An optimal algorithm for coverage hole healing

An Optimal Algorithm for Coverage Hole Healing
in Hybrid Sensor Networks
Dung T. Nguyen, Nam P. Nguyen, My T. Thai, Abdelsalam Helal
Department of Computer and Information Science and Engineering, University of Florida
Email: {dtnguyen, nanguyen, mythai, helal}@cise.ufl.edu
Abstract—Network coverage is one of the most decisive factors
for determining the efficiency of a wireless sensor network.
However, in dangerous or hostile environments such as battle
fields or active volcano areas, we can neither deterministically
or purposely deploy sensors as desired, thus the emergence
of coverage holes (the unmonitored areas) is unavoidable. In
addition, the introduction of new coverage holes during network
operation due to sensor failures due to energy depletion shall
significantly reduce coverage efficacy. Therefore, we need to either
remotely control or set up a protocol to heal them as soon as
possible in an automated fashion.
In this paper, we focus on how to schedule mobile sensors
in order to cope with coverage hole issues in a hybrid sensor
network containing both static and mobile sensors. To this end,
we introduce a new metric, namely to maximize the minimum
remaining energy of all moved sensor since the more energy
remains, the longer the network can operate. Based on this
metric, we propose an efficient coverage healing algorithm that
always determines an optimal location for each mobile sensor in
order to heal all coverage holes, after all mobile sensors locations
and coverage holes are located. Simulation results confirm the
efficiency and utilization of our proposed method.
Index Terms—hybrid sensor network; coverage hole, move-
ment schedule, mobile sensor, coverage hole healing
I. INTRODUCTION
Wireless sensor networks (WSN) have been developing
quickly in the recent decade due to their low costs and wide
ranges of applications. Autonomous sensors can be deployed
in an area and incorporate to each other to fulfill a specific
task such as habitat monitoring, environment observation or
battle field surveillance [1]. A sensor can only sense and detect
events within its sensing range which usually forms a circle
centered at its location. The union of areas monitored by all
sensors is defined as the covered area of the whole network.
To guarantee the monitoring quality, it is crucial that every
single point in the target field must be within the network’s
covered area, or in other words, every point in the target field
must be monitored by at least one sensor.
The development of sensing coverage holes, i.e., unmoni-
tored areas, during network deployment and operation is one
of the most difficult problems to address. In many dangerous
and hostile regions such as battle fields or active volcano areas,
we can neither deterministically nor purposely deploy sensors,
we can only do so randomly. However, a notable problem with
random sensor deployment is the unbalance of sensor density
in the targeting area: the sensor density could be high in some
subareas but might be extremely low in some of the others,
thus coverage holes are more likely to emerge in those regions.
Therefore, the development of coverage holes is unavoidable.
Additionally, during network operation, some sensors may
accidentally stop working due to either energy depletion or
failure under some adversarial events. Therefore, it is crucial
to have a mechanism for not only setting up and maintaining
network coverage but also healing the sensing holes as soon
as they arise in an automated manner. This is an interesting,
yet changeling problem in WSN.
In this paper, we investigate the coverage healing problem
in a hybrid sensor network containing both static and mobile
sensors. Wang [2] proposed an economical deployment strat-
egy for this type of networks that uses only a small fraction
of mobile sensors to save many static sensors. In this type of
networks, mobile sensors can be used to heal coverage holes.
Assume that the locations of coverage holes and mobile sen-
sors are already detected, the coverage healing problem asks
for a movement scheduling that relocates all or some of the
mobile sensors to their appropriate locations in order to cover
all holes. Wang [2] designed an algorithm for this problem
with an objective of minimizing the total consumed energy
of mobile sensors. However, this objective has a shortcoming,
especially with sensors having the lowest remaining energy:
these sensors may unpredictably stop working any time due
to lack of energy. This observation drives the need for a better
sensor scheduling to heal coverage holes.
