Survey of Deployment Algorithms in Wireless Sensor

Networks: Coverage and Connectivity Issues and

Challenges
Ines Khoufi, Pascale Minet, Anis Laouiti, Saoucene Mahfoudh

To cite this version:

Ines Khoufi, Pascale Minet, Anis Laouiti, Saoucene Mahfoudh. Survey of Deployment Algorithms in
Wireless Sensor Networks: Coverage and Connectivity Issues and Challenges. International journal of
autonomous and adaptive communications systems, Inderscience Publishers, 2017, 10 (4), pp.341-390.
�10.1504/IJAACS.2017.10009671�. �hal-01095749�

HAL Id: hal-01095749

https://hal.inria.fr/hal-01095749
Submitted on 16 Dec 2014

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Int. J. Autonomous and Adaptive Communications Systems, Vol. x, No. x, 20xx 1

Survey of Deployment Algorithms in Wireless Sensor

Networks: Coverage and Connectivity Issues and
Challenges

Ines Khoufi
INRIA, Rocquencourt,
78153 Le Chesnay Cedex, France,
Email: [email protected]

Pascale Minet
INRIA, Rocquencourt,
78153 Le Chesnay Cedex, France,
[email protected]

Anis Laouiti
TELECOM SudParis,
CNRS Samovar UMR 5157,
91011 Evry Cedex, France,
Email: [email protected]

Saoucene Mahfoudh
INRIA, Rocquencourt,
78153 Le Chesnay Cedex, France,
[email protected]

Abstract: Wireless Sensor Networks (WSNs) have many fields of application, including industrial,
environmental, military, health and home domains. Monitoring a given zone is one of the main
goals of this technology. This consists in deploying sensor nodes in order to detect any event
occurring in the zone of interest considered and report this event to the sink. The monitoring task
can vary depending on the application domain concerned. In the industrial domain, the fast and
easy deployment of wireless sensor nodes allows a better monitoring of the area of interest in
temporary worksites. This deployment must be able to cope with obstacles and be energy efficient
in order to maximize the network lifetime. If the deployment is made after a disaster, it will operate
in an unfriendly environment that is discovered dynamically. We present a survey that focuses on
two major issues in WSNs: coverage and connectivity. We motivate our study by giving different
use cases corresponding to different coverage, connectivity, latency and robustness requirements of
the applications considered. We present a general and detailed analysis of deployment problems,
while highlighting the impacting factors, the common assumptions and models adopted in the
literature, as well as performance criteria for evaluation purposes. Different deployment algorithms
for area, barrier, and points of interest are studied and classified according to their characteristics and
properties. Several recapitulative tables illustrate and summarize our study. The designer in charge of
setting up such a network will find some useful recommendations, as well as some pitfalls to avoid.
Before concluding, we look at current trends and discuss some open issues.

Keywords: area coverage, barrier coverage, coverage, deployment algorithms, full connectivity, grid,
intermittent connectivity, node activity scheduling, point of interest coverage, sensor deployment,
virtual forces, wireless sensors, WSN
Biographical notes: Ines Khoufi received her Computer Science Engineering and Master degrees
from the National College of Computer Science (ENSI) in 2010 and 2011 respectively. Currently,
she is a Ph.D. student in HIPERCOM2 research team at Inria. Her current research interest is on
wireless sensor networks deployment and progressive discovery of an unfriendly environment.

Pascale MINET works at the Inria research center of Rocquencourt, near Versailles. She is head of the
HIPERCOM (High Performance Communication) team. She got her qualification in advising PhD
students in 1998 from the University of Versailles. Previously she got her PhD diploma in Computer
Science, the University of Toulouse and her Engineer diploma in Computer Science in 1980 from

Copyright © 20xx Inderscience Enterprises Ltd.

2 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

ENSEEIHT (Engineering school of Toulouse). Her research topics relate to wireless sensor networks
and mobile ad hoc networks and more particularly energy efficiency, routing, node activity
scheduling, multichannel communication, redeployment and quality of service in these networks.
She is co-author of the OLSR routing protocol standardized at IETF.

Anis Laouiti is an associate professor at Telecom SudParis since 2006. Before, he did his Phd
research work and worked as a research engineer within Hipercom team at Inria-Rocquencourt
where he participated to the OLSR routing protocol design (RFC3626). His research covers different
aspects in wireless ad hoc and mesh networks including protocol design, performance evaluation
and implementation testbed.

Saoucene MAHFOUDH received her Computer Science Engineering degree from ENSI in 2005
and her Master degree from the University of Paris 6 in 2006. She obtained her Ph D diploma in
2010 from the University of Paris 6. Her research topics deal with energy efficiency, cross-layering,
routing and redeployment in wireless sensor networks. After a post-doctoral fellow at Inria, she has
been working at King AbdelAziz University since 2013.

1 Introduction Depending on the size of the entity (area, barrier or point

of interest) monitored, a multi-hop network may need to be
deployed to enable the monitoring of this area as well as
With the emergence of the Internet of Things (IoT),
the delivery of the collected data. To meet the application
several billion electronic devices and machines will be
requirements, the deployment of these communicating objects
able to be connected to one another via the Internet.
(e.g. sensor nodes) must ensure coverage and connectivity
The devices will be able to communicate without the
properties. Roughly speaking, coverage refers to the ability
intervention of humans. Such communication is called
to detect events occurring in the entity monitored (e.g. area,
Machine-to-Machine communication, denoted M2M, and
barrier, point of interest) whereas connectivity refers to the
mobile wireless technology is an ideal technology to support
ability to report this event to the sink.
M2M communication. These emerging technologies based
on wireless communication can be autonomous and cope
with changes without any human help. As an example, the 1.1 Motivation
fast and easy deployment of mobile wireless sensors in a
temporary worksite is very useful for maintenance purposes There are several types of coverage and connectivity problems
and quality insurance. As these wireless sensor nodes are in WSNs.
battery equipped, the deployment should be energy efficient With regard to the coverage of the entity monitored
in order to maximize network lifetime. Sensors can also be we distinguish between area, barrier and point of interest
used for damage assessment after a disaster, when the network coverage. Furthermore, the coverage can be full (i.e. any
infrastructure has been damaged. In this case, the Wireless point of the entity is covered) or partial, depending on the
Sensor Network (WSN) operates in an unfriendly environment application requirements. If the coverage is full, any point can
that can only be discovered progressively. However, some be monitored by a single sensor node (i.e. simple coverage) or
major challenges need to be tackled, such as: by several sensor nodes (i.e. multiple coverage), depending on
the degree of robustness required. If the application requires
short delays to detect an event, any point must be permanently
• How to deal with the large volume of data produced. covered. In other cases, any point is temporarily covered.
Collecting and exploiting these large data flows is With regard to connectivity, if the application requires
commonly known as big data transfer and processing. short delays to report the event detected to the sink, permanent
connectivity is needed. Otherwise, intermittent connectivity
is sufficient. An example is given by a mobile robot collecting
• How to organize communication. Communicating data from disconnected islands of sensor nodes and, more
objects will exist everywhere. They are generally used generally, delay tolerant networks taking advantage of the
to monitor a phenomenon. This phenomenon can be mobility of data mules such as robots or people, etc.. If the
monitored by a single object (e.g. the temperature of application requires a high degree of robustness, multiple
coffee in a mug) or require collaboration between several paths toward the sink are needed. Otherwise, a single path is
objects, which are usually deployed in a predefined area. sufficient.
In the latter case, these objects need to communicate and As concerns the quality of data gathered, the application
organize themselves in the form of a network. That is the specifies its requirements for the time and space consistency
focus of this paper. of the data gathered. If both are required, a regular and
uniform deployment is needed.
 Coverage and Connectivity Issues and Challenges in WSN 3

Usually, an initial deployment is provided. It can be UC4 monitoring of endangered wild species at some water
random (e.g. sensor nodes are dropped from a helicopter), all points: the idea is to compute statistics about the number
sensor nodes can be grouped together at an entry point, or they of individuals of this species from the number of
can form disconnected groups, each group consisting of a set individuals visiting the water point. A full or partial belt
of connected sensor nodes, etc. However, such a deployment of sensor nodes is built along the water point depending
usually fails to ensure the coverage and connectivity properties on its size. Intermittent connectivity is usually sufficient.
required by the application. For instance, some regions can
be highly covered whereas others are poorly covered and UC5 detection of intruders crossing a barrier (e.g. the border
may contain some coverage holes that are not monitored. of a country, a door or windows in an apartment). Such
Similarly, disconnected groups of sensors may fail to report applications require a barrier coverage with a permanent
the event detected to the sink. In both cases, the quality of connectivity. Depending on the application requirements,
data gathered is inappropriate, making, new a deployment one or several barriers are needed, the latter case being
necessary. called multiple barrier coverage.
To save energy and maximize network lifetime, it is UC6 air pollution monitoring in a smart city. Partial area
necessary after the final deployment to schedule node activity coverage is sufficient and intermittent connectivity can
to make nodes sleep (e.g. redundant nodes for full coverage, be compliant with the application requirements.
useless nodes for partial coverage) while meeting the
application requirements. Notice that node activity scheduling UC7 instantaneous snapshot of measures taken at locations
differs from sensor node deployment, because existing sensor predefined by the application. In precision agriculture,
nodes are only switched on or off but are not moved. the goal is to detect the appearance of diseases in the
crops. In a smart city, the goal is to track an air pollutant.
This paper is organized as follows. This section ends with a Such applications require the coverage of static points
description of representative use cases and the positioning of interest. Permanent connectivity may be not needed.
of our contribution with regard to other surveys. Section 2 Intermittent connectivity can be provided by mobile
defines the coverage and connectivity issues encountered robots (e.g. tractors for precision agriculture).
in wireless sensor networks, (WSNs). Section 3 deals
with analysis criteria for deployment algorithms, and more UC8 tracking of wild animals or a truck fleet with embedded
particularly presents factors impacting the deployment, sensors. In such a case, different technologies can
common assumptions and models adopted, as well as be used to track these mobile points of interest (e.g.
performance evaluation criteria. In Sections 4, 5 and 6, we Argos beacons for animals, 3G/4G systems for trucks).
focus on deployment algorithms ensuring coverage of area, Depending on the application requirements, connectivity
barrier and point of interest respectively. Section 7, explores may be intermittent (e.g. animals) or permanent (e.g a
some energy-efficient optimization of a deployment, based on truck fleet).
node activity scheduling. In Section 8, we provide guidelines
UC9 health monitoring of isolated workers, disabled people
to help the designer to select the deployment algorithm
or elderly. They are considered as mobile Points of
suitable for the application requirements. Finally, we discuss
interest that must be permanently covered. Permanent
some trends and open issues for deployment algorithms in
connectivity is required.
Section 9 before concluding in Section 10.
All these uses cases will enable us to classify the coverage
1.2 Representative use cases and connectivity problems encountered in the literature (see
Table 1), according to the criteria defined more precisely in
Depending on the application requirements, we can
Section 2.
distinguish the following use cases (UC) dealing with coverage
With the emergence of smart cities, different use cases can
and connectivity, and representative of most applications:
coexist simultaneously. For instance, air pollution monitoring,
UC1 monitoring of a temporary industrial worksite requires surveillance of parking lots, public lighting control, and
full area coverage, permanent network connectivity and pollutant tracking are examples of sensor deployments that
a uniform deployment of sensor nodes to reduce data will be very common in our cities in the near future.
gathering delays and provide a better balancing of node
energy. 1.3 Related work
UC2 forest fire detection requires full area coverage in dry
In this section, we position our work with regard to other
seasons and only 80% in rainy seasons. Permanent
existing surveys and highlight our contribution. Existing
connectivity is required in both cases to alert the
surveys (2; 3; 4; 5; 6; 7; 8) introduce basic concepts
firefighters.
related to coverage and connectivity. For instance, (2)
UC3 detecting and tracking of intruders in restricted areas. focuses on how to ensure area coverage and how to
Such applications require full area coverage; furthermore, deploy sensor nodes. (3) classifies coverage problems as
the most critical zones should be covered by more than coverage based on exposure and coverage exploiting mobility.
one sensor node (i.e. multiple coverage). Permanent Area coverage, point coverage and barrier coverage is
connectivity is also required. another classification proposed and detailed in (4) and (5).
4 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

Area coverage Barrier coverage PoI coverage

Full Partial Full Partial Static Mobile
Simple Multiple Simple Multiple
Permanent Simple or multiple UC1 UC3 UC2 UC5 UC4 UC8 UC9
Connectivity Intermittent UC6 UC4 UC4 UC7 UC8
Table 1 Classification of use cases.

