the distribution.
Fig. 2. Voronoi blind-zone polygon: Case 1 The procedure of the BCBS is given by Algorithm 1.
ED

Case 2: The Voronoi polygon vertices are not fully covered


by its coordinate sensor and there must be a Voronoi blind- B. Disturbed CBS (DCBS)
zone when the sensor is removed. In this case, the vertices of For the deployment strategies based on CBS, an important
PT

the corresponding Voronoi blind-zone polygon are composed step is to find a new location for each sensor according to
of two parts. One part is the uncovered vertex (vertices) of the Voronoi diagram. The new location replaces the sensor’s
the Voronoi polygon when the sensor is removed. For the current location if the coverage area w.r.t the new location in
remaining covered vertices of the Voronoi polygon, they are the old Voronoi polygon increases. By the iterative procedure,
CE

covered by the neighbors of the sensor. Then the other part the total coverage can increase. These strategies can reduce
is the intersections between the sensing disks of those sensors the complexity of the deployment problem. However, the
where the covered Voronoi vertices lie in. overlapping coverage area w.r.t. the sensor’s neighbor(s) is
Fig. 3 presents the second case of Voronoi blind-zone poly- not considered in these strategies, which may not ensure the
AC

gon. For the sensor O, only the vertex v1 of its corresponding coverage for the sensor in its coordinate Voronoi polygon.
Voronoi polygon v1 v2 v3 v4 v5 is in its sensing range which can Another problem of these strategies is that just regarding the
be seen from Fig. 3(a). As shown in Fig.3(b), the sensor O is coverage as the objective in each iteration limits the diffusion
removed, part area of its corresponding Voronoi polygon is still among the sensors and leads to stagnate of sensors termed
covered by its neighbors A, B, C and D, while the remaining as local convergence of the deployment algorithms. Here we
area can not be covered by any sensor. Then the Voronoi blind- firstly show the cases of local convergence and then propose
zone arises. The vertex v5 is now covered by the sensors B the operators in DCBS, which are local reconstruction and
and C. There is an intersection v7 between the sensing disks local perturbation.
of B and C. Similarly, the vertex v4 is covered by the sensors 1) Local convergence: Case 1: Fig. 4(a) shows an instance
C and D and the intersection v6 is between the sensing disks that most of the sensors have found their location where their
of C and D. Therefore, by the three uncovered vertices v1 , coordinate Voronoi polygon can be maximally covered. In
v2 , and v3 and two intersections v6 and v7 , the Voronoi blind- Fig. 4(a), the sensor s13 is also covered by the sensing disk
zone polygon can be obtained by connecting these five points. of its neighbor s17 and it is clearly that these two sensing
ACCEPTED MANUSCRIPT

SHELL et al.: BARE DEMO OF IEEETRAN.CLS FOR JOURNALS 7

Algorithm 1 Procedure of BCBS


1: Notations: 12
5
2: S = {s1 , s2 , · · · , sN } : N sensors 3
3: V = {v1 , v2 , · · · , vN } : Voronoi polygon set of N sensors
7
27 2 2
4: Zi = {zi1 , zi2 , · · · : n neighbors of the sensor si
, zin } 18 18
5: Ki = {si } ∪ Zi = {si , zi1 , zi2 , · · · , zin }
6: VKi = {vi , vzi1 , vzi2 , · · · , vzin , } : Voronoi polygon set of 22 22
the sensors in the set Ki 20
13
25 25
7: Main Procedure: 17 17
8: Randomly deploy N sensors in the 2D sensor field;
(a) (b)
9: repeat

T
10: Construct the Voronoi polygons for all the N sensors; Fig. 4. Local convergence: Case 1 (a) and Local reconstruction (b)
11: for i = 1 to N do

IP
12: Consider the sensor si and then determine the sets
Zi , Ki , and VKi ; operator on the sensors regardless of whether the coverage
13: if All the vertices of vi are covered by si then is increased after the movement of the sensor. When the

CR
14: // Voronoi blind-zone polygon: Case 1 distance between a sensor and any of its neighbors is less
15: if The Voronoi blind-zone doesn’t exist then than the sensing range Rs , the Voronoi polygons should
16: Move si to the centroid Ci of vi ; be locally reconstructed where the sensor is assumed to be
17: else {The Voronoi blind-zone exists} removed and only its neighbors are considered. If any vertex
18:
19:
20:
21:
Call SubProcedure;
end if
else {The vertices of vi are not fully covered by si }
// Voronoi blind-zone polygon: Case 2 & 3
US of the new locally reconstructed Voronoi polygons, which is
in the original Voronoi polygon, can not be covered by the
neighbors, the vertex must be the most difficult to be covered
by the neighbors [40]. Therefore, the sensor can move to the
AN
22: Call SubProcedure; location of such vertices in order not to be covered by its
23: end if neighbors. Taken the sensor s13 in Fig. 4(a) as an example,
24: end for its neighbors are s2 , s18 , s22 , s17 , and s25 . As the distance
25: until termination criterion is met. between s13 and s17 is less than Rs , s13 should carry out the
local reconstruction. For the local reconstruction of s13 , s13
M

26: SubProcedure:
27: Calculate the centroid Ci of the Voronoi blind-zone poly- is removed and only its neighbors are taken into account to
gon; reconstruct the local Voronoi polygons as shown in Fig. 4(b).
28: Calculate the coverage Qicen and Qicur of si to vi if si is The reconstructed Voronoi polygons are plotted by black dash
ED

at Ci and its current location [36]; lines. The origin Voronoi polygon of s13 is plotted by black
29: if Qicen ≥ Qicur then bold lines and used to discover the uncovered vertices. From
30: Move si to (cxi , cyi ); Fig. 4(b), it can be found that three Voronoi vertices (denoted
31: else {Qicen < Qicur } by black triangles) can not be covered by any sensors and then
PT