To overcome this limitation, we present a new metric to
evaluate a sensor movement scheduling: maximize the mini-
mum remaining energy over all moving sensors. In particular,
we introduce an algorithm which not only provides the optimal
solution but also very efficient and easy to implement in
practice. In a big picture, our proposed algorithm first tries
to attain a possible optimal value of remaining energy and
then, based on the gained information, schedules an optimal
movement which achieves this value. In some cases, if the
number of failed sensors are small and consequently, there
are not too many coverage holes to heal, computing the
movement schedule of the whole network is of high costs and
unnecessary. Therefore, we introduce a simple and efficient
distributed scheme to deal with these special cases. This
scheme not only requires low computational costs but also
quickly locates the appropriate mobile sensors to heal a new
hole as soon as its location is detected.
Our main contributions in this paper are threefold. Firstly,
we propose an exact O(
√
nm log m) algorithm that provides
an optimal solution for coverage healing problem on a hybrid
sensor network, where n and m are the numbers of nodes and
edges in the underlying bipartite graph. Secondly, we intro-
duce a distributed heuristic scheme to quickly and efficiently
978-1-4577-9538-2/11/$26.00 ©2011 IEEE 494
Authorized licensed use limited to: IEEE Xplore. Downloaded on October 18,2011 at 21:38:36 UTC from IEEE Xplore. Restrictions apply.

Head node
Inactive mobile
sensor
Active mobile
sensor
Coverage hole
Direct movement
Cascaded
movement
Fig. 1. Network model
heal coverage holes when not too many of them show up.
This heuristic scheme can identify a specific mobile sensor
to instantly heal any newly introduced coverage hole with
very low message and computational complexity. Thirdly, we
extensively test our algorithm on various networks against the
one suggested by Wang [2]. Experimental results show that our
new metric, maximizing the lowest remaining energy, is more
suitable in WSN than the total consumed energy in [2]. In
addition, this distributed scheme usually provides near optimal
values in most networks.
The rest of the paper is organized as follows: Section II de-
scribes the network models and problem statement. Section III
presents our algorithm for computing the optimal movement
scheduling. A distributed scheme to deal with some special
cases is introduced in the Section IV. In section V, we show
experiment results. Section VI briefly summarizes the related
work, and finally Section VII concludes our work.
II. NETWORK MODEL AND PROBLEM FORMULATION
A. Network model
A hybrid sensor network consists of both static and mobile
sensors working together in a targeted field. Each sensor is
powered by a different amount of energy and has its own com-
munication as well as sensing range. When a mobile sensor
moves to a new location, it consumes energy proportionally
to the distance traveled. In our model, each sensor is able to
estimate its remaining energy thanks to the method suggested
in [3] and identify its current location by the equipped GPS
device [4].
There are two types of mobile sensors during network oper-
ation: active and inactive mobile sensors. Active mobile sen-
sors are those actually working in the field and in cooperation
with the other sensors to monitor the target area. In contrast,
inactive ones are mobile sensors in the sleeping mode. In
addition, there is a base station equipped with rich computing
resources to manage the network and exchange information
with the remote controller. Sensors gather monitored data and
other network information and send them back to the base
station for processing. The base station then sends back control
messages to sensors.
For the ease of management, the target area is divided
into autonomous subareas (called cells) of the same size as
illustrated in Fig. 1. Each subareas is controlled by a head
node which manages activities within its cell. In fact, a head
node is a regular designated sensor and any sensor in the cell
can take the role of the head node. In our model, sensors
within a cell will take turn to be the head node. In particular,
a head node knows the following information within its area
(1) Static and mobile sensor locations and (2) Coverage hole
locations. Furthermore, for each mobile sensor, the following
information is made available to the head node: its remaining
energy and working status (active or inactive).
The size of a coverage hole in our model can be estimated
using the proposed method in [5]. This implies the number
of mobile sensors for healing the hole can be estimated
approximately from the size of the hole and their sensing
ranges. In this paper, we assume that any newly introduced
coverage hole can be covered by a single mobile sensor. This
assumption makes sense since any large coverage hole can
be decomposed as a collection of smaller ones, in which each
hole can be fully monitored by a sensor. When a sensor detects
a hole near it, it sends information to the head node. Thus,
the information of coverage holes in a grid is made available
to the head node as well as the base station.