In (6), the authors distinguish two coverage problems: Applications such as battlefield monitoring require full area
static coverage and dynamic coverage. They also propose coverage. In these applications, every location is covered by
a study of sleep scheduling mechanisms to reduce energy at least one sensor node (1-coverage) or by k > 1 sensor
consumption and analyze the relationship between coverage nodes (k-coverage). Deploying sensors over a large area while
and connectivity. An overview of existing centralized and ensuring full coverage and network connectivity may be
distributed deployment algorithms is given in (7). The authors expensive. However, full coverage with connectivity provides
in (8) discuss the different deployment algorithm strategies the best surveillance quality. In the following we detail one-
such as forces, computational geometry and pattern based coverage defined as simple coverage and k-coverage defined
deployment. These surveys are good references to have an as multiple coverage, depending on the degree of robustness
overall view of coverage and connectivity issues in WSNs. required by the application.
In this survey, we define the coverage and connectivity − Simple coverage
problems separately to provide a better understanding. The In WSNs, it is necessary to ensure full coverage of the area
originality of our approach lies in a different viewing considered while deploying the minimum number of sensor
angle. We provide comprehensive definitions of coverage and nodes. This can be satisfied by covering every location in the
connectivity with their possible variants. Indeed, these variants field using at least one sensor node. Then information detected
depend on the latency and robustness requirements that differ in this location should be reported to the sink. Many studies
in the applications considered, leading to representative use aim to minimize the number of nodes deployed while ensuring
cases. For each use case, we list some deployment algorithms coverage and connectivity. For instance, the triangular lattice
found in the literature. We give a global analysis of the deployment provides full coverage, connectivity and uniform
deployment problem by discussing the impacting factors, deployment using the minimum number of sensor nodes.
detailing the common assumptions and models adopted in the − Multiple coverage
literature. Moreover, we propose some performance criteria to Multiple coverage is defined as an extension of simple
evaluate deployment algorithms. In order to help the designer coverage and is denoted by k-coverage. It is specific to
to choose the most suitable deployment algorithms, we applications such as distributed detection, mobility tracking,
dedicate an entire section to questions and recommendations monitoring in high security areas and military intelligence in a
regarding coverage and connectivity problems. We also battlefield. Since the failure of a single node may result in the
provide a section summarizing current trends and discuss open loss or corruption of important data, one degree of coverage
issues for deployment algorithms. is not sufficient for these applications. Such applications
require highly accurate information in order to provide fault
tolerance and allow good decisions to be made. The k-
coverage deployment is defined as a sensor deployment pattern
2 Coverage and connectivity problems in WSNs
where each point in the area is covered by at least k deployed
sensor nodes. Then, k-coverage tolerates at least k − 1 node
2.1 Coverage failures while maintaining coverage.

An area is said to be covered if and only if each location

of this area is within the sensing range of at least one active
sensor node.

In our work, we distinguish three types of coverage problems

: Area coverage, Point coverage and Barrier coverage.

2.1.1 Area coverage

a Simple coverage. b Multiple coverage.
In the area coverage problem, the goal is to cover the whole
area. Depending on the application requirements, full or partial Figure 1: Full area coverage.
coverage is required. However, if the number of sensors is
not sufficient, full coverage cannot be achieved and the goal
becomes maximizing the coverage rate.

• Full coverage • Partial coverage

 Coverage and Connectivity Issues and Challenges in WSN 5

In some applications, full coverage of a given area is not

required, in which case partial coverage ensuring a given
degree of coverage is sufficient and acceptable. Partial
coverage can be defined as the set of sensor nodes that
cover at least θ percent of the entire area and is referred
to θ-coverage where 0 < θ < 1. Generally, environment
monitoring applications require only partial coverage. An
example of such an application is given by temperature
applications where it is sufficient to sense the temperature of
80% of the region to know the temperature in this region.
Another example is given by forest fire applications where full Figure 3: Static PoIs coverage.
coverage of the forest is required in the dry season whereas
only an 80% coverage rate is required in the rainy season.
Partial coverage is a way of saving the energy consumption A point of interest is considered mobile if it changes its
of sensor nodes and prolonging the network lifetime since location. We distinguish two solutions to cover this mobile
the number of sensor nodes deployed is less than the number PoI. If mobile sensors are used, then they should be deployed
required to fully cover the area considered. Figures 2, 15 and in such a way as to cover this mobile PoI and keep track of it
16 depict sensors deployment ensuring partial coverage. when it moves to a new position. If static sensors are deployed
then they should be placed such that for each new position of
the PoI there is at least one sensor node that can cover it.

The monitoring of these points can be permanent (each point

is permanently monitored by at least one sensor) or not. In the
latter case, a mobile sensor should visit this point to collect
its data.

2.1.3 Barrier coverage

In several important applications, sensors are not designed

Figure 2: Partial coverage. to monitor events inside the area considered but to detect
intruders that attempt to penetrate this area. Examples of
such applications involving movement detection are the
deployment of sensors along international borders to detect
2.1.2 Point coverage illegal intrusion, around forests to detect the spread of forest
fire, around a chemical factory to detect the spread of lethal
In many applications, monitoring the whole area might be chemicals, and on both sides of a gas pipeline to detect
unnecessary, it is being sufficient to monitor only some specific potential sabotage. A major goal of these applications is to
points. Each specific point should be covered by at least one detect intruders as they cross a border or as they penetrate a
sensor node. Consequently, monitoring only these Points of protected area. Barrier coverage, which guarantees that every
Interest (PoI) will increase the sensing performance since all movement crossing a barrier of sensors will be detected,
the available sensors are used to monitor the PoIs, ensuring is known to be an appropriate model of coverage for such
a better coverage of these PoIs. Furthermore, the deployment applications. There are two types of barrier coverage: full
cost will decrease because of the smaller number of sensors barrier coverage or partial barrier coverage.
used compared to the number required to cover the entire area.
• Full barrier coverage
Examples of point of interest monitoring, include monitoring
of enemy troops and bases, capturing the real-time video A barrier is fully covered if every location of this barrier is
material of possibly mobile targets. In such applications, covered by at least one sensor node as it is shown in Figure 4.
mobile flying sensors can be deployed to monitor a point of
interest. The PoI can be either fixed or mobile.
• Fixed PoI
A PoI is fixed if it always has the same location. It is simpler to
cover a fixed PoI with prior knowledge of its position than to Figure 4: Full barrier coverage.
follow a mobile PoI. Figure 3 depicts an example of static PoI
monitoring. In this example sensor nodes do not only cover the
PoI but also maintain the connectivity with the sink to report • Partial barrier coverage
detected events.
When the number of sensors is insufficient to fully cover
• Mobile PoI the barrier, sensor nodes will provide partial coverage. The
6 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

deployment algorithm should ensure that by moving, the intermittent connectivity by using a mobile sink that moves
sensor nodes will detect an intruder trying to cross the barrier, and collects information from disconnected nodes. There are
with a probability that is higher than a given threshold. two types of intermittent connectivity: the first one uses only
one or several mobile sinks and the second uses a mobile sink
2.2 Connectivity and multiple throwboxes (Cluster heads).

• Isolated nodes
Two sensor nodes are said to be connected if and only
if they can communicate directly (one-hop connectivity) or When the radio range is less than the sensing range,
indirectly (multi-hop connectivity). In WSNs, the network full coverage can be achieved but without maintaining
is considered to be connected if there is at least one path connectivity between neighboring nodes. Consequently, these
between the sink and each sensor node in the considered area. nodes will be isolated. One solution to collect the detected
information from isolated nodes is to use one or several mobile
To monitor a specific area it is not enough to ensure coverage sinks. One or several nodes are in charge of visiting any sensor
without considering connectivity. When an event is detected, node that is not connected to the sink.
it should be reported to a sink. Consequently, it is necessary to • Connected components
ensure the connectivity between the sensor nodes and the sink
in order to guarantee the transfer of information to the sink. In any connected component, all sensor nodes of this
There are two types of network connectivity: full connectivity component are connected to each other. However, they are
and intermittent connectivity. disconnected from nodes in another connected component
and they can also be disconnected from the sink. To take
2.2.1 Full connectivity advantage of the connectivity within a connected component,
a throwbox, illustrated in Figure 5 by green nodes, can be
As connectivity is essential to guarantee the transfer of assigned to each connected component. A throwbox has the
information, it cannot be neglected and should have the same task of collecting the information of each node belonging to its
degree of importance as coverage. Thus, to efficiently monitor component. Then, a mobile sink (blue node in Figure 5) will
a given area, many applications require not only full coverage not collect information from each node in the network but just
but also full connectivity in order to collect information and take information from throwboxes. One or several nodes are in
report it. charge of visiting the throwbox of each connected component.
As we saw in the previous section dealing with full
coverage, full network connectivity can also be either simple
(1-connectivity) or multiple (k-connectivity). In addition,
full connectivity can be maintained during the deployment
procedure or it can be provided only when sensors have been
deployed in the area. In the following, we use connectivity to
represent full connectivity.
• Simple/Multiple connectivity
Full connectivity is said to be simple if there is a single path
from any sensor node to the sink.
Full connectivity is termed multiple if there are multiple
Figure 5: Intermittent connectivity using a mobile sink and
disjoint paths between any sensor node and the sink.
throwbox.
• Preserved connectivity
Considering only initial sensor deployments where all the
nodes are connected to each other and to the sink, this
connectivity is maintained during the deployment procedure. 2.3 Classification with regard to coverage and
This means that at any time during the deployment, there is a connectivity problems
path connecting every sensor node to the sink.
Table 2 provides a classification of the deployment algorithms
• Connectivity at the end of the algorithm
studied in this survey. For each of them we give the coverage
During the deployment process connectivity can be lost. and connectivity problem addressed. Notice that we use the
However, at the end of its execution, the deployment algorithm same classification criteria as in Table 1 of Section 1.2.
should guarantee full connectivity.

2.2.2 Intermittent connectivity 3 Analysis criteria of deployment algorithms

In some applications, it is not necessary to ensure full In this section, we analyze the different factors which have
connectivity in the area considered. It is sufficient to guarantee a positive or negative impact on the deployment. We discuss
 Coverage and Connectivity Issues and Challenges in WSN 7

Area coverage Barrier coverage PoI coverage

Full Partial Full Partial Static Mobile
Simple Multiple Simple Multiple
(20), (28), (16), (29), (27), (57) (49), (50) (49) (46), (47) (47)
Permanent
Connectivity (22), (23), (30) (52)
Simple (32), (33),
(34), (35),
(43), (40),
(41)
Multiple (16), (29), (52) (46)
(30)
Intermittent (54), (55) (53) (44), (45)
Table 2 Classification of deployment algorithms.

the common assumptions and models found in the literature deployment algorithm is chosen, this algorithm should include
before focusing on the relationship between the sensing range, a neighborhood discovery phase as well as a spreading phase
r, and the communication range, R, which highly impact the to allow sensor nodes to quickly discover other connected
behavior of the deployment algorithm. Moreover, we define components.
performance criteria for evaluation purposes. We end this • The energy of sensor nodes is difficult or impossible
section by highlighting the salient features of representative to renew, and this fact is of great importance. In the
deployment algorithms. deployment phase, the main reason for energy consumption
is the movement of the nodes, whereas in the data gathering
phase it is communication between the nodes. In both phases,
3.1 Factors impacting the deployment
energy-efficient techniques must be used.
Several factors impact the deployment provided and determine • The presence of obstacles makes the deployment more
how satisfactory the application is. They concern: complex: no sensor node should be placed within an obstacle.
• The assumptions and models used concerning r the Hence, the obstacles must be detected and a strategy must be
sensing range and R the communication range. Such used by the deployment algorithm to get around the obstacles.
assumptions and models are discussed in the next section. Furthermore, if the shape of the entity monitored is complex
The discrepancy between these oversimplified models and with irregular borders, some extensions to the deployment
reality may explain why the results obtained are not those algorithm will be needed.
which might be expected. The values of r and R determine • The quality of the data gathering required by the
the minimum number of sensors needed to fully cover the application may lead to a uniform and regular deployment.
entity monitored (i.e. area, barrier or PoI). The deployment Such a deployment provides smaller data gathering delays (9),
algorithms that use exactly this number are said to be optimal. a better time and space consistency of the data gathered, which
Depending on the relationship between r and R, detailed in leads to a more accurate snapshot of the measures taken.
Section 3.3, some algorithms either work or not. Others are • The positioning system may introduce some inaccuracy
valid whatever the relationship between r and R, but are not, in the position of the nodes; such a positioning error is very
however, optimal in all cases. common with GPS. To meet the application requirements, the
• The number of sensor nodes available for the deployment deployment algorithm should not accumulate the positioning
and the dimensions of the entity monitored will determine errors during the deployment.
whether this number is sufficient to fully cover the entity
monitored. It is usually assumed that this entity has a regular 3.2 Common assumptions and models
shape (e.g. rectangle, disk, etc). However, the reality is often
more complex with irregular borders. The common assumptions and models found in the literature
• The sensor nodes’ ability to move is a determining concern:
factor. If sensor nodes are unable to move, the only possible • Communication:
deployment is an assisted one, where a mobile robot for
example is in charge of placing the static sensor nodes at their − A unit disk graph model is generally adopted, where
final location. Otherwise, self-deployment is done, where each any two nodes whose Euclidean distance from each other
sensor node is autonomous and able to move. Notice that in is less than or equal to the communication range R, have a
such a case, the sensor nodes’ movement will consume more communication link: they are able to communicate in both
energy than communication during the deployment. directions. This binary model is, however, too simple and does
• The initial topology may require some extensions to the not match the real world. Some authors have introduced more
deployment algorithm. For instance, if the initial topology complex models where the probability of success falls less
comprises several disconnected components and a centralized abruptly when the distance increases up to R (10).
deployment algorithm is used, a mobile robot should be − A consequence of the unit disk graph model is that
used to collect the initial positions of the nodes needed by any wireless link is assumed to be symmetric. This assumption
the centralized deployment algorithm to compute the final is not always true in the real world.
positions of these nodes and this information should be − A frequent assumption is that all sensor nodes have
disseminated to them. If on the other hand, a distributed the same communication range. Sensor nodes may differ
8 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