32: si stays without movement; the next location of s13 is randomly chosen from the location
33: end if of these three vertices. By the local reconstruction on s13 , it
is helpful for s2 moving to the coverage holes.
3) Local convergence: Case 2: In Fig. 5, the biggest cover-
CE

disks have large overlapping area, which wastes the sensing age hole, which is shown in white in the left bottom corner, is
ability of the sensors. Similar situations can be found on the located in the Voronoi polygons of the sensors s21 and s25 . The
sensors s17 and s22 . As these sensors have fully covered their minimum circumscribed circles and the maximum inscribed
coordinate Voronoi polygons, they can hardly move to the new circles for the Voronoi polygons of s21 and s25 are plotted in
AC

location through evaluating the coverage. At the same time, the figure with red lines, and their centers are marked as stars
the sensing disks of the other sensors have been used too much and crosses respectively. The centroid of these two Voronoi
and therefore there is not enough sensing ability to cover the polygons are marked as triangles in the figure. If the sensors
holes. Such situation will be frequently happened during the s21 and s25 move to the center of the minimum circumscribed
later iterations. In this paper, we regard the situation as the circle or the maximum inscribed circle or the centroid of the
first case of local convergence. Voronoi polygon, the biggest uncovered area can be covered
2) Local reconstruction for solving local convergence of more but new coverage holes will be generated. Then the
case 1: From the illustration for local convergence of case sensors s21 and s25 may not move to these locations w.r.t.
1, a main reason leads to the local convergence is that two the coverage. Here we term this situation as the second case
sensors are too close to each other. It is very easy to calculate of local convergence. In such case, if the sensors move to
the distance between two sensors, but the difficulty is how the farthest location of the coverage hole directly regardless
to move the sensors and help to increase the coverage with of the coverage, the coverage will decrease but it will benefit
a minimum cost. Here we propose the local reconstruction the movement of all the sensors in order to avoid the local
ACCEPTED MANUSCRIPT

8 JOURNAL OF LATEX CLASS FILES, VOL. 0, NO. 0, MAY 2017

Algorithm 2 Procedure of DCBS


1: Notations:
2: S, V, and Zi share the same meaning as those in Algorithm
1
3: Procedure:
4: Randomly deploy N sensors in the 2D sensor field;
5: repeat
6: Construct the Voronoi polygons for all the N sensors;
7: for i = 1 to N do
8: Consider the sensor si and determine Zi ;
9: Calculate the centroid Ci of vi and the coverage Qicen
Fig. 5. Local convergence: Case 2
of si to vi if si is at Ci [36];

T
10: Calculate the coverage Qicur of si to vi if si is at its
convergence. current location [36];

IP
4) Local perturbation for solving local convergence: In 11: if Qicen ≥ Qicur then
this paper, we propose the local perturbation operator on the 12: Move si to Ci ;
13: else {Qicen < Qicur }

CR
sensors to move them. Under the local perturbation operator,
if a sensor detects the existence of the uncovered vertices 14: si stays without movement;
of the Voronoi polygon, the sensor will move to the farthest 15: end if
uncovered Voronoi vertex and stop when the farthest uncov- 16: // Local perturbation begins
ered Voronoi vertex can be covered by the sensor. Another 17: if All the vertices of vi have been covered by si then
possibility in local perturbation is that all the Voronoi vertices
are already covered, then a sensor will move away from
the farthest Voronoi vertex and stop when the farthest vertex
can be covered by the sensor. Fig. 6 shows two operational
US 18:

19:
Move si away from the farthest Voronoi vertex and
stop when the farthest Voronoi vertex reaches the
sensing range of si ;
else {Any vertex of vi has not been covered by si }
AN
example of these two situations where O0 is the new location 20: Move si to the farthest uncovered Voronoi vertex
of the sensor O after the local perturbation operator. The and stop when the farthest uncovered Voronoi
movement of the sensor smoothes the differences of the vertex can be covered by the sensor;
distance between the sensor and the Voronoi vertices. 21: end if
M

22: // Local reconstruction begins


23: if The distance between si and any of its neighbors
in Zi is less than Rs then
24: Reconstruct the Voronoi polygons for the sensors
ED

in Zi after removing si ;
25: Move si to one of the uncovered Voronoi vertices
which are located in vi ;
26: end if
PT

27: end for


28: until termination criterion is met.
(a) (b)
CE

Fig. 6. Snapshots of the movement of the sensor with uncovered and fully
covered Voronoi vertices under local perturbation operation initialization. Therefore these three sensors only stay where
they are and need not partition the Voronoi polygons until
According to the description of local convergence and the they communicate at least two other sensors. The sensors 1,
proposed local perturbation operator, the procedure of the 5, and 14 form an independent area and partition the Voronoi
AC

DCBS is given by Algorithm 2. polygons by themselves as they cannot communicate with