B. Coverage Hole Healing Problem
After the network is deployed, two important tasks are main-
taining its coverage and healing coverage holes as soon as they
arise. Generally, one would expect that the network will work
stably and last as long as possible. This means the rearranged
mobile sensors, after the hole healing process, should have as
much remaining energy as possible. However, since sensing
holes can simultaneously and unpredictably appear at different
regions in the targeting area, how can we quickly find an
optimal movement scheduling for mobile sensors satisfying the
above network requirements? Informally, the Coverage Hole
Healing Problem asks for a sensor scheduling that not only
heals all introduced coverage holes but also maximizes the
lowest remaining energy over all moving sensors, assuming
the knowledge of holes’ locations.
Cascaded Movement: There are two possible scenarios
when a mobile sensor is ordered to move to a new location.
An inactive one simply moves directly to the desired location
whereas an active one may require an extra consideration: any
active mobile sensor immediately creates a new hole behind
itself as soon as it moves to a new location. Clearly, that
hole needs to be covered by another mobile sensor and the
movement of that sensor may further lead to another uncovered
hole as a result. These actions reassemble the concept of a
cascaded movement of mobile sensors on WSN.
Cascaded movement represents an opportunity in our ap-
proach since allowing this will help in significantly reducing
the consumed energy of each sensor, thus enhancing its
remaining energy. To give a sense of its effect, let us consider
the example illustrated in Fig. 2(a). In this example, we have
two possibilities. In the first option, we move inactive sensor
S1 directly to the hole. In the other option, we can move S1
to S2’s location, S2 to S3’s location and then move S3 to
heal the hole. After all sensors move, it is obvious that the
remaining energy of S1 in the first option may be less than
the minimum energy of S1, S2 and S3 in the second option,
since S1 has traveled a much longer distance and thus, has
consumed much more energy than S2 or S3 has. In other
words, the second option provides a better solution than the
495

first one. A big picture of the whole movement schedule is
illustrated in Fig. 1.
The Coverage Hole Healing Problem (CHP) can be formally
described as follows: Given a hybrid sensor network with a
set M of mobile sensors, a set S of static sensors and a
set H of locations of coverage holes, find a scheduling for
mobile sensors, allowing cascaded movement, to quickly and
efficiently heal the holes in H such that the lowest remaining
energy over all moving sensors is maximized.
III. COVERAGE HOLE HEALING ALGORITHM
In this section, we present our optimal solution for CHP
based on the graph theory approach. Basically, our solution
consists of two main steps: 1) construct a bipartite graph
G representing the relationship between mobile sensors and
coverage holes. 2) Find a complete matching T of G that
maximize the lightest weight in T. Based on T, mobile sensors
will move to correct location to heal the coverage holes.
A. Building bipartite graph
Based on the gathered information at the base station, we
construct a bipartite graph G = ((M, H), E) describing the
network and prove that CHP is, in fact, equivalent to finding a
complete matching (in terms of holes) on G. We first construct
G with directed movement scheduling using only inactive
mobile sensors and then, extend G to capture schedules that
allow cascaded movements.
To capture direct movements, G is constructed as follows:
Let M and H be sets of nodes representing inactive mobile
sensors and coverage holes, respectively. For any pair of a
mobile sensor Mi and a hole Hj, an edge eij is added to
the edge set E if Mi has enough energy to move to Hj.
The weight w(eij) of this edge is the remaining energy of
Mi after moving to Hj. A scheduling that orders mobile
sensors M1, M2, . . . , Mh healing holes H1, H2, . . . , Hh can
be represented by a complete matching {e11, e22, . . . , ehh} on
G, where h = |H|. By constructing this way, the minimum
energy remains after a scheduling is indeed the weight of the
lightest edge in the matching.
Now we modify G to allow cascaded movements by intro-
ducing a dummy hole dh for each working sensors ws and add
dh to H as well as ws to M. The weights of edges connecting
a dummy hole to other mobile sensors are computed the same
as others, except for the edge connecting dh to ws whose
weight is infinite. This modification transforms the problem
back to the case of direct movements.
Lemma 1: The CHP problem is equivalent to finding a
complete matching on G (in terms of coverage holes) that
maximizes the weight of the lightest edge.
Proof: We will prove that if the lowest remaining energy
is maximized in CHP (say its value is Rmin) then there exists
an optimal complete matching on G whose the lightest edge
weight is exactly Rmin, and vice versa.
Let MS be an optimal movement scheduling in CHP.