in their age, their manufacturer, and their communication 3.3.1 Sensor deployment algorithms based on the
capacity. Hence some sensor nodes may have a higher relationship between R and r
transmission range than others.
− The initial topology considered in centralized
deployment algorithms is usually connected with the sink. • Case R ≥ 2r: Full coverage implies connectivity
This may not be the case in the real world (see the discussion in
In (11) and (12), the authors prove that when R ≥ 2r the full
Section 3.1). In distributed deployment algorithms, the initial
coverage of a convex area implies full network connectivity.
topology is generally random, as it facilitates the spreading
This result is extended to k-coverage and k-connectivity in
of nodes, leading to shorter convergence delays. For instance
(12). Then, using this assumption, it is sufficient to ensure full
Figure 6a depicts an initial topology where some sensor nodes
coverage, and connectivity will be a consequence.
are unable to communicate with the sink. In addition, Figure 6b
√
depicts another initial topology where all the sensor nodes are • Case R ≥ 3r: Full coverage implies connectivity
grouped at an entry point but unable to communicate with the √
sink. In (13), it is proved that when R ≥ 3r, ensuring full coverage
implies full connectivity. Moreover, the number of sensors
needed is optimal, when the triangular lattice is used as a
deployment pattern. For instance, in (22), the authors propose
a deployment algorithm where each sensor node should √ be
placed in a vertex of an equilateral triangle of edge 3r.

• Case R = r

An optimal deployment algorithm is proposed in (16) to

ensure full coverage and 1-connectivity when R = r. In this
algorithm, sensor nodes are deployed along an horizontal line,
a Random Topology. b Entry point topology. each two neighboring nodes are at a distance of r. Adjacent
√
lines are at a distance of ( 23 + 1)r. In such a deployment, full
Figure 6: Intial disconnected topology.
coverage is ensured but only sensor nodes located in the same
line are connected. That is why the authors propose adding
a sensor node between each two adjacent lines in order to
connect them, such that these nodes form a vertical line. Then
• Sensing:
1-connectivity is ensured. The optimality of this deployment
− A unit disk graph model is used to model the sensing in terms of the number of sensor nodes was proved in (13).
of a sensor node. Any event occurring within the disk of radius √
• Case R < 3r
the sensing range r, centered at the sensor node is detected. √
This assumption is too optimistic in the presence of obstacles, When R < 3r, full coverage does not imply network
for instance. connectivity. Network connectivity is necessary to report
− The homogeneity of sensors (i.e. the same sensing information and it is an important part of the monitoring
model with the same sensing range) is generally assumed. This task. Then, ensuring connectivity while maximizing the area
may not be the case in the real world. coverage becomes the goal of the deployment algorithm. The
deployment algorithm proposed in (16) which deploys sensor
• The presence of obstacles: nodes in horizontal lines and connects these lines by placing
− Most authors assume that the entity to monitor is sensor nodes between two adjacent lines, is generalized in (13)
flat and nodes can move freely without obstacles. Such an as illustrated in Figure 7. In addition, this deployment is
assumption cannot be made for rescue applications after a optimal when the distance between neighboring sensor nodes
disaster, for instance. q line R and the distance between two adjacent lines
in the same
R2
is r + r2 − 4 .
In the next section, we study the relationship between r and
R in more detail. • Case arbitrary R and r

3.3 Relationship between coverage and connectivity In (27), the authors propose an algorithm that aims
at preserving network connectivity while maximizing area
Some deployment algorithms only work when a given coverage. Starting with an initial deployment where all sensor
relationship exists between the radio range R and the sensing nodes are connected to the sink, a virtual force algorithm
range r. For instance, if R ≥ 2r, it is sufficient to ensure full is applied in order to redeploy sensor nodes in the area
coverage, and connectivity will be provided as a consequence. considered. As the sensing
√ and radio ranges do not meet
In the following, we study the different cases considered in the assumption R ≥ 3r, when sensor nodes move to their
the literature. Furthermore, we recall some results concerning new positions they check whether they are still connected
optimal deployments based on regular patterns. to the sink. If they are not, they move towards the sink
 Coverage and Connectivity Issues and Challenges in WSN 9

a Triangular deployment. b Square deployment.

Figure 7: Sensor deployment with added sensors to ensure

connectivity.

until connectivity is established. This algorithm preserves full

network connectivity during the deployment process and tries
to maximize the area coverage with any given values of R and
r. In (20), the authors propose a deployment algorithm that c Hexagonal deployment. d Rhomboid deployment.
aims at ensuring full coverage and full network connectivity of
an area containing obstacles of different shapes. The authors Figure 8: Regular deployment pattern.
propose dividing the area into two different types of region:
small regions or large regions which can contain boundaries
and obstacles. As there are no assumptions concerning R and 1 3/4 √
• 3 ≤ Rr ≤ 2, the square grid is the best deployment
r, in the small regions (like a belt), sensors are deployed 2
along the bisectors of this region and are separated by rmin = pattern. See Figure 8b.
min{R, r}. In the large region, sensor nodes are deployed in √ √
• 2 ≤ Rr ≤ 3, the rhomboid pattern is the best
rows. The distances which separate sensor nodes and rows are deployment pattern. See Figure 8d.
determined according to the values of R and r.
√
• Rr ≥ 3 the triangular lattice is the best deployment
3.3.2 Optimal number of sensor nodes for regular pattern. See Figure 8a.
deployment patterns
3.4 Criteria for performance evaluation
Sensor nodes can be deployed in a regular pattern. This pattern
can be a triangular lattice, a square grid, an hexagonal grid or Each pattern may fit some application requirements. The
a rhomboid grid. In (5), the authors specify for each pattern question is then how to evaluate and select the best one.
a condition that ensures coverage of the area and guarantees Different evaluation criteria have been introduced:
network connectivity as a consequence. • coverage: (e.g. area, barrier, point of interest) is the main
criteria to evaluate the efficiency of the algorithm. Usually,
• If R ≥ r and the hexagonal grid pattern is used, then full coverage is computed as follows: the area to cover is divided
area coverage is ensured and the network is connected. virtually into LxW grid units. A grid unit is considered to
√ be covered if and only if its centered point is covered by at
• If R ≥ 2r and the square grid or rhomboid pattern is least one sensor node. The coverage rate is computed as the
used, then full area coverage is ensured and the network percentage of grid units covered.
is connected. • connectivity: is also important. The type of connectivity
√ (e.g. full or intermittent) is application dependent. For some
• if R ≥ 3r and the triangular lattice pattern is used, applications, maintaining full connectivity is required in
then full area coverage is ensured and the network order to report any detected event immediately to the sink.
is connected. The triangular lattice is the optimal Other applications with fewer constraints require intermittent
deployment pattern to ensure full area coverage and connectivity: usually a data mule.
guarantee network connectivity. • convergence and stability: convergence is evaluated by
the convergence time defined as the time needed to achieve the
These conditions are studied in (13) with regard to the optimal required coverage and connectivity. In distributed deployment
number of sensor nodes and the regular pattern used. It was algorithms, the convergence may be difficult to reach because
proved that when: of node oscillations. Hence, the stability of the deployment
1 is an important criterion to detect the completion of the
• 0 < Rr ≤ 33/4 , the hexagonal grid is the best deployment.
2
deployment pattern (i.e. it requires the minimum number • energy and distance traveled: during the deployment,
of sensor nodes). See Figure 8c. the main cause of energy consumption is the mobility of the
10 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

Goal(s) Relationship between R and r Deployment pattern: examples

1 3/4
0< R r ≤ 3 Hexagonal grid (13)
2
1 3/4 R
√
Full coverage 3 ≤ r ≤ 2 Square grid (13)
2√ √
R
(Coverage implies connectivity) 2≤ r√ ≤ 3 Rhomboid pattern (13)
R
r ≥ 3 Triangular lattice (13)

R=r - Horizontal lines + a node between two adjacent lines (16)

1-Full or partial coverage by horizontal lines
2-Connectivity by an additional vertical line - Horizontal lines + a√
node between two adjacent lines (13)
Optimal when R < 3r, distance betweenq nodes R
R2
R<r and distance between adjacent lines = r + r2 − 4

Ensuring connectivity and maximizing coverage No assumptions - Floors (27)

- Dividing the area into small and large regions (20)
Full coverage and Full connectivity No assumptions Sensors are deployed along the bisectors of small regions
and in rows in the large regions
Table 3 Relationship between r and R.

nodes. That is why the total distance traveled by the nodes must 4 Area coverage and connectivity algorithms
be measured, as this measure reflects the energy consumed.
Obviously, minimizing the total distance traveled leads to 4.1 Full coverage
savings in energy. Notice that the convergence and stability
performance has a strong impact on the distance traveled and Many deployment algorithms aim to ensure full coverage of
the energy consumed. Once the deployment has been carried the area considered. These algorithms are classified into three
out and the nodes are stationary, the data gathering takes strategies. We distinguish the forces-based strategy, the grid-
place. The main cause of energy consumption in this phase is based strategy and the computational geometry-based strategy.
communication. To maximize network lifetime, node activity
scheduling can be used to make nodes sleep when they are not
needed for the data gathering. 4.1.1 Forces-based strategy
• communication overhead: comes from the control
messages exchanged between the nodes to organize the The forces-based strategy is known by its simple deployment
deployment and the data gathering. In the case of contention- principle. This principle is based on virtual forces that can
based medium access, collisions imply retransmission and be attractive, repulsive or null. In this strategy, a sensor node
increase the overall bandwidth and energy consumption. The should maintain a fixed threshold distance called Dth with
aim is to reduce this overhead. its 1-hop neighbors. Then, if the distance separating two
• uniformity, regularity and optimality of the deployment: neighboring nodes is greater than Dth , an attractive force is
if the space consistency of measures taken is expected, a exerted, whereas if this distance is less than Dth , a repulsive
uniform deployment is needed: all the nodes (except the border force is exerted. Otherwise, the force is null since the distance
ones) should have the same number of neighbors. Similarly, separating neighboring sensor nodes is equal to Dth , the
if the measures should be taken at equidistant positions, a required distance. This principle is illustrated in Figure 9,
−→
uniform and regular deployment is needed. Usually, such where Fij denotes the force exerted by sensor node j on sensor
a deployment reproduces the same geometric pattern (e.g. node i.
triangle, hexagon, square , etc). Depending on the relationship
between r and R, some patterns are optimal. This optimality is
useful because it requires the smallest number of sensor nodes
to meet the application requirements. A uniform and regular
deployment is also mandatory when the application requires
time and space consistency of the data gathered.