Fig. 7(a)-(c) shows the three snapshots of deployment proce- the other sensors. The same situation can be found among
dure of DCBS with 30 sensors in a 50m×50m 2D sensor field. the sensors 8, 9, 21, and 26. The sensors move according
The sensing range Rs and communication range Rc of a sensor to the deployment procedure of DCBS in each round before
is 6m and 12m respectively. The maximum step stepmax for all the 30 sensors are fully connected. Once all the sensors
every move of a sensor is half of the communication range Rc are connected, the sensors will stay at their current location
[26]. The sensors, which can communicate with each other, and move to the final location directly until the deployment
are connected with green lines. From the initial round to the procedure terminates.
fourth round, the coverage increases from 69.97% to 89.15%.
After 200 rounds, the final coverage reaches 99.35% and the C. Computational complexity of BCBS and DCBS
sensors are distributed more uniformly. As shown in Fig. 7(a), From the description of BCBS, the computational burden of
the sensor 27 cannot communicate with any other sensors and BCBS mainly comes from calculating the centroid of Voronoi
the sensors 20 and 29 only communicate with one sensor after blind-zone polygon and Voronoi polygon. To develop the
ACCEPTED MANUSCRIPT

SHELL et al.: BARE DEMO OF IEEETRAN.CLS FOR JOURNALS 9

27 29
29 27 29 27
20 20
20

14
14

14 1 5 1

5 5
1
9
9 26
9
26 26
21 8
8 21 8
21

T
(a) Initial Round (69.97%) (b) Round 4 (89.15%) (c) Final Round (99.35%)

IP
Fig. 7. Snapshots of the movement of sensors with Voronoi polygons, sensing circles and connections using DCBS

CR
maximum inscribed circle and the minimum circumscribed TABLE III
circle form the main items of the computational complexity of PARAMETER SETTINGS FOR E XAMPLE 1
DCBS. Suppose the number of edges and vertices of a Voronoi Rc Minimal Maximum Rc Minimal Maximum
polygon to be m and k respectively, then the complexity of increment round increment round
Case 1 1% 30 Case 3 0.1% 100

US
BCBS and DCBS is O(m3 ) and max{O(k 3 ), O(m4 )} [27]. 12m 12m
Case 2 20m 1% 30 Case 4 20m 0.1% 100

V. P ERFORMANCE EVALUATION
In this section, we evaluate and compare the performance Fig. 8 gives the average coverage at each round using six
AN
of BCBS and DCBS with four other deployment algorithms, algorithms under four parameter settings. As shown in the
which are CBS [28], VOR [26], Minimax [26] and VEDGE figure, DCBS shows the best coverage performance followed
[27]. Both examples are conducted in the 2D sensor field by BCBS in all four cases. It can be observed from the
measuring 50m × 50m. The sensing range Rs of a sensor is figure that the coverage curve by DCBS is not as smooth as
6m. Then according to (5)-(6), the minimal number of sensors
M

those by the other five schemes as local reconstruction and


to fully cover the 50m × 50m 2D sensor field is 30 under local perturbation operators in DCBS are executed without
the ideal distribution. The maximum step stepmax for every measuring the coverage. With the deployment procedure of
movement of a sensor is set to half of the communication DCBS goes on, the convergence curve ranges smaller and
ED

range Rc . smaller. In terms of the termination condition, the coverage


under the second termination condition is always better than
A. Example 1 that under the first termination condition for all algorithms.
In the first example, the number of sensors is 30, which are This is not surprise as the sensors need more rounds to
PT

randomly deployed in the sensor field at the beginning. Two stop under the second termination condition and then higher
kinds of communication range (Rc = 12m and Rc = 20m) are coverage can be reached. From Figs. 8(a) ∼ (b), the coverage
considered individually to compare the performance change is approaching convergence after about 10 rounds using BCBS,
on the schemes. One termination condition for all the six CBS, VEDGE, Minimax, and VOR, but the coverage by
CE

deployment strategies is when no sensor’s coverage in its DCBS keeps on increasing until the final round. In Case 3
Voronoi polygon increases by more than 1% in its next move and 4, the coverage by DCBS also continues to increase until
and the maximum round is 30. This termination condition is the end while the coverage by other five schemes is tending
suitable for the situations that the sensors are not necessary towards stable after about 20 rounds. From Figs. 8(a) ∼ (d), we
AC

to cover the smaller coverage holes in order to save energy can see that the coverage increases very quickly by BCBS with
and reduce the deployment time. For another termination good performance during the first several rounds. It should be
condition, the deployment algorithm stops when no sensor’s noted that, in the same simulation case, the average stopping
coverage in its Voronoi polygon increases by more than 0.1% rounds for each algorithm is quite different. For convenience of
in its next move and the maximum round is 100. Therefore comparing the average coverage from the figures, we plot the
there are four cases w.r.t. the communication ranges and curves of the average coverage for six algorithm in each case
termination conditions, which are listed in Table III. Each from the initialization to the maximum number of iteration.
simulation case runs 100 times with different initial locations Table IV shows the average values and standard deviations
of sensors. The initial sensor deployment in six deployment of stopping rounds, uniformity at the final round, connected
algorithms are the same for fair comparison. The deployment rounds when the network is conneted, deployment time, and
algorithms are compared in terms of coverage, deployment moving distance by using six algorithms after 100 individual
time, moving distance, connected round, stopping round, and runs in four cases. We also record the number of disconnected
uniformity. times after 100 random simulations in Table IV. From the
ACCEPTED MANUSCRIPT

10 JOURNAL OF LATEX CLASS FILES, VOL. 0, NO. 0, MAY 2017

1 1

0.95 0.95

Coverage(%)
Coverage(%) 0.9 0.9

0.85 VOR 0.85 VOR


Minimax Minimax
0.8 VEDGE 0.8 VEDGE
CBS CBS
0.75 BCBS 0.75 BCBS
DCBS DCBS
0.7 0.7