Denote by Rmin the minimum remaining energy over all
moving sensors. We will construct a complete matching T
on G whose lightest edge weight is Rmin. Initialize T = ∅.
For each hole Hj, if it can be covered by a sensor Mi, we add
S1
S3
S2
hole
S1
S3
S2
hole
(a) (b)
Fig. 2. 2(a): Cascaded movement. 2(b): An optimal matching (shown in thick
edges) with mobile sensors on the left and holes on the right. The lowest edge
weight Rmin is shown in red. The removals of edges whose weights are less
than Rmin (shown in dotted lines) will not affect the optimal solution
edge eij into T. According to the definition of edge weight,
the weight of the lightest edge in T is Rmin. Now, there are
only dummy nodes in H that are not covered by any moved
sensor. For each dh, we add an edge connecting dh to its
corresponding inactive mobile sensor ws into T; these edges
have weight infinite. Since each hole is covered by one sensor,
it implies that T is a complete matching on G with the weight
of lightest edge is Rmin.
On the other hand, if there exists a complete matching T
whose lightest edge weight is Rmin, then for each edge eij
in T, if Hj is not the dummy hole created by Mi, move Mi
to heal Hj. Since T is a complete matching, each hole is
healed by exactly one sensor and the minimum energy over
all moving sensors is Rmin, which completes the proof.
B. Scheduling algorithm
The key idea of our approach is that, if Rmin is the weight
of the lightest edge in the optimal matching T, then the
removals of edges whose weights are less than Rmin will
not affect T. Therefore, we need to identify Rmin and find
the maximum matching on G after removing all edges whose
weights less than Rmin. In the next part, we show the optimal
condition of Rmin and describe the algorithm in detail.
Given a value R we denote G(R)−
as the subgraph of G
after removing all edges whose weights are strictly less than
R. We also define G(R) be the subgraph of G when all edges
of weights at most R are removed. Let M(G) denote the size
of the maximum matching on the bipartite graph G. Suppose
that Rmin is the optimal value corresponding to the matching
M, then removing edges whose weights are less than Rmin
does not effect M. But if all edges of weights at most Rmin
are removed, at least one edge in M is removed, i.e. M is no
longer a valid matching. Based on this property, the following
lemma states the optimality condition for Rmin.
Lemma 2: Rmin is the optimal value iff the size of
the maximum matching on M(G(Rmin)−
) = |H| and
M(G(Rmin)) < |H|.
Proof: Suppose that Rmin is the optimal value, then
there exists a complete matching T whose edges have weight
at least Rmin (by Lemma 1). After the removals of edges
whose weights are less than Rmin, the edges of T still
remain in G(Rmin)−
. Thus the size of a maximum match-
ing on G(Rmin) is at least the size of T. In addition,
M(G(Rmin)) < |H| since if M(G(Rmin)) = |H| then
there exists a matching of size H such as every edge has
496

weight greater than Rmin (a contradiction). On the other hand,
if M(G(Rmin)−
) = |H| and M(G(Rmin)) < |H| then
Rmin is optimal. First M(G(Rmin)−
) = |H|, there exists
a complete matching whose edges are of weights at least
Rmin. Second, there is no complete matching whose edges
have weights greater than Rmin. Otherwise, there exists a such
complete matching, let R be the weight of its lightest edge.
Removing all edges of weight less than R does not affect this
matching, then it is reserved in G(Rmin), i.e. G(Rmin) has a
maximum matching of size |H| (a contradiction).
Corollary 1: Given a value R, if the size of maximum
matching on G(R)−
is less than |H| then Rmin < R,
otherwise Rmin ≥ R
Proof: We prove this by contradiction. Suppose that the
size of maximum matching on G(R)−
is less than |H| but
Rmin ≥ R. We have G(Rmin)−
is the subgraph of G(R)−
.
It implies that M(G(Rmin)−
) ≤ M(G(R)−
) ≤ |H|, i.e.,
the optimal condition of Rmin is violated. On the other hand,
suppose that the size of maximum matching on G(R)−
equals
|H| but Rmin < R. Then G(R)−
is the subgraph of G(Rmin).
It means that M(G(Rmin)) ≥ M(G(R)−
) = |H| i.e the
optimal condition of Rmin is also violated.