3.5 Salient features of deployment algorithms

For each protocol studied, we define the coverage and

connectivity problem solved, the strategy used, its type
(centralized, distributed) and its specific assumptions. For area
coverage, we distinguish different strategies that are detailed Figure 9: Forces based strategy.
in Sections 4, 5 and 6 for area, barrier and PoI coverage,
respectively.
 Coverage and Connectivity Issues and Challenges in WSN 11

Area coverage
Protocol Coverage problem Connectivity problem Strategy Cent/Dist Specific assumptions
VFA (21) Full coverage Permanent connectivity Forces based Centralized
R R
Extended VFA(24) Full coverage Permanent connectivity Forces based Distributed r > 2.5 and r < 2.5
IVFA (25) Full coverage Permanent connectivity Forces based Distributed
EVFA (25) Full coverage Permanent connectivity Forces based Distributed √
DVFA (22) (23) Full and uniform coverage Permanent connectivity Forces based Distributed R ≥ 3r
CPVF (27) Maximized coverage Permanent connectivity Forces based Distributed arbitrary R and r
Push&Pull (28) Maximized coverage Permanent connectivity Forces based Distributed Triangular lattice
Forces based Square√grid
VFCSO(26) Full coverage Full connectivity Grid based Centralized R ≥ 5r
Node activity scheduling
(29) Full coverage Permanent connectivity Grid based Distributed
Multiple Multiple
(30) Full coverage Permanent connectivity Grid based Distributed
Multiple Multiple
HGSDA(32) Full coverage Permanent connectivity Grid based Centralized Triangular
√ lattice
R ≥ 3r
C 2 (34) Full coverage Permanent connectivity Grid based Distributed Triangular lattice
Energy saving
(33) Full coverage Permanent connectivity Grid based Distributed Square pattern
Square pattern
(35) Full coverage Permanent connectivity Grid based Distributed Static node
Assisted by robot
(56) Partial coverage Permanent connectivity Grid based Cent/Dist
VEC, VOR and Maximized coverage Permanent connectivity Computational Distributed Voronoi diagram
Minimax (40) geometry based
(43) Full coverage Permanent connectivity Computational Centralized Delaunay triangulation
geometrie based Obstacles
(41) Full coverage Permanent connectivity Computational Centralized Static nodes
geometry based
Robot collector
(54) Full coverage Intermittent connectivity Random Centralized Cluster head
Energy saving
(55) Full coverage Intermittent connectivity Random Centralized Ferries
(15) Full coverage Permanent connectivity Random Distributed Node activity scheduling
(17) Full coverage Permanent connectivity Random Centralized Node activity scheduling
Connected graph based
(12) Full coverage Permanent connectivity Random Distributed R ≥ 2r
Simple-Multiple Node activity scheduling
(19) Full coverage Permanent connectivity Random Distributed Arbitrary R and r
Node activity scheduling
(57) Partial coverage Permanent connectivity Random Dist/Cent Node activity scheduling

Barrier coverage
Protocol Coverage problem Connectivity problem Strategy Cent/Dist Specific assumptions
(49) Full coverage Permanent connectivity Grid based Distributed Mobile sensors
Simple-Multiple
(50) Full coverage Permanent connectivity Random Centralized Random offset < r
(51) Full coverage Permanent connectivity Random Centralized
MBC (52) Full coverage Permanent connectivity Deterministic Distributed R ≥ 2r
Simple-Multiple Dynamic object
CSP (53) Partial coverage Intermittent connectivity Probabilistic Centralized
PMS (53) Partial coverage Intermittent connectivity Probabilistic Centralized
(59) Full coverage Permanent connectivity Random Centralized Node activity scheduling

Point of Interest coverage

Protocol Coverage problem Connectivity problem Strategy Cent/Dist Specific assumptions
(47) Full coverage Permanent connectivity Forces based Distributed RNG for connectivity
Static PoI
(44) Temporary coverage Intermittent connectivity Random Distributed Ferries
R ≥ 2r
DSWEEP (45) Temporary coverage Intermittent connectivity Distributed √
(46) Full coverage Permanent Connectivity Grid based Distributed R≥ 3r
Table 4 Salient features.
12 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

The virtual forces algorithm (VFA) is proposed in (21) certain threshold. In this way, energy consumption is reduced
as a centralized redeployment algorithm to enhance an initial during the deployment which provides a fast convergence to
random deployment. In the initial deployment, any sensor a coverage rate close to 100%. DVFA is also used in (23) to
node is able to communicate with the sink in a one-hop or cope with obstacles of different shapes. By using the virtual
multi-hop manner. Then, the sink computes the appropriate forces principle and a method to avoid the obstacles, full area
new position of each sensor node based on the coverage coverage is ensured even when an obstacle has a confined
requirements and using the virtual forces mechanism. In shape.
this work, obstacles exert a repulsive force and an area of
preferential coverage exerts an attractive force on sensor Usually, the virtual forces strategy is used to ensure full
nodes. During the execution of the virtual forces algorithm, area coverage as the attractive and repulsive forces spread
sensor nodes do not change their positions. It is only when sensor nodes over the whole area and consequently achieve a
they receive their final positions from the sink that they move high coverage rate rapidly. Furthermore, this strategy is used
directly to them. VFA is a centralized algorithm that offers a in (27) with the goal of preserving network connectivity. This
good coverage rate of the area considered while maintaining deployment algorithm, called CPVF, Connectivity-Preserved
network connectivity. However, a central entity must know Virtual Force, is used to monitor an unknown area with
the initial positions of all sensor nodes, compute their final an arbitrary ratio Rr . To achieve that, a sink periodically
positions and disseminate the positions to all sensor nodes. broadcasts a message to neighboring sensors which in turn
This principle is problematic when network connectivity flood the message to all connecting nodes. A sensor node is
is not initially ensured. Furthermore, when the network is considered to be disconnected from the network if it does not
very dense, this algorithm has a poor performance due to the receive the flooding message. Then, it moves toward the sink
gathering of the initial positions of sensor nodes. in order to reconnect. This algorithm induces a high overhead
To cope with the scalability problem, distributed versions in terms of messages broadcast in the network to check the
of VFA are proposed in the literature. For instance, the connectivity of the nodes with the sink. This paper also
extended virtual forces-based approach proposed in (24) proposes a floor-based scheme to improve the global network
copes with two drawbacks of the virtual forces algorithm: coverage by reducing overlapping. This scheme is based on
the connectivity maintenance and nodes stacking problems the division of the area into equidistant floors (distant of
(i.e. two or more sensor nodes occupy the same position). 2r) and encourages sensors to stay in the floor lines. Sensor
The connectivity maintenance problem occurs when the nodes are added in a column between floor lines to ensure
communication range is low, Rr < 2.5. Thus, the authors connectivity. Although this work aims at preserving network
propose adding an orientation force which is exerted only if connectivity when the ratio Rr is arbitrary, it requires a high
the node has fewer than 6 neighbors. This force aims to keep number of sensor nodes, as illustrated in Figure 10, because
the angle formed by one node and its two neighbors equal to the inter-floor distance is fixed to 2r for any value of R and r.
π
3 in order to provide a reliable connectivity and eliminate
coverage holes. Notice that these authors observe a stacking
problem, where several nodes are located in almost the same
position. This is because the coefficient of the attractive
forces is not well tuned. As a solution, the authors propose
an exponential force model to adjust the distance between a
node and its distant neighbors. However, the threshold value
of Rr = 2.5 is not explained and the maintained connectivity
is not proved in the paper. Furthermore, the additional
orientation force may induce node oscillations.
IVFA, Improved Virtual Force Algorithm, and EVFA,
Exponential Virtual Force Algorithm are two distributed Figure 10: Floor based deployment.
deployment algorithms proposed in (25). EVFA aims
at speeding up convergence because forces increase
exponentially with the distance between sensors. IVFA limits
the scope of virtual forces: only nodes in radio range of a
given node exert virtual forces on it. Furthermore, the stacking
4.1.2 Grid-based Strategy
problem is solved by using a very small attractive force with
regard to the repulsive force. IVFA converges to a steady The grid-based strategy provides a deterministic deployment
state faster than the basic virtual forces algorithm, and defines where the position of the sensor nodes is fixed according to a
a maximum movement in each iteration to reduce useless special grid pattern such as a triangular lattice, a square grid
moves and save energy. or a hexagonal grid (see Figures 11b, 12 and 13 respectively).
DVFA, proposed in (22), is another example of the distributed Then, the area is divided into virtual cells and depending on the
algorithm that uses the virtual forces to spread sensor nodes deployment algorithm used, sensor nodes are located either in
until the entire area is covered. The main drawback of this cell vertices or at the cell center.
algorithm is node oscillations. To deal with this problem, the The grid deployment is also a regular deployment pattern
authors of DVFA limit the distance sensor nodes move to a as all the generated grid cells have the same shape and
 Coverage and Connectivity Issues and Challenges in WSN 13

size. The regular deployment pattern is studied in (29) in

order to provide multiple coverage (p-coverage) and multiple
connectivity (q-connectivity) using the triangular lattice,
square or hexagonal pattern. The value of p and q are
provided by adjusting the distance separating sensor nodes and
limiting the ratio Rr . A comparative study of regular pattern
performance in terms of the number of nodes required is also
provided to achieve√1, 3 and 5-coverage and q-connectivity. a Triangular lattice b Triangular lattice.
With the ratio Rr ≥ 3, the triangular lattice is better than the with hexagonal cells.
square grid, which is better
√ than the hexagonal grid. However,
with the value of Rr < 3, the triangular lattice becomes Figure 11: Triangular lattice.
the worst. Multiple coverage and connectivity with regard to
the regular deployment pattern is also studied in (30). The
authors propose the optimal deployment patterns to ensure full of the deployment, the area is not yet divided into hexagonal
coverage and q-connectivity while q ≤ 6 for certain values cells. An initiative sensor node starts by snapping itself at
of Rr . They consider the hexagonal deployment pattern as a the center of the first hexagonal cell and selects six sensor
universal basic pattern that can generate all optimal patterns. nodes in its vicinity to snap them in the adjacent hexagonal
Then, they present different forms derived from the hexagonal cells. The selected sensor nodes move to their cells and in turn
pattern by changing the edge length and the angle between select other sensor nodes to occupy their adjacent cells. Then,
adjacent edges. hexagonal cells are built progressively in a distributed way:
When the applications require time and space consistency of the hexagonal side length is equal to the sensing range. Since
the measures taken by sensor nodes regularly distributed in the the sensor occupies the center of the cell, the triangular lattice
area, the regular deployment pattern can be a good solution is used as the deployment pattern.
to provide a high level of coverage and connectivity with a The deployment algorithm C 2 proposed in (34) is a triangular
minimum number of sensor nodes. lattice based strategy where a sensor node occupies a
In the following we present some research studies proposing hexagonal cell. Hexagonal cells are built progressively in a
a regular deployment pattern based on a triangular lattice and distributed manner by sensor nodes. This algorithm proceeds
a square grid. in two phases. In the first phase, called cluster heads selection,
the sink which is the first cluster head in the area considered,
Triangular grid starts by building its hexagonal cell and defines its position as
the cell center. The distance between the cell center and one
In (31), it was proved that the triangular lattice shown in of the vertices is R3 and the distance between two neighboring
Figure 11b offers the smallest overlapping area and requires cell centers is 2 R3 in order to maintain network connectivity
the smallest number of sensor nodes. When the triangular during the deployment process. Then, the sink determines the
lattice is used as a deployment pattern, each sensor node center of each neighboring cell and informs sensor nodes in
occupies a hexagonal cell. However, the deployment is not its neighborhood. The nearest sensor node to the cell center is
considered to be a hexagonal deployment (see Figure 13) since selected as a cluster head of its hexagonal cell. It should move
a sensor node is at the center of a hexagon and neighboring towards its cell center. In turn, the new cluster heads define the
sensors form a triangular pattern (see Figure 11). For instance, center of their neighboring cells. The second phase is called
the authors in (32) propose a deployment algorithm called node balancing and its goal is to improve area coverage by
HGSDA that deploys sensor nodes in a triangular lattice. This balancing the number of sensor nodes between cells. To do
deployment starts by dividing the area into small hexagonal so, if the difference between sensor nodes in two neighboring
cells and each cell center corresponds to a sensor position. cells is greater than 1, some sensor nodes will move to the cells
Although the cells are hexagonal, sensor nodes are deployed with a deficit number of nodes. In this deployment algorithm, a
in a√triangular lattice since the distance between two neighbors hexagonal grid is used to ensure full coverage and maintain full
is 3r and there is a sensor node at the cell center. HGSDA connectivity. Energy saving is achieved by selecting a cluster
identifies redundant sensor nodes in order to place them in head for each cell and balancing the number of sensor nodes
empty hexagonal cells. Since the size of a hexagonal cell between adjacent cells. This algorithm performs well when
is computed according to sensor sensing range and the area the sink is located at the center of the area and all the nodes
size, full coverage is achieved using the smallest number of are grouped around the sink.
sensor nodes. This algorithm is carried out by a sink. Then,
all the sensor nodes receive their final position from the Square grid
sink and move to it. HGSDA is a centralized algorithm that
ensures full coverage using the minimum number of sensor The square grid strategy is used in (33) where the area
nodes while ensuring simple connectivity with the sink in monitored is divided into square cells, as shown in Figure 12.
the final deployment. This centralized algorithm can only be Each cell represents the maximum square size that is covered
used if connectivity with the sink is ensured in the initial by one sensor node. Each sensor node occupies a cell center
deployment. The same deployment pattern is presented in to cover the corresponding square cell. If an empty cell exists,
(28), but in a distributed version. However, at the beginning neighboring sensor nodes should decide to which one will
14 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