T
0 10 20 30 0 10 20 30
Round Round

IP
(a) Case 1 (b) Case 2

1 1

CR
0.95 0.95

0.9 0.9
Coverage(%)
Coverage(%)

0.85 0.85

0.8

0.75
VOR
Minimax
VEDGE
CBS
BCBS
US 0.8

0.75
VOR
Minimax
VEDGE
CBS
BCBS
AN
DCBS DCBS
0.7 0.7
0 20 40 60 80 100 0 20 40 60 80 100
Round Round

(c) Case 3 (d) Case 4

Fig. 8. The average coverage at each round using six schemes with four parameter settings
M

TABLE IV
T HE EXPERIMENTAL RESULTS BASED ON COMMONLY USED METRICS BY USING SIX ALGORITHMS IN FOUR CASES
ED

VOR Minimax VEDGE CBS BCBS DCBS


The average stopping rounds by standard deviations
Case 1 6.42 ± 1.68 9.69 ± 2.89 8.45 ± 2.32 9.34 ± 2.31 9.82 ± 2.48 29.17 ± 3.09
Case 2 6.03 ± 1.55 9.38 ± 2.53 8.43 ± 2.74 9.46 ± 3.01 9.82 ± 2.91 29.07 ± 3.08
Case 3
PT

10.01 ± 2.72 20.71 ± 5.11 20.58 ± 6.05 28.91 ± 21.98 20.69 ± 6.08 99.79 ± 2.10
Case 4 9.49 ± 2.66 21.26 ± 7.07 20.9 ± 5.80 30.03 ± 22.31 21.28 ± 7.10 99.75 ± 1.77
The average uniformity by standard deviations
Case 1 1.97 ± 0.23 1.83 ± 0.22 1.72 ± 0.19 1.62 ± 0.17 1.50 ± 0.15 1.40 ± 0.14
Case 2 2.03 ± 0.24 1.85 ± 0.21 1.74 ± 0.20 1.62 ± 0.15 1.50 ± 0.15 1.39 ± 0.14
CE

Case 3 1.82 ± 0.24 1.54 ± 0.21 1.44 ± 0.18 1.32 ± 0.13 1.24 ± 0.12 1.18 ± 0.10
Case 4 1.88 ± 0.23 1.55 ± 0.21 1.45 ± 0.14 1.32 ± 0.12 1.24 ± 0.12 1.23 ± 0.11
The average connected rounds by standard deviations
Case 1 4.33 ± 1.42 3.42 ± 1.28 3.20 ± 1.09 3.14 ± 0.96 2.35 ± 0.90 2.40 ± 0.92
Case 2 1.01 ± 0.10 1.01 ± 0.10 1.01 ± 0.10 1.01 ± 0.10 1.02 ± 0.20 1.01 ± 0.10
Case 3 4.20 ± 2.50 4.05 ± 1.39 3.36 ± 1.41 2.40 ± 1.17 2.34 ± 0.89 2.37 ± 1.03
AC

Case 4 1.01 ± 0.10 1.01 ± 0.10 1.01 ± 0.10 1.01 ± 0.10 1.02 ± 0.20 1.01 ± 0.10
Number of disconnected times
Case 1 7 3 3 3 0 0
Case 2 0 0 0 0 0 0
Case 3 6 4 3 0 0 0
Case 4 0 0 0 0 0 0
The average deployment time(sec) by standard deviations
Case 1 45.92 ± 34.84 54.53 ± 32.50 62.21 ± 44.00 42.28 ± 32.50 52.15 ± 23.56 63.57 ± 34.21
Case 2 12.45 ± 3.64 17.37 ± 5.61 16.16 ± 6.46 15.08 ± 5.55 14.46 ± 4.33 27.75 ± 6.41
Case 3 49.85 ± 39.00 66.00 ± 43.77 72.05 ± 47.46 57.71 ± 43.93 53.41 ± 21.87 112.25 ± 50.25
Case 4 16.60 ± 5.96 31.98 ± 11.14 28.66 ± 12.32 28.82 ± 8.66 28.19 ± 10.56 81.50 ± 14.54
The average moving distance by standard deviations
Case 1 5.67 ± 1.23 4.21 ± 1.74 4.24 ± 1.77 4.49 ± 1.95 4.88 ± 1.82 5.39 ± 1.15
Case 2 2.63 ± 1.20 3.21 ± 1.26 3.27 ± 1.18 3.39 ± 1.08 3.87 ± 1.30 6.18 ± 1.82
Case 3 4.43 ± 1.82 5.52 ± 2.23 5.43 ± 2.20 5.65 ± 2.10 5.88 ± 1.79 5.61 ± 1.15
Case 4 2.97 ± 1.21 4.10 ± 1.50 4.14 ± 1.32 4.38 ± 1.36 4.80 ± 1.54 6.60 ± 2.08
ACCEPTED MANUSCRIPT