Based on the above corollary, we design an effective proce-
dure to find the optimal value Rmin among all edge weights
by iteratively shrinking its value range. Initially, the range of
Rmin is between the maximum weight and minimum weight.
Consider the value R in the middle of the range, we check
on the size of the maximum matching on G(R)−
. If it is less
than |H|, then Rmin must belong to the lower haft of the
range. Otherwise, Rmin belongs to the upper haft. Due this
property, we use binary search approach to find the value of
Rmin. In our approach, we use Hopcroft-Karp algorithm [6]
as a subroutine for finding a maximum matching on G(R)−
.
The whole algorithm is described in Algorithm 1.
Algorithm 1 Maximized Minimum-Edge-Weight Matching
Input: A bipartite graph G = (M, H, E)
Output: A maximized minimum-edge-weight matching.
max ← size of the maximum matching
Sort list of edge E in increasing order of weights.
low ← 1; high ← |E|;
while 1 + low < high do
mid ← low + (high − low)/2
s ← |the maximum matching with weights ≥ E[mid]|
if s < max then
high ← mid
else
low ← mid
end if
end while
return The min-cost maximum matching whose edge weighs ≥ E[low]
Theorem 1: The matching provided by the Algorithm 1 is a
maximum one that maximizes the minimum remaining energy
over all moving sensors. In addition, the running time of the
algorithm is O(
√
nm log m) where n = |M∪H| and m = |E|.
Proof: In each while loop, since mid > low, the dif-
ference high − low is reduced at least by max{1, (high −
low)/2 } after each iteration. Thus the number of while
loops is O(log m). In addition, the following inequalities
always hold for the size of the maximum matching on
G(E[low]) and G(E[high]): M(G(E[low])) = |H| and and
NW
SW
SE
NE
(i,j+1)
(i,j)(i-1,j) (i+1,j)
(i+1,j-1)
(i+1,j+1)(i-1,j+1)
(i-1,j-1) (i-1,j)
(i,j+1)
(i,j) (i+1,j)
(i+1,j+1)
NE area of cell (i,j)
Fig. 3. Four candidate sensors of each cell
M(G(E[high])) < |H|. When the algorithm terminates,
low + 1 = high, (G(E[low])+
and (G(E[high]) are the
same. From Lemma 2, E[low] satisfies the optimal condition
of Rmin. Thus the produced matching maximizes the sensor’s
lowest remaining energy.
There is O(log m) loop, each loop calls Hopcroft-Karp
algorithm only one time. Since the running time of Hopcroft-
Karp algorithm is O(
√
nm), the running time of our algorithm
is O(
√
nm log m).
IV. DISTRIBUTED SCHEME
In some circumstances where the number of failed sensors
is small and consequently, the number of new coverage holes
is small, Algorithm 1 still works perfectly but it may require
much more time and computing resources than what is actually
needed. Therefore, we introduce a distributed protocol that
can efficiently identify the appropriate mobile sensor(s) to
quickly heal the newly introduced hole while keeping both
the computational and communication costs low. Due to space
limit, we only sketch the main idea and exclude the detailed
description of message protocol among sensors.
The main idea is that each head of a cell node maintains a
list of best candidates which contains mobile sensors having
the maximum remaining energy as they move toward the cell.
Generally, when a hole appears in a cell, the corresponding
head node contacts its best candidate in order to heal the
hole. However, when many mobile sensors are ordered to heal
coverage holes simultaneously, the candidate lists stored at the
corresponding head nodes may be out-of-date. Therefore, each
head node must be able to update its candidate list as quick as
possible. The update protocol should be fast, light weight and
is only executed when a mobile sensor is ordered to move.
We observe that the best mobile sensor for a cell is usually
the best one for its neighbors. This is a very important
observation since it helps us to design an efficient protocol that
notifies the neighboring head nodes to update their candidate
lists once a mobile sensor in a cell moves. In this case, the
head node of the moved sensor first queries the candidate
lists of its eight neighboring head nodes and then, it picks
the best sensors among these lists to be its new candidates.
After updating its own list, this head node will notify all of its
neighbors about this change and some of them may eventually
want to update as well. In particular, each head node stores the
locations of four best mobile sensors, namely SE, SW, NW
and NE which are the best sensors in the SE, SW, NW and
NE regions. As illustrated in Fig. 3, the NE region is shaded.