move to cover it, such that if new empty cells appear, they will of partitioning the area into a number of polygons based
be around the sink. Redundant nodes should move toward the on distances to a specific discrete set of nodes. Each node
sink in order to cover empty cells that can occur along the occupies only one polygon and is closer to any point in this
path to the sink. polygon than any other node in the neighboring polygons.
These polygons can be obtained by drawing the mediator of
A grid-based approach is also used for robot-assisted sensor each two neighboring nodes. Consequently, the edges of the
deployment. As an example in (35), a robot places sensor polygons are equidistant from neighboring nodes. Delaunay
nodes at the vertices of a square cell. Then, each deployed triangulation is the dual graph of the Voronoi diagram. It can
sensor node colors itself white if it is adjacent to an empty be constructed by connecting each two neighboring points
cell and black otherwise. Neighboring sensor nodes exchange in the Voronoi diagram whose polygons share a common
hello messages to inform each other about white nodes edge. Voronoi diagram and Delaunay triangulation are used in
(empty cells) and maintain a back pointer corresponding WSNs to deal with coverage hole problems. The occurrence
to the nearest empty cell along the backward path of the of coverage holes after the deployment of sensor nodes in a
robot. Then, the robot backtracks this back pointer to drop given area can be considered as a cause of a low coverage
sensor node in the empty cell. This algorithm guarantees full rate. By detecting and healing these holes, the coverage rate
coverage in a failure free environment using a mobile robot can be maximized.
in a square grid.
It is assumed that the robot carries enough sensors to heal
any coverage hole (i.e. empty cell) that is detected. Such Deployment algorithms based on Voronoi diagram
strategies are used when the sensor nodes are static, and a
mobile robot is used to ensure coverage by repairing any Some schemes proposed are based on Voronoi diagram to
coverage hole detected by the sensor nodes. The new problem detect coverage holes. Sensor nodes are able to construct their
is that of detecting coverage holes and optimizing the robot Voronoi polygons based on location information received
movements. from their neighbors. Due to these Voronoi polygons, nodes
can determine coverage holes. Then, they move in order
to reduce or eliminate these holes while maximizing the
coverage rate of the area considered.
In (40), three distributed moving algorithms are proposed:
VEC, VOR and Minimax algorithms. The VECtor
based algorithm (VEC) is inspired by the behavior of
electromagnetic particles. When two electromagnetic particles
are too close to each other, an expelling force pushes them
a Sensors in the cell b Sensor in cell vertice. apart. VEC pushes sensor nodes away from a densely covered
centers. area. In contrast to the VEC algorithm, the VORonoi based
algorithm (VOR) pulls sensor nodes to the sparsely covered
Figure 12: Grid Based Strategy. area. The Minimax algorithm is similar to VOR. It fixes
coverage holes by moving sensor nodes closer to the furthest
Voronoi vertex. However, it does not go as far as VOR
to avoid situations in which a vertex that was originally
closer now becomes the furthest. Minimax chooses the
node target position as the point inside the Voronoi polygon
whose distance to the furthest Voronoi vertex is minimized.
Minimax and Vor do not ensure uniform coverage of the
final deployment since the algorithm stops as soon as full
coverage is obtained. Moreover, if the number of sensors is
not sufficient to cover the whole area, node oscillations may
occur.
Figure 13: Hexagonal pattern.

Deployment algorithms based on Delaunay

Triangulation

4.1.3 Computational geometry based Strategy In (43), a centralized algorithm is proposed to cope with the
boundaries and obstacles coverage problem. In their paper,
The computational geometry strategy is used to solve the authors propose a deterministic sensor node placement
problems based on geometrical objects: points, polygons, line to ensure full coverage of an area containing obstacles of
segments, etc. Voronoi diagram and Delaunay triangulation arbitrary shapes. Sensor nodes are deployed in a triangular
are two computational geometry methods used in WSNs lattice over the whole area as if there were no obstacles.
to solve static problems. The Voronoi diagram is a method Then, sensor nodes inside the obstacles are eliminated and
 Coverage and Connectivity Issues and Challenges in WSN 15

a Voronoi diagram. b Delaunay triangulation. a A large uncovered area. b Regular distribution of

uncovered area.
Figure 14: Computational geometry approach.
Figure 15: Partial coverage.

so coverage holes may occur around these obstacles. To deal

with this problem, Delaunay triangulation is used to partition
these coverage holes into triangles of edges less than r, and
then, a sensor node is placed in one of the triangle vertices to
cover it.

Other computational geometry deployment algorithms

Another study based on computational geometry strategy is
a Bad distribution. b Good distribution.
proposed in (41) to detect any coverage hole and calculate its
size. In this work, the authors do not rely on Voronoi diagram Figure 16: Different distributions of uncovered area.
or Delaunay triangulation, but, they propose a triangular
oriented diagram called HSTT that connects static sensor
nodes such that every three neighboring nodes form a triangle. specific parameter fixed by the application. The goal is to cover
Using a HSTT diagram, coverage holes can be detected and at least θ percent of the area considered while maintaining
the required number of mobile sensors to heal these holes a connected graph between these nodes. Partial coverage is
can be determined. Although this HSTT diagram presents useful to measure the temperature and humidity, to detect
some advantages compared to a Voronoi diagram, such as smoke and to provide an early warning of a potential forest
its simplicity and its accuracy when computing the size of fire (58), for instance.
the coverage holes, it requires a high energy consumption to In addition, to avoid a large uncovered area (see Figure 15a),
achieve its goal. the uncovered areas should be regularly distributed (see
Figure 15b). For that purpose, the authors in (56) propose
4.1.4 Other deployment strategies dividing the area to be monitored into subregions of equal size.
The goal is then to cover θ-percent of each subregion.
Other deployment strategies exist. They include off-line
optimization algorithms which compute off-line the best
4.3 Intermittent connectivity
position of each sensor node with the goal of ensuring the
coverage and connectivity required by the application. For this The deployment algorithms presented above ensure full
purpose, they employ optimization techniques, usually based or partial coverage with permanent connectivity. When
on a linear programming of the problem considered. They permanent connectivity is not required, intermittent
discretize the area of interest and decide for each point in the connectivity is provided, exploiting the mobility of some
area whether a sensor should be located there or not, taking into nodes. The strategies differ in:
account the application requirements (e.g. maximum number
of sensors, maximum cost, etc). See for instance (42). • the number of mobile nodes: one mobile node or several.
If several, how do the mobile nodes coordinate their
4.2 Partial coverage action to visit nodes and gather their data?
• the trajectory type of mobile nodes:
The area coverage problem has been widely studied in the
literature. As we have shown previously, much effort has been – a fixed predefined geometrical trajectory like a line
made to cope with full area coverage. However, only a few or a circle, for instance.
studies have focused on partial area coverage. – a trajectory that visits all the nodes or a subset
Generally, partial coverage is a solution to prolong the network of nodes depending on the deployment architecture
lifetime when full coverage is not required. The foremost (e.g. clustering).
requirement in this case is that the coverage rate provided
should be higher than some predefined bound which is a More particularly, we distinguish:
16 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

• The coverage obtained is generally very good. However,

in the distributed version it is achieved at the expense
of nodes moving large distances. This is due to node
oscillations that occur even when maximum coverage
has been reached. Such oscillations cause high energy
consumption and are detrimental to the network lifetime.

a Line ferry. b Path ferry c Two annular • With the enhancements brought by many authors ((22),
ferries. (24) and (25) for instance), maximum coverage is reached
faster.
Figure 17: Different Ferry trajectories.
Nevertheless, some issues remain unsolved, like the node
oscillations mentioned previously and the detection of the
Mobile sink with multiple cluster heads (throwboxes): end of the distributed algorithm.
In (54), a large number of sensor nodes are randomly dispersed
in a square area. Those sensor nodes are grouped into clusters The grid-based strategy has the following advantages:
and a cluster head is elected for each one. Obviously, sensor • It provides a regular deployment with deterministic
nodes are connected to their cluster head in order to report positions of sensor nodes (e.g. a triangular lattice, square
the detected information to it. The cluster head has the role pattern, etc), if a virtual grid is used.
of storing this information and waiting for the mobile sink.
A moving strategy for the mobile sink is proposed to collect • It requires a minimum number of sensor nodes to achieve
the information detected in the whole area while saving the the required coverage. The optimal deployment pattern
energy consumption. The mobile sink starts from a fixed (i.e. the pattern requiring the minimum number of sensor
point, follows a specific trajectory to visit each cluster head nodes) varies according to the relationship between R
and gathers information, and finally it returns to its starting and r.
point. Intermittent connectivity is provided using a mobile sink
communicating with cluster heads and coverage is maximized. • It can easily achieve k-coverage and connectivity.

• It exists in centralized and distributed versions.

Ferries: a ferry is a mobile robot that has a geometrical
trajectory like a line or circle. Sensor nodes can be randomly Usually, the distributed version is more complex. If a
deployed with no connectivity with the sink. The ferry virtual grid is not used, a sophisticated management of grid
will act as a relay between sensor nodes and the sink to cells is needed ((28),(34)). The complexity of this strategy
ensure communication, distribution and gathering of the data comes from managing the movement of nodes and the
collected by the nodes. Based on this principle, (55) studies the positions of newly built cells. Coverage holes can appear.
ferry trajectory that can be a line, path (multiple) or annular,
as depicted in Figure 17. Its goal is to optimize the route of Computational geometry-based strategies aim at improving
the ferries that collect information from the sensor nodes. the area coverage by healing previously detected coverage
holes. Like the other strategies, the computational geometry
based strategy exists in centralized and distributed versions.
4.4 Summary
The main drawback lies in the complexity of detecting
Area coverage has been widely studied in the literature. We coverage holes and computing the new nodes’ positions.
have distinguished three deployment strategies: a force-based Furthermore, the new deployment obtained is not uniform.
strategy, a grid-based strategy and a computational geometry- In addition, all these strategies have been enhanced to deal
based strategy. Based on the studies cited previously, we can with the existence of obstacles within the network area. A
observe that force-based strategies exhibit many advantages: better adaptability to the environment is still a challenge.
There are two types of wireless sensor networks, depending on
• The simplicity of the basic principle, which performs the mobility of sensor nodes. If the sensor nodes are mobile,
well both in centralized and distributed versions. In all the redeployment strategies (virtual forces strategy, grid
the distributed version, all the nodes apply the same based strategy and computational geometry strategy) can be
algorithm and play the same role. The distributed considered as self-deployment. Otherwise, sensor nodes are
version is based only on local information (coordinates static and mobile robots are used to put the sensor nodes in
of the nodes and their neighbors). It allows nodes to their final position. In this case the redeployment is said to be
progressively discover their environment and react to assisted.
changes in this environment without the need for a central
entity to manage these changes.
5 Barrier coverage and connectivity algorithms
• The uniformity of the redeployment obtained: the density
obtained is nearly the same and the same distance is Intruder detection and border monitoring are two important
maintained between the neighboring nodes. applications of WSNs. Barrier coverage is considered to be
 Coverage and Connectivity Issues and Challenges in WSN 17

an appropriate model for such applications. A deployment of 5.1.2 Mobile sensor nodes
sensor nodes along a barrier is necessary to detect an intruder
crossing, for example, an international border. Depending A deployment strategy to ensure (simple or multiple) barrier
on the application requirements and the number of sensor coverage using mobile sensor nodes is proposed in (49). This
nodes provided, this deployment can ensure either full barrier strategy consists in dividing the area into virtual lines (i.e.
coverage or partial barrier coverage. barriers) where the number of virtual lines matches the desired
robustness of barrier coverage. In each line, sensor nodes
should occupy grid points at a distance of 2r. Starting from a
5.1 Full barrier coverage random deployment in a rectangular area, mobile sensor nodes
should execute two phases to reach their final positions. In the
Full barrier coverage can be either simple or multiple. It is
first phase, each sensor node moves vertically to reach a line.
simple, if there is just one barrier that is fully covered by
Then, in the second phase, it moves horizontally along the line
sensor nodes. The barrier coverage is multiple if there are k
to a predetermined grid point position. When each grid point
successive barriers of sensor nodes.
is occupied by a sensor node, full barrier coverage is provided.
The authors in (48) are the first to address the problem of
providing the minimum number of deployed sensor nodes
to ensure simple or multiple barrier coverage. They define
a simple barrier coverage by a belt of successive sensor
nodes such that their sensing areas overlap. A multiple barrier
coverage is defined by the fact that every two successive
barriers have two overlapping sensor nodes, as depicted in
Figure 19: LNRO barrier deployment.
Figure 18b. Based on a theoretical study, the authors prove
that the optimal number of sensor nodes deployed along a
l
barrier is 2r , where l is the length of the barrier and r the
sensing range. Then, every two successive sensor nodes are (51) focuses on finding and healing barrier holes using
at a distance of 2r in order to optimize the overlapping (see mobile sensor nodes. This work is an extension of (50).
Figure 18a). To ensure full barrier coverage, two types of After the deployment, sensor nodes may fail due to many
deployment algorithms can be used, depending on whether factors, such as battery depletion, environmental conditions
sensor nodes are static or mobile. or malfunctioning. Then, a redeployment is needed to heal
coverage holes. The proposed algorithm proceeds in two
phases. In the first phase, it scans the network from the
beginning to the end of the barrier to check coverage holes. The
second phase consists in computing which sensor nodes should
move to which position such that the total distance traveled
by the nodes is minimized. This algorithm takes advantage of
the LNRO distribution as all sensor nodes are concentrated
a Optimal 1-barrier b The above zone is along a line, as depicted in Figure 19, allowing quick and easy
coverage. 2-barrier covered. replacement of failed nodes.
Figure 18: Barrier coverage.