SHELL et al.: BARE DEMO OF IEEETRAN.CLS FOR JOURNALS 11

TABLE V
T HE EXPERIMENTAL RESULTS BASED ON THE METRICS IN T-MQM BY USING SIX ALGORITHMS IN FOUR CASES

VOR Minimax VEDGE CBS BCBS DCBS


Effective Coverage Area
Case 1 0.6626 0.6822 0.6890 0.6962 0.7040 0.7194
Case 2 0.6548 0.6785 0.6855 0.6937 0.7027 0.7198
Case 3 0.6728 0.7017 0.7063 0.7161 0.7206 0.7297
Case 4 0.6647 0.6999 0.7063 0.7153 0.7206 0.7279
Net Effective Coverage Area
Case 1 50.5071 52.9972 54.0528 54.5398 56.8470 60.9832
Case 2 49.6586 52.5742 53.5824 54.4204 56.5113 60.8061
Case 3 51.8329 56.5542 57.4658 57.5779 59.3326 62.1753
Case 4 50.9417 56.3303 57.1011 57.4659 59.3066 62.0856
Net Effective Coverage Area Ratio

T
Case 1 0.4466 0.4686 0.4779 0.4822 0.5026 0.5392
Case 2 0.4391 0.4649 0.4738 0.4812 0.4997 0.5376
Case 3 0.4583 0.5000 0.5081 0.5091 0.5246 0.5498

IP
Case 4 0.4504 0.4981 0.5049 0.5081 0.5244 0.5490
Total Non-Overlapped Coverage Area
Case 1 1515.2117 1589.9163 1621.5837 1636.1942 1705.4104 1829.4965

CR
Case 2 1489.7577 1577.2271 1607.4715 1632.6125 1695.3382 1824.1829
Case 3 1554.9862 1696.6245 1723.9750 1727.3380 1779.9767 1865.2601
Case 4 1528.2514 1689.9089 1713.0318 1723.9781 1779.1971 1862.5687
Total Overlapped Coverage Area
Case 1 758.3410 724.6743 716.0194 725.9712 683.3402 611.3861
Case 2 732.0718 724.9060 718.5209 721.0445 688.9759 617.9741
Case
Case

Case
Case
3
4

1
2
727.8491
727.1488

1144.5184
1117.9350
684.0549
684.6915

1062.2881
1068.3779
US
672.5230
683.4662
702.3384
702.9817
Quality of Connectivity
1039.2682
1048.1049
1042.7561
1039.4235
665.1266
665.8082

970.5989
980.4246
610.6715
607.1763

849.8500
858.5594
AN
Case 3 1081.7836 974.9081 952.1463 980.7800 922.9555 836.8404
Case 4 1093.8891 978.3135 967.6395 982.7771 923.9383 834.1349

table, it is obviously that DCBS requires the most rounds to average deployment time in all simulations, the coverage by
M

stop and VOR needs the least in all four cases. Therefore, from VOR is the worst among all six algorithms.
Fig. 8 and Table IV, we know that DCBS can keep increasing From the average moving distance listed in Table IV, the
the coverage from initialization to finalization, while for the sensors deployed by VOR travel the shortest path in cases 2,
other five algorithms they generally stop in a few rounds. From 3, and 4. For Minimax, VEDGE, CBS, and BCBS, they have
ED

Fig. 8(a) and (b), we can find that the average coverage in similar performance in terms of the average moving distance
case 1 is slightly higher than that in case 2 as the sensors in four cases.
have smaller communication range in case 1, which leads to In order to further evaluate and compare the coverage and
more movements in the deployment. The same situation can connectivity among the algorithms, the metrics in T-MQM are
PT

be found from Figs. 8(c) and (d). also taken into account in this experiment. Table V lists the
Regarding the average uniformity, the results shown in Table values of seven metrics in T-MQM by using six algorithms
IV quite agree with the coverage by the six algorithms. The after 100 individual runs in four cases. It should be noted
sensors deployed by DCBS are distributed the most evenly that the metrics of total coverage area and coverage rate share
CE

with the highest coverage than that by the other algorithms. the same meaning. Therefore the values of total coverage
As the average connected rounds is concerned, among the area are not given in the Table. According to the meaning
six algorithms in case 1 and case 3, the sensors deployed by of the other six metrics, bigger values of effective coverage
BCBS can connect with the least rounds and DCBS is the area, net effective coverage area, net effective coverage area
AC

second one. For case 2 and case 4, the average rounds are ratio, and total non-overlapped coverage area reflect better
almost the same for all six algorithms. From another point of performance of the deployment schemes. It can be found from
view, all the sensors can be connected using DCBS and BCBS Table V that DCBS obtained the biggest values in all the six
in all four cases, which can be seen from the the values of metrics and follows by BCBS. On the contrary, lower values
the number of disconnected times. For case 2 and case 4, of total overlapped coverage area and quality of connectivity
there is no disconnected situation using any algorithm, which means better performance of the deployment patterns. One can
is attributed to the big communication range of the sensors. also observed that DCBS and BCBS show better performance
It can be found that DCBS takes the longest deployment compared to the other four algorithms.
time in all four cases. This is inevitable for DCBS as it deploys
the sensors almost during the whole rounds which can be B. Example 2
seen from the average stopping rounds. Compared to Minimax, In the second example, the simulation cases are carried out
VEDGE, and CBS, BCBS uses the shortest deployment time. with more sensors than that in case 1 and two kinds of termi-
Although VOR shows the best performance in terms of the nation conditions are used. The parameter settings, including
ACCEPTED MANUSCRIPT