When cell (i, j) wants to update its NE sensor, it queries
497

the candidate lists of three cells (i + 1, j), (i + 1, j + 1) and
(i, j + 1) and if NE candidate is actually updated, the head
node of cell (i, j) sends notifications to head nodes of cells
(i − 1, j), (i − 1, j − 1) and (i, j − 1) so that they can decide
whether they should update their NE candidates or not. The
updating process terminates when no head nodes decide to
change or update their candidate lists.
V. PERFORMANCE EVALUATION
This section consists of three parts: First, we evaluate the
new metric utilization. Next, we show the advantage of the
scheduling with cascaded movement and finally, we certify
the efficiency of our distributed scheme in the last part.
A. The utilization of the new objective function
To evaluate the efficiency of our proposed optimization
function, we compare it to the current best solution suggested
for hybrid sensor networks [2]. The testbed is set up as
follows: we consider a network deployed in an area of different
sizes 100m × 100m, 200m × 200m and 400m × 400m. All
networks contain 20 randomly located coverage holes. We
randomly deploy various numbers mobile sensors, each sensor
of which is uniformly initialized a random amount of energy
ranging from 2500J to 3000J. We allow each mobile sensor
to consume about 30J per meter [7].
In each configuration of the area size and mobile sensors,
we move mobile sensors according to two optimal schedules
(1) Maximizing the minimum remaining energy (Ours) and
(2) Minimizing the total movement distance (Wang’s [2]). By
the time the sensors with minimum energy in [2] run out of
energy, we again execute Wang’s method to heal all holes
created by these sensors.
As depicted in Fig. 4, our approach most of the time
produces very competitive, if not to say better, results in
comparison with the current best method. In particular, our
total movement costs are always lower than those of Wang’s
in all test cases (top figures), except for some unusual cases
where the density of mobile sensors is extremely high (e.g.,
when number of sensors is 300 in the top left figure). In
particular, our results are around 3% to 9% better than that
of Wang’s. These results confirm the utilization strength of
our proposed metric.
Competitions on the network coverage (Fig. 4) reveal that
our algorithm, again, achieves competitive coverage percent-
age with its competitor, especially in the first test case where
both cover 100% of the network (left down figure). In the other
test cases, our method is a little bit off when the number of
sensors is small (e.g., less than 60 sensors); however, in a long
run, ours is better than Wang’s method with 2% to 5% and
2% to 3% more of the target field covered as shown in middle
and right down figures, respectively. One of the key reasons is
that in Wang’s method, mobile sensors are moved regardless
of their remaining energy, thereby increasing the posiblity
of re-executing the movement schedule more frequently. It
implies that during the re-execution, coverage holes are left
unhealed. Supported by these experimental results, we believe
that our new optimization metric is more efficient and robust
in comparison to the current best one proposed in [2].
20
40
60
80
100
120
50 100 150 200 250
TotalMovementCost(m)
Minimized Total Movement Cost
Maximized Remain Energy
50
100
150
200
250
300
50 100 150 200 250
100
150
200
250
300
350
50 100 150 200 250
90
95
100
105
110
50 100 150 200 250
PercentageofFullCoverage(%)
80
85
90
95
100
105
110
50 100 150 200 250
Number of Mobile Sensors
60
70
80
90
100
110
50 100 150 200 250
Fig. 4. Two optimal movement schedules with different objectives: max-
imized minimum remained energy and minimized total movement cost.
Two left figures are of size 100m × 100m. Middle figures are of size
200m × 200m. Two right figures are of size 400m × 400m.
B. The advantage of cascaded movement
We use the same testbed as above with the appearance of
coverage holes. Each network contains 20 randomly located
coverage holes and 50 randomly placed inactive mobile sen-
sors. To allow the cascaded movement, we also deploy active
mobile sensors which are randomly assigned ranging from
1500J to 3000J [7]. We run 100 times and compute the
average of the lowest remained energy overall moved sensors
with two options: with and without cascaded movement.