5.1.1 Static sensor nodes

When sensor nodes are static, they are generally deployed
uniformly over the whole area based on a Poisson Point
Process model. Using this kind of deployment, barrier
coverage can be provided by selecting a chain of overlapping a Big object: 1-barrier b Small object:
sensor nodes. However, when static sensor nodes are dropped coverage. 2-barrier coverage.
by an aircraft, they will deviate from their expected location
due to mechanical inaccuracy or environmental factors such Figure 20: Dynamic object.
as wind, terrain characteristics, etc. To cope with this problem,
(50) proposes a concentrated deployment of sensor nodes
along the deployment line with some random offsets, using The monitored object may be dynamic, (i.e. changing its
for example aircraft (see Figure 19). This distribution is called shape). As a consequence, sensor nodes have to move to adapt
LNRO, Line based normal random offset distribution, and in the belts they form around the object to be monitored. In
terms of barrier coverage, it outperforms the Poisson model (52) the problem of mobile barrier for dynamic coverage is
when the random offset in LNRO is relatively small compared formulated as: for a given number n of sensor nodes, how
to r. do sensor nodes move with the objective to minimize the
18 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

total distance traveled under the constraint that the number of is not always met. In such a case, strategies to ensure
barriers is maximized at any time. Sensor nodes are placed connectivity should be provided.
around the dynamic object, neighboring sensors are at a Sensor nodes may be dropped randomly, trying to follow a
distance less than or equal to 2r forming a belt around the barrier line (e.g. (50)). In this case, coverage can be improved
dynamic object without any coverage holes. The authors by a centralized algorithm, as in (51) in charge of detecting and
assume that R ≥ 2r, in order to ensure full connectivity. A healing holes in barrier coverage. However, when coverage
dynamic belt region provides k-mobile barrier coverage if and holes are present, the central entity may fail to collect all sensor
only if there are k vertex disjoint belts in its coverage graph. nodes’ positions since these holes may produce disconnected
The maximum number of barriers k changes in response to components.
changes of the dynamic object, k becomes smaller when the
dynamic object becomes larger, as illustrated in Figure 20.
6 Point coverage and connectivity algorithms
5.2 Partial barrier coverage
The last type of coverage is given by the coverage of Points of
In the barrier coverage problem, the optimal number of nodes Interest (PoI). Examples of applications include the detection
(denoted m points) required to fully cover the barrier, can be of some static or moving target, using the smallest number of
determined based on the sensors’ sensing range and the barrier sensors. We distinguish between static PoIs and dynamic PoIs.
length. However, if the number of available nodes is less than
optimal, the barrier coverage problem will be formulated as 6.1 Static PoI
how to move n mobile sensor nodes to monitor n points among
the m points so as to maximize the average intruder detection In (47), the authors are interested in the deployment of
while minimizing the average sensor movement distance. To mobile sensors to cover predefined PoIs, while preserving
solve this problem, two algorithms PMS and CSP are proposed connectivity with the sink. The sink has the task of
in (53). PMS, periodic monitoring scheduling, lets sensor disseminating information about the PoI locations to the
nodes monitor each point of the barrier periodically, regardless sensors as well as collecting the information reported from
of any arrival by an intruder and without any coordination the sensors about the events happening at the PoI. The basic
between sensors. Each sensor moves to the point j and stays idea of this deployment algorithm for PoI coverage is as
there for T time slots. Then, it moves to point mod(j + n, m) follows: initially all the sensors are within radio range of the
and stays there, also for T time slots. This is repeated until sink. All the sensors run the same algorithm but the motion
all the sensors run out of energy. CSP, Coordinated Sensor decision is taken individually by each sensor node. The sensors
patrolling, is a centralized algorithm that uses the temporal move toward one predefined point that could be the PoI or
correlation of intruder arrival times. CSP runs in two steps. the barycenter of the PoIs. Then they form straight lines
Firstly, it selects the point with the highest priority of intruder between the PoI and the sink. The distance the sensors move
arrival to be monitored at the current time. Then, it determines is bounded in order to maintain connectivity. Finally a sensor
how to move sensors to the selected point while minimizing stops moving, when it covers the PoI (i.e. the PoI is in the
the total distance traveled, using the information collected in sensing range of the sensor). The strategy of this deployment
the past time slot. It has been shown that the CSP algorithm algorithm minimizes the number of sensors used to maintain
outperforms PMS. connectivity by using the RNG graph (Relative Neighborhood
Graph).
5.3 Summary If multiple PoIs exist in the area considered, two
approaches can be adopted:
Generally, the barrier coverage problem refers to critical
applications such as intruder detection which require special • Random PoI deployment: the sensor chooses one of the
attention. The high degree of robustness (multiple barrier PoI at random;
coverage) is generally chosen for critical applications to prove • Barycenter PoI deployment: Every sensor calculates the
the efficiency and reliability of the monitoring task. barycenter of all the PoIs and the sink to cover it. Then
Furthermore, the zone monitored, such as a battlefield or any sensor chooses a PoI at random and covers it.
country borders very often includes obstacles and is not always
flat in these applications. Many environment constraints In (44), a distributed deployment scheme is proposed
may occur, such as in a battlefield or international borders. where mobile sensors nodes move following concentric
Obstacles can also occur in the monitoring barrier. The circular paths (ferries with annular trajectories) that cover
solutions proposed in the literature do not take into account static PoIs (See Figure 21). The goal of this work is to ensure
these constraints which have a negative impact on the PoI coverage and that events are reported to the sink. This sink
deployment algorithm. is located at the barycenter. Two neighboring circular paths
The issue of connectivity is very important in critical are at a distance of R. The authors assume that R ≥ 2r and
applications since it allows information to be reported to mobile sensors have no global knowledge of the PoIs in the
the sink. All the papers cited in this section, assume that area considered. This work combines three aspects which are:
connectivity between neighboring nodes and with the sink is PoI discovery, PoI coverage and connectivity with the sink.
ensured: R ≥ 2r. However, in real deployments, this condition To achieve these three aspects, a mobile sensor should move
 Coverage and Connectivity Issues and Challenges in WSN 19

constantly to execute the PoI discovery task. Then, it should 6.2 Mobile PoIs
adjust its movement velocity with sensors in the neighboring
circular paths to satisfy the constraints regarding coverage and In the case of mobile PoIs, the authors of (47) propose three
connectivity with the sink in order to report the information strategies to reach the mobile PoI:
about the PoIs. • In the first strategy, sensor nodes move back to the
sink before deploying toward the new location of the
PoI. This strategy provides a high coverage quality but
increases the deployment duration and the amount of
energy consumed.
• In the second strategy, sensors try to move directly toward
the new location of the PoI without going back to the
sink. This strategy reduces the time needed to cover the
new PoI but also reduces the coverage quality as it needs
a greater number of sensors to maintain connectivity.
• In the third strategy, a sensor moves toward the straight
line between the sink and the new location of PoI, then
Figure 21: PoI coverage using annular ferries. it moves toward the PoI. This strategy provides a higher
coverage quality and reduces the time needed to cover
the PoI.

Temporary coverage of Multiple PoIs is studied in (45) 6.3 Summary

and is called the sweep coverage problem as sensor nodes
sweep between PoIs and cover them periodically. A distributed Any PoI needs only one sensor to be covered. If permanent
algorithm DSWEEP is proposed to address this problem. connectivity is required, a sufficient number of sensor nodes
A sensor node covers a PoI for a determined duration and are deployed to ensure connectivity with the sink. However,
then moves to a new one. When a sensor node is moving, it if intermittent connectivity is sufficient, one sensor node will
encounters other sensor nodes and exchanges information that cover a PoI, and a mobile node (that can be the sink or a
serves to decide which PoI should be monitored next. This collector Robot) will operate like a data mule. This can be a
deployment algorithm needs a small number of sensor nodes solution to deploy a minimum number of sensor nodes and
to cover a large number of PoIs. DSWEEP provides temporary save energy.
coverage and partial network connectivity. When the PoI is static, a static sensor node can be used to cover
In some applications, the PoI, as well as the area surrounding it. If the PoI is mobile, however, autonomous sensor nodes are
it need to be covered. In (46) a localized self deployment deployed to track the PoI and avoid the use of a robot that
algorithm is proposed to meet this goal. This algorithm
√ is based would pick up and deploy sensor nodes each time the position
on a virtual triangular lattice grid of edge√ 3r to maintain of the PoI changes.
connectivity since it is assumed that R ≥ 3r. Sensor nodes Notice that PoI coverage needs just one sensor to be covered,
are autonomous and know the position of the PoI. They move whereas a zone of interest requires at least one sensor to be
through the triangular vertexes and organize themselves by covered as the zone may be larger than the sensors’ sensing
respecting rules that avoid collisions between sensors, to reach range. When many sensor nodes are deployed to cover a
the vertices around the PoI. Based on this principle no coverage zone (area) of interest, they are usually deployed with various
holes will occur if all the vertexes around the PoI are occupied densities: high density in the center of the zone of interest and
by sensor nodes. then the density decreases with the distance to the center of
the zone.