12 JOURNAL OF LATEX CLASS FILES, VOL. 0, NO. 0, MAY 2017

the number of sensors, maximum round, and communication sensors to maximize the sensing coverage in a given sensing
range are listed in Table VI. Both simulation cases run 100 field. The proposed two deployment algorithms are designed
times with the sensors randomly deployed in the initial round based on the Voronoi diagram and the centroid of Voronoi
and the results are the average values with standard deviations. polygon. The main contribution of BCBS is the introduction of
the Voronoi blind-zone and the Voronoi blind-zone polygon.
A sensor’s Voronoi blind-zone is help to find the coverage
TABLE VI holes, which is de termined by considering the neighbors’
PARAMETER S ETTINGS FOR E XAMPLE 2
sensing range to its Voronoi polygon. For a sensor in BCBS,
Rc Minimal Maximum Number of the centroid of its Voronoi blind-zone polygon is the candidate
increment round sensors location instead of the centroid of Voronoi polygon. DCBS
Case 5 12m 1% 30 31/32/33/34/35
Case 6 12m 0.1% 100 31/32/33/34/35
is a combination of CBS and two local disturbed operators.
After each sensor deploys according to CBS, the sensor carries

T
out the local perturbation operator by checking whether all
Fig. 9 shows comparisons of the average coverage and
the Voronoi vertices are inside its sensing range. If yes, the

IP
standard deviations with different number of sensors by six
sensor moves away from the farthest Voronoi vertex, which
algorithms in two simulation cases. As can be seen from the
can reduce the variants of the distances between the sensor
figure, DCBS and BCBS get better coverage performance with
and the Voronoi vertices; otherwise, the sensor is forced to

CR
less standard deviation than CBS, VEDGE, Mnimax, and VOR
move towards the farthest uncovered Voronoi vertex in order
irrespectively the number of sensors and the threshold of the
to heal the coverage hole. The local reconstruction operator
termination conditions. When the number of sensors increases,
for a sensor in DCBS works following local perturbation
the tendency of the coverage for all the compared algorithms
operator if the distance to any of its neighbors is less than
also increases in two simulation cases.
In Fig. 10, the average number of stopping rounds with
standard deviations of six algorithms for different number
of sensors is shown. From the figure, it is clear that for
US the sensing range. In such case, only the sensor’s neigh-
bors need to reconstruct their Voronoi polygons followed by
checking whether their Voronoi vertices are covered. Once
the uncovered Voronoi vertices exist, the sensor selects one
AN
DCBS the number of stopping rounds decreases sharply as the
vertex randomly as its new location. Compared with the other
number of sensors increases from 31 to 35 in both cases. And
four known algorithms based on commonly used performance
among VOR, Minimax, VEDGE, CBS, and BCBS, there is not
metrics and seven correlated performance metrics with high
much difference on the number of stopping rounds when the
precision, simulation results show that DCBS is quite good
number of sensors changes. However, the number of stopping
M

at maximizing the coverage with the most evenly distribution.


rounds varies considerably for CBS in case 6 when the number
BCBS can converge with higher coverage in a shorter time
of sensors is different. VOR is the quickest algorithm to
and moderate in the deployment time, moving distance and
terminate the deployment procedure in all scenarios as the
uniformity.
ED

sensor densities are not high that it is ineffective to move the


sensors frequently [26]. ACKNOWLEDGMENT
Fig. 11 depicts the average uniformity with standard devia-
tion for different number of sensors using six algorithms. It can Thanks are due to anonymous referees for their con-
be observed from the figure that the general trend of uniformity structive comments. This work was partially supported by
PT

for the algorithms in two cases is increasing as the number of the National Natural Science foundation of China (Grant
sensors increases. This is due to the fact that for the 50m∗50m Nos. 61673194, 61105128), the Natural Science Founda-
2D sensor field in the simulation, 30 sensors with Rs = 12m tion of Jiangsu Province, China (Grant No. BK20131106),
the Postdoctoral Science Foundation of China (Grant No.
CE

can maximize sensor coverage under the ideal distribution and


result in the best uniformity. With more sensors than 30, the 2014M560390), the Fundamental Research Funds for the
uniformity gets worse. Central Universities, China (Grant No. JUSRP51410B), Six
Fig. 12 shows the average moving distance from the initial Talent Peaks Project of Jiangsu Province (Grant No. DZXX-
025).
AC

round to the stopping round for different number of sensors


using six algorithms. When the number of sensors increases,
R EFERENCES
it can be seen that the average moving distance deceases for
all six algorithms in two cases. This is because the sensors [1] G. Y. Keung, L. Bo, Z. Qian, and Y. Hai-Dong, “The target tracking
in mobile sensor networks,” in 2011 IEEE Global Telecommunications
can cover their coordinate Voronoi polygons with higher Conference (GLOBECOM 2011), 2011, pp. 1–5.
probability when the number of sensor increases beyond a [2] K. Ramya, K. P. Kumar, and V. S. Rao, “A survey on target tracking
certain amount. Therefore, the deployment procedure will stop techniques in wireless sensor networks,” International Journal of Com-
puter Science and Engineering Survey, vol. 3, no. 4, pp. 93–108, 2012.
in a short period to meet the termination condition, which will [3] A. Abrardo, M. Martal, and G. Ferrari, “Information fusion for efficient
result in the decrease of moving distance. target detection in large-scale surveillance wireless sensor networks,”
Information Fusion, vol. 38, pp. 55–64, 2017. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1566253517300751
VI. C ONCLUSION [4] I. Benkhelifa, N. Nouali-Taboudjemat, and S. Moussaoui, “Disaster man-
agement projects using wireless sensor networks: An overview,” in Ad-
In this paper, two novel deployment algorithms, BCBS and vanced Information Networking and Applications Workshops (WAINA),
DCBS, are proposed to address the problem of placing the 2014 28th International Conference on. IEEE, 2014, pp. 605–610.
ACCEPTED MANUSCRIPT