The dominance of mobile sensor scheduling with over
without cascaded movement is shown in Fig. 5. As indicated in
this figure, the allowance of cascaded movement significantly
helps the mobile sensors to attain much more remaining
energy, especially when the number of sensors increases. In
particular, a scheduling with cascaded movement helps sensors
save up to 7%, 21% and 71% more energy in the hole healing
process, as depicted in the top left, top middle and top right
figure, respectively. The reason behind this superiority is due
to the lesser distance each sensors has to travel in order to
heal a coverage hole. These results promise the advantage
of a scheduling with cascaded movement in a hybrid sensor
network.
Experiments on network coverage, again, confirm the ef-
fectiveness of a scheduling with cascaded movement. As one
can observe from the results depicted in the left, middle and
right down figures, the percentage of the targeting area covered
by mobile sensors with cascaded movements increases signif-
icantly and tends to double that of non-cascaded movement
when the number of deployed sensor increases. These results
prove the strength of our proposed algorithm with cascaded
movement.
C. The efficiency of the distributed scheme
Finally, we test the distributed scheme when the network
has only one coverage hole and compare the results with the
optimal solutions. We use the same set up as above with a
slight modification: the size of a cell is 10m × 10m, which
implies a 400m × 400m-field will have 1600 cells. A set of
mobile sensors are randomly located in the field of target. We
498

1900
1950
2000
2050
2100
2150
2200
0 20 40 60 80 100
MinimumRemainedEnergy(J)
Direct Movement
Cascaded Movement
1300
1400
1500
1600
1700
1800
0 20 40 60 80 100
Direct Movement
Cascaded Movement
700
800
900
1000
1100
1200
1300
0 20 40 60 80 100
Direct Movement
Cascaded Movement
90
95
100
105
110
0 20 40 60 80 100
PercentageofFullCoverage(%)
Direct Movement
Cascaded Movement
90
95
100
105
110
0 20 40 60 80 100
Number of Active Mobile Sensors
Direct Movement
Cascaded Movement
50
55
60
65
70
0 20 40 60 80 100
Direct Movement
Cascaded Movement
Fig. 5. Minimum remained energy of moved sensors in cascaded movement
and direct movement. Two left figures are of size 100m × 100m. Middle
figures are of size 200m×200m. Two right figures are of size 400m×400m.
2300
2400
2500
2600
2700
2800
2900
3000
50 100 150 200 250
RemainedEnergy(J)
Optimal Remained Energy
Distributed Scheme
2300
2400
2500
2600
2700
2800
2900
3000
50 100 150 200 250
Number of Mobile Sensors
Distributed Scheme
2300
2400
2500
2600
2700
2800
2900
3000
50 100 150 200 250
Distributed Scheme
Fig. 6. The near optimality of the Distributed Scheme. Left figure is of size
100m × 100m. Middle figure is of size 200m × 200m. Right figure is of
size 400m ∗ 400m.
generate 100 networks, each of them contains 1000 coverage
holes randomly deployed in the targeting area.
As shown in Fig. 6, the energy differences between the
our heuristic results and the optimal solutions are quite small:
the remaining energy of our distributed algorithm covers up to
95%, 96% and 98% of the energy produced by the optimal so-
lution. Moreover, our results tend to closely approximate those
of the optimal methods as the number of sensors increases.
This result indicates that when only a small number of holes
develops, our distributed algorithm efficiently computes a
scheduling that relatively preserves as much remaining energy
as the optimal strategy does.
VI. RELATED WORK
On hybrid sensor networks, Wang et al.[8] proposed a
bidding protocol between static and mobile sensors to guide
the movement of mobile sensors. Static sensors detect the
coverage holes and then send bidding messages to mobile
ones. After several rounds of exchanging information, a mobile
sensor moves to the nearest matched hole. This protocol is time
consuming and computationally expensive. In another work
[2], Wang et al. designed an optimal algorithm to move mobile
sensors such that the total movement distance is minimized.
However, this approach has the limitation as we discussed in
Section I, which gives rise to our metric. Using a somehow
similar metric with ours, Czyzowicz et al. [9] proposed an
O(n2
) optimal algorithm that maps mobile sensors in a line
segment. Because in the line segment, mobile sensors can only
choose two directions to move, the solution is not extended to
wireless sensor networks that monitor a square area.