7 Node activity scheduling with regard to coverage

Assuming an initial deployment of static sensor nodes meeting

the application requirements (e.g. full or partial coverage),
the node activity scheduling problem consists in determining
a connected set of active nodes to ensure the application
requirement. Only nodes in this set are active, the other nodes
are sleeping to save energy. Hence, the network lifetime is
a Initial deployment. b Final deployment.
maximized. The problem here is not to deploy the sensor
Figure 22: PoI coverage using Grid. nodes but only to select which sensor nodes will be active to
maximize coverage and connectivity. Figure 23 depicts an
example where blue sensor nodes are sleeping, while coverage
and connectivity are ensured by white sensor nodes in red.
20 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

and connectivity when R < 2r. SPAN is a connectivity

maintenance protocol. This protocol connects all active nodes
via a communication backbone, and connects inactive nodes
to at least one active node. Then, when R < 2r, network
connectivity is ensured.
Several other distributed protocols are proposed in (18), to
ensure area coverage with a low communication overhead. In
Figure 23: Node Activity Scheduling. these protocols sensor nodes select a waiting time for each
round and receive neighboring messages which are used to
compute the area coverage. If the sensing area of a sensor
We distinguish two categories of node activity scheduling node is not fully covered, the node should stay in an active
with regard to coverage: state during the current round and announce its state when its
waiting time expires.
7.1 Node activity scheduling based on message Sensing range and radio range may be different and they may
exchanges between neighbors also differ between sensor nodes. The authors in (19) adopt
this assumption and aim to minimize the number of active
Sensor nodes rely on message exchanges to decide which nodes in the region queried that is fully covered. Then, each
sensor nodes should be in an active state while others are sensor node should determine whether it switches to an active
sleeping, with the goal of ensuring full coverage and saving state to respond to the query request originating from the sink,
energy. This mechanism can either be centralized, where a based on information collected from its neighbors.
central entity collects all the nodes’ positions and assigns
a state (active or sleep) to each node, or distributed, where 7.2 Node activity scheduling based on implicit
neighboring sensor nodes exchange messages to decide which coordination
of them will be active while others are sleeping.
An example of a centralized algorithm is given in (17). This Implicit coordination algorithms are proposed to save
work is based on the construction of a connected subgraph of the energy of sensor nodes, assuming full coverage and
sensor nodes based on local information. It focuses on finding connectivity. Such algorithms are distributed and based on a
the smallest subset of sensor nodes that ensures full coverage grid. Each node knows from its position in the grid whether
of the monitored area while maintaining connectivity with the it must be active or it can sleep. An example is given in (15)
sink. for a square pattern and a hexagonal pattern: each sensor
Another centralized algorithm is proposed in (59) to build a node located in the vertex of the grid switches to the active
camera barrier from an initial arbitrary deployment of camera state, while other nodes are sleeping. Another example of a
sensors. The aim is to guarantee that each point of the barrier square pattern is given in by VFCSO, Virtual Force-Based
is fully covered visually. The method consists in building a Coverage Optimization Strategy (26). VFCSO is a dynamic
graph of nodes where each of them covers a small subregion, deployment algorithm that aims at ensuring full area coverage
and every two adjacent nodes are connected. The idea is to using a minimum number of sensor nodes while saving energy
select a path from one boundary to another such that the nodes consumption. In this work, the considered area is divided into
of the path are full-view covered. Only nodes belonging to that square cells with edges equal to r. Many sensor nodes may be
path are active. in the same square cell. Starting from a random deployment,
Node activity scheduling based on message exchanges is also the virtual forces strategy is applied by sensor nodes belonging
adopted to ensure partial coverage. In (57), a centralized to the same square cell. Only one node in each cell will be
algorithm is proposed to ensure partial coverage. It aims to active, the others should switch to the sleep state: the active
select the smallest number of nodes to monitor p-percent of node being the closest sensor node to the center of the cell with
the area. The authors also propose a distributed algorithm that the highest residual energy. Both full coverage and √ network
determines a set of nodes to cover p-percent of the considered connectivity are guaranteed in this work as R ≥ 5r.
area. The main idea of these two algorithms is to divide the
whole area into sub-regions and select specific nodes, while
respecting some criteria (for example, a starter node selects 8 Some considereations for practitioners
its furthest neighbor) in order to cover p-percent of each sub-
region. In this section we set out to help designers to select
CCP, Coverage Configuration Protocol (12) is a distributed a deployment algorithm that meets their application
algorithm based on message exchanges to provide the degree requirements, and give some recommendations.
of coverage required by applications when R ≥ 2r. In CCP, The two main questions that the designer has to consider are
according to information about its sensing neighbors, a sensor the following:
node can be in a sleep state to save energy, a listen state to
collect neighboring messages and decide its new state, or an • What does the application need in term of coverage and
active state to sense the environment. Without assuming that connectivity?
R ≥ 2r, CCP cannot guarantee network connectivity. In (12),
CCP is combined with SPAN (14) to achieve both coverage • Which assumptions and constraints are given?
 Coverage and Connectivity Issues and Challenges in WSN 21

In the following we discuss various ways to answer these assisted in their deployment by a mobile robot.
questions. − The sensing range r, the communication range R and
I Definition of the coverage and connectivity problem that the associated models: for more details see Section 3.2.
must be solved: Furthermore, the relationship between r and R will be
used to select the appropriate deployment algorithms in
• Coverage Table 3.
− If the goal is to monitor an area, then the problem − The number of sensor nodes must be sufficient to meet
concerns area coverage, which may be full or partial. the application requirements, otherwise the problem is
If it is to detect barrier crossing, the problem is barrier impossible to solve.
coverage, which, again, may be full or partial. − Energy: if sensor nodes are equipped with a battery,
If the goal is to track/monitor a target, the problem deals the deployment algorithm must be energy efficient.
with point of interest (PoI) coverage. The PoI may be
static or mobile. • The sink
− If coverage must be full and the degree of robustness It is in charge of collecting the data generated by the
required by the application is high, multiple coverage is sensor nodes deployed. It can be static or mobile. If the
needed, otherwise simple coverage is sufficient. sink is static, either it is connected to sensor nodes, or
− If long delays to detect an event are tolerated by the a mobile robot visits the disconnected sensor nodes to
application, the coverage of any point can be temporary. collect their data and report them to the sink. If the sink
Otherwise, it is permanent. is mobile, it moves to collect data.
• Connectivity I Recommendations:
− If short delays to report detected events to the sink
are required by the application, permanent connectivity • Coverage problem
must be ensured. Otherwise, intermittent connectivity is Depending on the application needs, the problem is an
sufficient. area, barrier or PoI (Point of interest) coverage problem.
− If the application needs a high degree of robustness,
multiple paths to the sink should be maintained. • Relationship between the transmission range R and the
Otherwise, a simple path sensing range r
The relationship between R and r influences
√ the choice
• Type of deployment of the solution. If for instance R ≥ 3r, it is sufficient
− depending on the application requirements, a uniform to solve the coverage problem to obtain connectivity as
and regular deployment should be provided, based on a a consequence of coverage. If the transmission range R
pattern (see Section 3.3). is strictly less than the sensing range r, a distributed
deployment would require a smaller target distance
I Assumptions and constraints
between sensor nodes than that required by full coverage
In most cases, the designer will be faced with multiple
of the area. Hence, a higher number of sensor nodes is
assumptions and constraints that must be taken into account
used, leading to a more expensive solution. If the designer
when selecting the appropriate deployment algorithm. These
has a small budget, he/she will prefer a centralized
include:
solution with a mobile robot/agent to deploy the sensor
• Environment nodes to their final position, and to collect data from
− The dimensions and position of the area, barrier or these nodes in the data gathering phase. Similarly, such a
PoI to cover should be provided in order to compute solution is also preferred when the application tolerates
the minimum number of sensor nodes required. If this delays (e.g. delay tolerant networks, ferries). In contrast,
number is large, the deployment algorithm must be a permanent path must exist from any sensor node to the
scalable. The initial topology influences the deployment sink. Additional sensor nodes are required to ensure this
algorithm, specially when some sensor nodes are permanent connectivity.
disconnected, or when they are all grouped together at an
entry point (see the discussion in Section 3.1). • Centralized versus Distributed solution
− The choice of the radio propagation model must Depending on the area/barrier size, a centralized
be compliant with the environment (e.g. free space or /distributed solution will be preferable. Indeed, if
confined) which may suffer from perturbations caused by the monitoring requires a high number of sensors,
other wireless networks (e.g. WiFi) or electronic devices a distributed solution is chosen because of its better
(e.g. microwaves), and may also contain obstacles. scalability, provided that the energy constraints are taken
− In the presence of obstacles, detection and get around into account, as discussed below. A centralized solution
techniques should be provided. requires that the central entity in charge of the deployment
computation has perfect knowledge of the positions of all
• Sensor nodes the sensor nodes. If the initial topology is disconnected,
− Mobility: sensor nodes may be mobile and a mobile robot is needed to collect the initial positions of
autonomous, and this condition is necessary for self- all disconnected nodes to compute the final deployment.
deployment. On the other hand, static nodes should be If all sensor nodes are static, the centralized solution is
22 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

the only possible one. A mobile robot is needed to deploy objects communicate through wired or wireless networks that
the sensor nodes to their final position. tend to use common internet protocols (e.g. IP).
The deployment algorithms presented in this survey deal only
• Energy constraints with 2D. With new applications in smart cities, sensors will
When sensor nodes are equipped with a battery of limited be deployed in 3D. For instance, sensors can be fixed on the
capacity, energy efficient techniques should be used. ground, as well as at different heights on buildings to form a
Special care must be given to node activity scheduling network.
that allows nodes to sleep for energy saving purposes. The radio propagation models are different and new
Another advantage of node activity scheduling is to deployment algorithms should be designed. With the
make the deployment adaptive to varying coverage introduction of UAVs (Unmanned Aerial Vehicles) and
requirements, ranging from full to partial. However, the MUAVs, (Micro UAVs), swarms of flying and communicating
energy consumed by nodes movements is considerable UAVs and MUAVs will be used in monitoring applications.
and should be limited. For instance, nodes oscillations The main characteristics of such networks are the coordinated
occurring in some distributed solutions should be moves to deploy and redeploy in a 3D space to meet the
avoided. If the designer wants to keep the energy of sensor application requirements. Similarly to 3D deployments in the
nodes for data gathering, a mobile robot/agent should be air, exist 3D underwater deployments based on ultrasound
used to deploy the sensor nodes to their final position. communication. Both 2D and 3D deployment algorithms
• Uniform and regular deployment have to cope with positioning errors, which are common in
A uniform and regular deployment reduces the energy positioning systems like GPS. Few algorithms take this issue
consumed during the data gathering phase and minimizes into account.
the data gathering delay. Moreover, it provides better time Security is an important issue in deployment algorithms, but is
and space consistency of the measures reported to the usually not addressed. An intruder could harm the deployment
sink. process and corrupt the data gathering.
We expect that wireless sensor network technologies will be
• Obstacles mature very soon and will be widely deployed in a large variety
An area/barrier with obstacles needs mechanisms to of industrial applications requiring multihop communications
detect obstacles and strategies to get around them, as well and needing tight constraints. In these applications, specific
as ensuring the required coverage. attention should be paid to robustness. The deployment
algorithms should be able to tolerate some message losses
and some failed sensor nodes. Furthermore, the deployment
9 Trends and open issues for deployment algorithms algorithms should be able to adapt to the dynamic changes of
the application requirements during the monitoring task.
Deployment is a fundamental issue in WSNs. Many challenges Deployment algorithms need to be smarter in adapting
have been studied and published in the literature. In this survey, themselves to changes in the environment or the application
we have discussed and classified some of them according to requirements that could be more frequent.
their main characteristics with a special focus on coverage and
connectivity.
Nevertheless, several issues are still unresolved or under 10 Conclusion
study to achieve an optimized deployment that can be
adapted to different situations and emerging technologies. We In this survey, we studied two major challenges in
believe that the sensors’ devices will become more powerful WSNs: coverage and connectivity, while bearing in mind
with enhanced capacities, like more sophisticated processing, the importance of energy issue. We provided indications
greater sensing range, fully equipped with positioning system for analyzing deployment algorithms and evaluating their
and integrating more and more various sensing components performances. We distinguish two types of deployment
as well as actuators. Some of them could even embed several algorithms depending on the mobility of sensor nodes:
communication technologies (802.15.4, WiFi, 3G, 4G). self deployment for mobile sensor nodes and assisted
In short, sensor devices are becoming faster, more intelligent deployment for static sensor nodes deployed by mobile
and heterogeneous. However, their autonomy is still limited robots. Deployment algorithms are designed to meet the
by their embedded energy. Pushing back this boundary, energy application requirements such as coverage, connectivity,
harvesting and renewable energy are promising techniques to latency and robustness. We established a classification of
extend the lifetime of a battery and hence make them more deployment algorithms based on these requirements. We
autonomous. provide several recapitulative tables to help the reader gain a
Since sensor devices are becoming cheaper, they can be better understanding of the advantages and shortcomings of
deployed on a large scale to build more complex wireless the problems and algorithms studied. In fact, the deployment
networks offering more functionalities. To provide scalability, of sensor nodes and sinks can be considered as the first
distributed algorithms are more suitable than the centralized step in the design of a data gathering application. As a
ones. second step, node activity scheduling is used to optimize
These trends will increase with the emergence of the Internet energy consumption by switching off redundant nodes to
of Things, where sensors and actuators embedded in physical maximize network lifetime, while ensuring the coverage and
 Coverage and Connectivity Issues and Challenges in WSN 23