SHELL et al.: BARE DEMO OF IEEETRAN.CLS FOR JOURNALS 13

1 1

0.98 0.98
Average Coverage

Average Coverage
0.96 0.96

0.94 0.94

0.92 0.92

0.9 0.9

0.88 0.88
31 32 33 34 35 31 32 33 34 35
Number of sensors Number of sensors

T
(a) Case 5 (b) Case 6

Fig. 9. Average coverage by standard deviations for different number of sensors using six algorithms with two parameter settings

IP
30 100

CR
Average stoping round
Average stoping round

25 80

20 60

15

10

5
US
40

20

0
AN
31 32 33 34 35 31 32 33 34 35
Number of sensors Number of sensors

(a) Case 5 (b) Case 6

Fig. 10. Averge number of stopping rounds by standard deviations for different number of sensors using six algorithms with two parameter settings
M

2.2

1.8
2
Average uniformity

Average uniformity
ED

1.6
1.8

1.6 1.4
PT

1.4 1.2

1.2 1
31 32 33 34 35 31 32 33 34 35
Number of sensors Number of sensors
CE

(a) Case 5 (b) Case 6

Fig. 11. Average uniformity by standard deviations for different number of sensors using six algorithms with two parameter settings
AC

7 6.5
Average moving distance

Average moving distance

6 6

5 5.5

5
4
4.5
3
4
2
3.5

1 3
31 32 33 34 35 31 32 33 34 35
Number of sensors Number of sensors

(a) Case 5 (b) Case 6

Fig. 12. Average moving distance by standard deviations for different number of sensors using different algorithms
ACCEPTED MANUSCRIPT

14 JOURNAL OF LATEX CLASS FILES, VOL. 0, NO. 0, MAY 2017

[5] L. M. Oliveira and J. J. Rodrigues, “Wireless sensor networks: a survey International Conference on Distributed Computing in Sensor Systems
on environmental monitoring,” Journal of communications, vol. 6, no. 2, (DCOSS). IEEE, 2012, pp. 359–363.
pp. 143–151, 2011. [23] Y. H. Han, Y.-h. Kim, W. Kim, and Y.-S. Jeong, “An energy-efficient
[6] M. F. Othman and K. Shazali, “Wireless sensor network applications: self-deployment with the centroid-directed virtual force in mobile sensor
A study in environment monitoring system,” Procedia Engineering, networks,” Simulation, vol. 88, no. 10, pp. 1152–1165, 2012.
vol. 41, pp. 1204–1210, 2012. [24] N. Boufares, I. Khoufi, P. Minet, L. Saidane, and Y. Ben Saied, “Three
[7] O. Kaiwartya, A. H. Abdullah, Y. Cao, R. S. Raw, S. Kumar, D. K. dimensional mobile wireless sensor networks redeployment based on
Lobiyal, I. F. Isnin, X. Liu, and R. R. Shah, “T-mqm: Testbed-based virtual forces,” in 2015 International Wireless Communications and
multi-metric quality measurement of sensor deployment for precision Mobile Computing Conference (IWCMC). IEEE, 2015, pp. 563–568.
agriculture: A case study,” IEEE Sensors Journal, vol. 16, no. 23, pp.
[25] F. Aurenhammer, R. Klein, D.-T. Lee, and R. Klein, Voronoi diagrams
8649–8664, 2016.
and Delaunay triangulations. World Scientific, 2013.
[8] M. Z. A. Bhuiyan, W. Guojun, C. Jiannong, and W. Jie, “Deploying
wireless sensor networks with fault-tolerance for structural health mon- [26] G. Wang, G. Cao, and T. La Porta, “Movement-assisted sensor deploy-
itoring,” IEEE Transactions on Computers, vol. 64, no. 2, pp. 382–395, ment,” IEEE Transactions on Mobile Computing, vol. 5, no. 6, pp. 640–
2015. 652, 2006.
[9] R. Jafari, A. Encarnacao, A. Zahoory, F. Dabiri, H. Noshadi, and [27] H. Mahboubi, K. Moezzi, A. G. Aghdam, K. Sayrafian-Pour, and V. Mar-

T
M. Sarrafzadeh, “Wireless sensor networks for health monitoring,” in bukh, “Distributed deployment algorithms for improved coverage in a
The Second Annual International Conference on Mobile and Ubiquitous network of wireless mobile sensors,” IEEE Transactions on Industrial

IP
Systems: Networking and Services, 2005, pp. 479–481. Informatics, vol. 10, no. 1, pp. 163–174, 2014.
[10] D. S. Deif and Y. Gadallah, “Classification of wireless sensor networks [28] H. J. Lee, Y. h. Kim, Y. H. Han, and C. Y. Park, “Centroid-based move-
deployment techniques,” IEEE Communications Surveys & Tutorials, ment assisted sensor deployment schemes in wireless sensor networks,”
vol. 16, no. 2, pp. 834–855, 2014. in 2009 IEEE 70th Vehicular Technology Conference Fall (VTC 2009-