Instead of healing several coverage holes simultaneously,
there exists some studies focused on healing one single hole
[10], [11], [12] (and references therein). The notable common
limitations of these methods are two fold: 1) high message and
computational complexity and 2) it is not easy to extend these
solutions for multiple coverage holes healing as addressed in
our paper.
VII. CONCLUSION
In this paper, we study the coverage healing problem with
the different metric on hybrid sensor networks where both
static and mobile sensors are utilized. We proposed a new
metric to maximize the minimum remaining energy over all
moved sensors. Based on this new metric, we proposed a graph
theory based approach for the coverage healing problem which
always returns an optimal solution according to this metric.
We also presented a near-optimal distributed algorithm for a
special case when the number of newly introduced coverage
holes is small. Experimental results show that our new metric
helps the WSN work much more stabler and last longer in
comparison to the total consumed energy metric in [2].
ACKNOWLEDGMENT
This work is partially supported by NSF CAREER 0953284
and DTRA HDTRA1-08-10.
REFERENCES
[1] N. Xu. A survey of sensor network applications. IEEE Communications
Magazine, 40(8):102–114, 2002.
[2] Wei Wang, Vikram Srinivasan, and Kee-Chaing Chua. Trade-offs
between mobility and density for coverage in wireless sensor networks.
MOBICOM, pages 39–50, 2007.
[3] R.A.F. Mini, M. Val Machado, A.A.F. Loureiro, and B. Nath. Prediction-
based energy map for wireless sensor networks. Ad Hoc Networks,
3(2):235–253, 2005.
[4] N. Bulusu, J. Heidemann, and D. Estrin. Gps-less low-cost outdoor
localization for very small devices. Personal Communications, IEEE,
7(5):28 –34, October 2000.
[5] Amitabha Ghosh. Estimating coverage holes and enhancing coverage
in mixed sensor networks. In Proceedings of the 29th Annual IEEE
International Conference on Local Computer Networks, LCN ’04, pages
68–76, Washington, DC, USA, 2004. IEEE Computer Society.
[6] John E. Hopcroft and Richard M. Karp. A n5/2 algorithm for maximum
matchings in bipartite. In Switching and Automata Theory, 1971., 12th
Annual Symposium on, pages 122 –125, oct. 1971.
[7] Gabe Sibley and Mohammad H. Rahimi andGaurav S. Sukhatme.
Robomote: A tiny mobile robot platform for large-scale ad-hoc sensor
networks. In ICRA, pages 1143–1148, 2002.
[8] Guiling Wang, Guohong Cao, and T. LaPorta. A bidding protocol for
deploying mobile sensors. In Network Protocols, 2003. Proceedings.
11th IEEE International Conference on, pages 315 – 324, 2003.
[9] J. Czyzowicz, E. Kranakis, D. Krizanc, I. Lambadaris, L. Narayanan,
J. Opatrny, L. Stacho, J. Urrutia, and M. Yazdani. On minimizing the
maximum sensor movement for barrier coverage of a line segment. In
ADHOC-NOW ’09: Proceedings of the 8th International Conference
on Ad-Hoc, Mobile and Wireless Networks, pages 194–212. Springer-
Verlag, 2009.
[10] Guiling Wang, Guohong Cao, T. La Porta, and Wensheng Zhang. Sensor
relocation in mobile sensor networks. In INFOCOM 2005. 24th Annual
Joint Conference of the IEEE Computer and Communications Societies.
Proceedings IEEE, volume 4, pages 2302 –2312 vol. 4, march 2005.
[11] Xu Li, Nicola Santoro, and Ivan Stojmenovic. Mesh-based sensor
relocation for coverage maintenance in mobile sensor networks. In
Ubiquitous Intelligence and Computing, Lecture Notes in Computer
Science. Springer Berlin / Heidelberg, 2007.
[12] Xu Li and N. Santoro. Zoner: A zone-based sensor relocation protocol
for mobile sensor networks. Local Computer Networks, Proceedings
2006 31st IEEE Conference on, 2006.
499

An optimal algorithm for coverage hole healing

More Related Content

What's hot (13)

Similar to An optimal algorithm for coverage hole healing (20)

Recently uploaded (20)

An optimal algorithm for coverage hole healing