connectivity required by the application. It should be noted [12] X. Wang, Y. Zhang, C. Lu, R. Pless and C. Gill (2003) ’Integred
that the deployment provided determines the consumption of Coverage and Connectivity Configuration in Wireless Sensor
energy and bandwidth during the data gathering. The reader Networks’, SenSys, Embedded networked sensor systems, p. 28-
will find useful guidelines to select the deployment algorithms, 39.
that are best-suited to his/her needs. The deployment of [13] X. Bai, S. Kumar, D. Xuan, Z. Yun and T.H. Lai (2006)
wireless sensor nodes to monitor a temporary worksite is ’Deploying Wireless Sensors to Achieve Both Coverage and
Connectivity’, MobiHoc, Mobile ad hoc networking and
a typical use case representative of industrial applications
computing, p. 131-142.
where full coverage and full connectivity are required.
[14] B. Chen, K. Jamieson, H. Balakrishnan and R. Morris
In precision farming, for instance, partial coverage with
(2001) ’Span: An Energy Efficient Coordination Algorithm for
intermittent connectivity are sufficient to meet the application Topology Maintenance in Ad Hoc Wireless Networks’, ACM
requirements. The reader is also given an overview of Wireless Networks Journal.
current trends and some interesting open issues. We strongly [15] R. Lyengar, K. Kar and S. Banerjee (2005) ’Low-coordination
believe that the deployment algorithms will undergo extensive Topologies for Redundancy In Sensor Networks’, Mobihoc.
development with the rapid emergence of the Internet of [16] K. Kar and S. Banerjee (2003) ’Node Placement for
Things. Connected Coverage in Sensor Networks’, WiOpt Modeling and
Optimization in Mobile, Ad Hoc and Wireless Networks.
[17] J. Carle and D. Simplot Ryl (2004) ’Energy-Efficient Area
Acknowledgment Monitoring for Sensor Networks’, IEEE Computer Society.
[18] A. Gallais, J. Carle, D. Simplot Ryl and I. Stojmenovic (2008)
This work has been partially funded by the Cluster Connexion ’Localized Sensor Area Coverage with Low Communication
Overhead’, IEEE Transaction on mobile computing, vol. 7, no.
project. For more details see http://www.cluster-connexion.fr/
5.
We would like to thank Richard James for his help to improve
[19] J. P. Sheu, S. C. Tu and C. H. Yu (2007) ’A Distributed
the quality of this survey.
Query Protocol in Wireless Sensor Networks’, Wireless Personal
Communication, p. 449-464.
[20] Y. C. Wang, C. C. Hu and Y. C. Tseng (2005) ’Efficient
References Deployment Algorithms for Ensuring Coverage and connectivity
of Wireless Sensor Networks’, WICON.
[1] Y. Miao (2005) ’Applications of sensor networks’, Seminar [21] Y. Zou and K. Chakrabarty (2003) ’Sensor Deployment and
Wireless Self-Organization Networks. Target Localization Based on Virtual Forces’ IEEE INFOCOM.
[2] R. Mulligan and H. M. Ammari (2010) ’Coverage in [22] K. Mougou, S. Mahfoudh, P. Minet and A. Laouiti (2012)
Wireless Sensor Networks: A Survey’, Networks protocols and ’Redeployment of Randomly Deployed Wireless Mobile Sensor
algorithms, vol. 2, no. 2. Nodes’ IEEE VTC, Quebec, Canada.
[3] A. Ghosh and S. K. Das (2008) ’Coverage and Connectivity [23] S. Mahfoudh, I. Khoufi P. Minet and A. Laouiti (2013)
Issues in Wireless Sensor Networks: A survey’, Pervasive and ’Relocation of Mobile Wireless Sensors in the Presence of
Mobile Computing, vol. 4, I. 3, p. 303-334. Obstacles’, ICT, CasaBlanca, Marroco.
[24] J. Li, B. Zhang, L. Cui and S. Chai, (2012) ’An Extended
[4] G. Fan and S. Jin (2010) ’Coverage Problem in Wireless Sensor
Virtual Force-Based Approach to Distributed Self-Deployment in
Networks: A Survey’, Journal of networks, vol. 5, no 9.
Mobile Sensor Networks’, International Journal of Distributed
[5] B. Wang (2011) ’Coverage Problem in Sensor Networks: A Sensor Networks.
Survey’, ACM Computing Surveys, vol. 43 no.4, article 32. [25] J. Chen, S. Li, and Y. Sun (2007) ’Novel Deployment Schemes
[6] C. Zhu, C. Zheng, L. Shu and G. Han (2012) ’A survey on for Mobile Sensor Networks’ Sensors, vol. 7, no. 11, pp. 2907-
coverage and connectivity issues in wireless sensor networks’, 2919.
Journal of Network and Computer Applications, p. 619-632. [26] Q. Xu and Q. Wang (2012) ’Coverage Optimization
[7] S. Mahfoudh, P. Minet and A. Laouiti (2012) ’Overview of Deployment Based on virtual Force Directed in Wireless Sensor
Deployment and Redeployment Algorithms for Mobile Wireless Networks’, IPCSIT, vol. 47, Singapore.
Sensor Networks’, Emerging Topic on Sensor Networks. [27] G. Tan, S. A. Jarvis, A. M. Kermarrec (2009) ’Connectivity-
[8] M. C. Akewar and N. V. Thkur (2012) "Study of Wireless Mobile Guaranteed and Obstacle-Adaptive Deployment Schemes for
Sensor Network Deployment" Internatioal Journal of Computer Mobile Sensor Networks’, IEEE Transactions on Mobile
and Wireless Communication, vol. 2, no 4. Computing.
[28] N. Bartolini, T. Calamoneri, E. G. Fusco, A. Massini and S.
[9] R. Soua, P. Minet, E. Livolant (2012) ’MODESA: an Optimized
Silvestri (2010) ’Autonomous deployment of mobile sensors for
Multichannel Slot Assignment for Raw Data Convergecast in
a complete coverage’, Journal Wireless Networks, vol. 16, I. 3,
Wireless Sensor Networks’, IPCCC 2012, the 31st IEEE
p. 607-625.
International Performance Computing and Communications
[29] Y. H. Kim, C. M. Kim, D. S. Yang, Y. J. Ohand Y. H. Han
Conference, Austin, Texas.
(2012) ’Regular Sensor Deployment Patterns for p-Coverage and
[10] H. Wang and W. Chung (2012) ’The generalized k-coverage q-Connectivity in Wireless Sensor Networks’, ICOIN.
under probabilistic sensing model in sensor network’ WCNC. [30] Z. Yun, X. Bai, D. Xuan, T. H. Lai and W. Jia (2010) ’Optimal
[11] H. Zhang and J. C. Hou (2005) ’Maintaining Sensing Coverage Deployment Patterns for Full Coverage and k-Connectivity (k ≤
and Connectivity in Large Sensor Networks’ Ad Hoc and Sensor 6) Wireless Sensor Networks’, IEEE Tansactions on Networking,
Wireless Networks, vol. 1, no. 1-2. vol. 18, no. 3.
24 I. Khoufi, P. Minet, A. Laouiti and S. Mahfoudh

[31] N. A. Ab. Aziz, K. Ab. Aziz, and W. Z. W. Ismail (2009) [48] S. Kumar, T. H. Lai and A. Arora (2005) ’Barrier Coverage with
’Coverage Strategies for Wireless Sensor Networks’, World Wireless Sensor’, MobiCom, Germany.
Academy of Science Engineering and Technology. [49] A. Saipulla, B. Liu, G. Xing and J. Wang (2010) ’Barrier
[32] S. Han, Y. Zhang and G. Xu (2010) ’Hexagonal grid- Coverage with sensor of limited mobility’, MobiHoc, USA.
based sensor deployment algorithm’, Control and Decision [50] A. Saipulla, C. Westphal, B. Liu, J. Wang (2009)
Conference, Chinese. ’Barrier Coverage of Line-Based Deployment Wireless Sensor
[33] P. Park, S. G. Min and Y. H. Han (2010) ’A Grid-based Self- Networks’, INFOCOM, Rio de Janeiro.
deployment Schemes in Mobile Sensor Networks’. [51] A. Saipulla, C. Westphal, B. Liu and J. Wang (2013) ’Barrier
[34] A. Mateska and L. Gavrilovska (2011) ’WSN Coverage and coverage with line-based deployed mobile sensors’ Journal Ad
Connectivity Improvement Utilizing Sensor Mobility’, European Hoc Network.
Wireless, Vienna, Austria. [52] L. Kong, Y. Zhu, M. Y. Wu and W. Shu (2012) ’Mobile Barrier
[35] L. Xu, A. Nayak and I. Stojmenovic (2010) ’Back Tracking Coverage for Dynamic Objects in Wireless Sensor Networks’,
based Sensor Deployment by a Robot Team’, Sensor Mesh and Mobile Adhoc and Sensor Systems MASS, Las Vegas, NV, USA.
Ad Hoc Communications and Networks, USA. [53] H. Shibo, C. Jiming, L. Xu, S. Xuemin and S. Youxian (2012)
[36] K. Chan-Myung, Y. Dong-Sun, O. Young-jun and H. Youn- ’Cost-Effective barrier coverage by mobile Sensor Network’,
Hee (2012) ’Regular Sensor Deployment Patterns for p-coverage IEEE INFOCOM.
and q-connectivity in Wireless Sensor Network’,Information [54] E. M. Saad, M. H. Awadalla and R. R. Darwish (2008) ’A Data
Networking ICOIN, 2012. Gathering Algorithm for a Mobile Sink in Large-Scale Sensor
[37] Y. C. Wang and Y. C. Tseng (2007) ’Distributed Deployment Networks’, Wireless and Mobile Communications, ICWMC,
Schemes for Mobile Wireless Sensor Networks to Ensure Multi- Athens.
level Coverage’ IEEE Transactions On Parallel And Distributed [55] V. Kavitha and E. Altman (2010) ’Analysis and Design of
Systems. Message Ferry Routes in Sensor Networks using Polling Models’
[38] A. Raha, S. Maity, M. K. Naskar, O. Alfandi and D. Hogrefe Modeling and Optimization in Mobile, Ad Hoc and Wireless
(2012) ’An Optimal Sensor Deployment Scheme to Ensure Multi Networks (WiOpt), Avignon.
Level Coverage and Connectivity in Wireless Sensor Networks’, [56] Y. Liu and W. Liang (2005) ’Approximate coverage in
Wireless Communications and Mobile Computing Conference wireless sensor networks’, IEEE Conference on Local Computer
IWCMC, Limassol. Networks, Sydney.
[39] J. Beaudaux, A. Gallais and T. Razafindralambo (2010) [57] Y. Wu, C. Ai, S. Gao, and Y. Li (2008) ’p-Percent Coverage in
’Multiple Coverage with Controlled Connectivity in Wireless Wireless Sensor Networks’, WASA, Berlin.
Sensor Networks’, ACM workshop on Performance evaluation [58] M. Hefeeda and M. Bagheri (2007) ’Wireless Sensor Networks
of wireless ad hoc sensor and ubiquitous networks. for Early Detection of Forest Fires’, Mobile Adhoc and Sensor
[40] G. Wang, G. Cao and T. F. La Porta (2006) ’Movement Assisted Systems, Pisa.
Sensor Deployment’, IEEE Transactions on Mobile Computing, [59] Y. Wang and G. Cao (2011) ’Barrier Coverage in Camera Sensor
vol 5, no 6. Networks’, MobiHoc, Paris.
[41] S. Babaie and S. S. Pirahesh (2012) (Hole Detection
Fore Increasing Coverage in Wireless Sensor Network Using
Triangular Structure’, International Journal of Computer Science
Issues IJCSI, vol. 9, I. 1, no 2.
[42] M.E. Keskin, I. K. Altmel, N. Aras, C. Ersoy (2013)
’Optimal Deployment, Scheduling and Routing for Maximizing
the Lifetime of a Wireless Sensor Networks With Multiple
Mobile Sinks’, Technical Report FBE-IE-02/2013-02, Institute
of Graduate Science and Engineering, Bogazici University,
Istanbul.
[43] H. Tan, Y. Wang, X. Hao, Q. S. Hua and F. C.M. Lau (2010)
’Arbitrary obstacles constrained full coverage in wireless sensor
networks’ In Wireless Algorithms Systems and Applications,
Springer Berlin Heidelberg.
[44] M. Erdelj, E. Natalizio and T. Razafindralambo (2012) ’Multiple
Point of Interest Discovery and Coverage with Mobile Wireless
Sensors’ Computing Networking and Communications ICNC.
[45] W. Cheng, M. Li, K. Liu, Y. Liu, X. Li and X. Liao (2011)
’Sweep Coverage with Mobile Sensors’, Mobile Computing
IEEE transactions, vol. 10, I. 11, November 2011.
[46] X. Li, H. Frey, N. Santoro and I. Stojmenovic (2011)
’Strictly Localized Sensor Self-Deployment for Optimal Focused
Coverage’, IEEE Transactions on Mobile Computing, Vol. 10,
No. 11.
[47] M. Erdelj, T. Razafindralambo and D. Simplot Ryl (2013)
’Covering Points of Interest with Mobile Sensors’, IEEE
Transaction on Parallel and Distributed Systems, p. 32-43.