CR
[11] Y. Miao, Z. Sun, N. Wang, Y. Cao, and H. Cruickshank, “Time efficient Fall),. IEEE, 2009, pp. 1–5.
data collection with mobile sink and vmimo technique in wireless sensor [29] N. Heo and P. K. Varshney, “Energy-efficient deployment of intelligent
networks,” IEEE Systems Journal, vol. PP, no. 99, pp. 1–9, 2016. mobile sensor networks,” IEEE Transactions on systems, Man and
[12] Y. Cao, N. Wang, Z. Sun, and H. Cruickshank, “A reliable and Cybernetics, Part A: Systems and Humans, vol. 35, no. 1, pp. 78–92,
efficient encounter-based routing framework for delaydisruption tolerant 2005.
networks,” IEEE Sensors Journal, vol. 15, no. 7, pp. 4004–4018, 2015.
[13] K. Xiao, R. Wang, T. Fu, J. Li, and P. Deng, “Divide-and-conquer
architecture based collaborative sensing for target monitoring in
wireless sensor networks,” Information Fusion, vol. 36, pp. 162–171, US
2017. [Online]. Available: http://www.sciencedirect.com/science/article/
pii/S1566253516301816
[30] A. Boukerche and X. Fei, “A voronoi approach for coverage protocols
in wireless sensor networks,” in IEEE Global Telecommunications
Conference. IEEE, 2007, pp. 5190–5194.
[31] H. Mahboubi, M. Vaezi, and F. Labeau, “Mobile sensors deployment
subject to location estimation error,” IEEE Transactions on Vehicular
AN
Technology, vol. 66, no. 1, pp. 668–678, 2017.
[14] M. R. Senouci, A. Mellouk, K. Asnoune, and F. Y. Bouhidel,
“Movement-assisted sensor deployment algorithms: A survey and tax- [32] N. Bartolini, G. Bongiovanni, T. La Porta, S. Silvestri, and F. Vincenti,
onomy,” IEEE Communications Surveys & Tutorials, vol. 17, no. 4, pp. “Voronoi-based deployment of mobile sensors in the face of adversaries,”
2493–2510, 2015. in 2014 IEEE International Conference on Communications (ICC),
[15] M. C. Akewar and N. V. Thakur, “A study of wireless mobile sensor 2014, pp. 532–537.
network deployment,” Internatioal Journal of Computer and Wireless [33] C. Qiu and H. Shen, “A delaunay-based coordinate-free mechanism
M

Communication, vol. 2, no. 4, 2012. for full coverage in wireless sensor networks,” IEEE Transactions on
[16] O. Khatib, “Real-time obstacle avoidance for manipulators and mobile Parallel and Distributed Systems, vol. 25, no. 4, pp. 828–839, 2014.
robots,” The International Journal of Robotics Research, vol. 5, no. 1, [34] T.-W. Sung and C.-S. Yang, “Voronoi-based coverage improvement
pp. 90–98, 1986. approach for wireless directional sensor networks,” Journal of Network
[17] A. Howard, M. Matarić, and G. Sukhatme, “Mobile sensor network and Computer Applications, vol. 39, pp. 202–213, 2014.
ED

deployment using potential fields: A distributed, scalable solution to the [35] M. Abo-Zahhad, N. Sabor, S. Sasaki, and S. M. Ahmed, “A centralized
area coverage problem,” in Distributed Autonomous Robotic Systems 5, immune-voronoi deployment algorithm for coverage maximization and
H. Asama, T. Arai, T. Fukuda, and T. Hasegawa, Eds. Springer Japan, energy conservation in mobile wireless sensor networks,” Information
2002, pp. 299–308. Fusion, vol. 30, pp. 36–51, 2016.
[18] S. Li, C. Xu, W. Pan, and Y. Pan, “Sensor deployment optimization for [36] B. Wang, “Coverage problems in sensor networks: A survey,” ACM
PT

detecting maneuvering targets,” in 2005 8th International Conference on Computing Surveys (CSUR), vol. 43, no. 4, p. 32, 2011.
Information Fusion, vol. 2. IEEE, 2005, pp. 1629–1635.
[37] K. Römer and F. Mattern, “The design space of wireless sensor net-
[19] Y. Zou and K. Chakrabarty, “Sensor deployment and target localization
works,” IEEE Wireless Communications, vol. 11, no. 6, pp. 54–61, 2004.
based on virtual forces,” in Twenty-Second Annual Joint Conference of
the IEEE Computer and Communications, vol. 2. IEEE, 2003, pp. [38] G. Tan, S. Jarvis, and A.-M. Kermarrec, “Connectivity-guaranteed and
CE

1293–1303. obstacle-adaptive deployment schemes for mobile sensor networks,”


[20] F. Kribi, P. Minet, and A. Laouiti, “Redeploying mobile wireless sensor IEEE Transactions on Mobile Computing, vol. 8, no. 6, pp. 836–848,
networks with virtual forces,” in 2009 2nd IFIP Wireless Days, 2009, 2009.
pp. 1–6. [39] H. Zhang and J. C. Hou, “Maintaining sensing coverage and connectivity
[21] M. Garetto, M. Gribaudo, C. Chiasserini, and E. Leonardi, “A distributed in large sensor networks,” Ad Hoc & Sensor Wireless Networks, vol. 1,
no. 1-2, pp. 89–124, 2005.
AC

sensor relocatlon scheme for environmental control,” in IEEE Interna-


tional Conference on Mobile Adhoc and Sensor Systems, 2007, pp. 1–10. [40] G. Wang, G. Cao, P. Berman, and T. F. La Porta, “Bidding protocols for
[22] X. Yu, W. Huang, J. Lan, and X. Qian, “A novel virtual force approach deploying mobile sensors,” IEEE Transactions on Mobile Computing,
for node deployment in wireless sensor network,” in 2012 IEEE 8th vol. 6, no. 5, pp. 563–576, 2007.

You might also like