Está en la página 1de 6

Critical Density Thresholds for Coverage in Wireless

Sensor Networks
Sachin Adlakha Mani Srivastava

Networked and Embedded Systems Lab (NESL), Electrical Engineering Department, UCLA
56-125B Eng. IV, UCLA-EE Dept., Los Angeles, CA 90095
{sachin, mbs}@ee.ucla.edu

Abstract- Sensor networks are deployed to monitor the physical
world and to provide relevant data to the users. An important
question in such networks is to estimate the number of sensors
required to achieve complete coverage of the desired region. The
number of sensors required would depend upon the physical
characteristics of the individual sensors as well as the nature of
the target. In this paper, we address the problem of finding the
critical density of sensors for complete coverage. We use an
exposure-based model to find the number of sensors required to
cover an area for given sensor and target characteristics. The
accuracy of the results is established via simulations.

Keywords: sensor networks, critical density, coverage

I. INTRODUCTION
A. Motivation

Sensor networks have generated flurry of research activity
because of variety of applications [1], [3]. Some of the
applications envisioned for such networks include but are not
limited to target detection, object tracking, environment
surveillance, intrusion detection etc. Typically such networks
consists of highly energy constrained nodes deployed over a
region, where they self organize [2] and monitor the region
for parameters of interest.
Sensor networks have two major aspects: Sensing and
Communications. The nodes sense the environment and
communicate the result to the user. A typical scenario for
these types of networks is a battlefield, where nodes are
randomly deployed to detect enemy movement. Upon
detection, nodes transmit the information to the user via
multi-hop communication. An important question in such
scenarios is to determine the number of sensors to be
deployed so that the entire area is covered and probability of
detection is high. Deploying small number of nodes might
leave blind spots or sensing holes, which can allow the enemy
to pass through. Thus knowing the sensing capacity as a
function of number of nodes to be deployed is crucial for
design of sensor networks. Density of nodes is also a crucial
parameter in scenarios where network is deployed to monitor
environmental variables. Leaving blind spots in such cases
can reduce the accuracy of the results obtained.
In the scenarios described above, typically the target is a
signal source, and the nodes receive the signal via a channel.
Depending upon the strength of the signal received the node
detects the target. Thus the sensing capacity of the sensor
network would depend upon the target characteristics as well
as sensor sensitivity and calibration. For example to detect an
enemy tank might require less number of nodes because the
acoustic signal emitted by them is stronger. However
detecting personnel movement might require more sensors
because the signal strength is low. Thus the density evaluation
must take into account the nature and characteristics of both
the sensor as well as the target.

B. Contribution

In this paper we present an analytical result for
approximate number of sensors required for coverage. We
evaluate the radius of influence of the sensor nodes depending
upon their physical characteristics and also the target
properties. Using the radius of influence of each sensor we
estimate the total number of sensor nodes required to cover
the area. We assume that the nodes are randomly deployed
over the area. We also evaluate the variation in density for
varying sensor and target characteristics. This enables us to do
complete network design knowing the sensor characteristics
and the target specifications.

C. Paper Organization

The remainder of the paper is organized as follows. We
first survey the related work. Section III gives the models and
various definitions used to evaluate the critical density. Next
we analytically evaluate the region of influence for a specific
case of a single sensor located at origin and the target moving
in a straight line and use this to find the critical number of
sensors required. In section V, we present the simulation
results to verify our results and extend them to a more general
case. Section VI concludes the paper.

II. RELATED WORK
Sensor networks have elicited immense interest in past
few years. From a communication perspective, the authors of
[6] give the capacity of wireless networks. However no work
has been done on the sensing capacity in terms of number of
nodes required for sensor networks. In [4] authors introduced
the idea of coverage, which they defined as quality of service.
They used graph theoretic techniques to evaluate the maximal
breach path. In [5] the authors extended the idea of coverage
to define exposure and also evaluated it for a single node case.
0-7803-7700-1/03/$17.00 (C) 2003 IEEE 1615

Some work in finding the number of disks required to cover
given set of point has been done in [7]. However, their work
considers grid-based placement of nodes that is not the case in
typical sensor networks. None of the above work takes into
account the specific target characteristic in evaluating the
probability of detection for random deployment of nodes or
their dependence on node characteristics.

III. DEFINITIONS AND MODEL
A. Sensor Model

Sensor nodes have a variety of physical characteristics. For
most types of sensors the ability to detect decreases as the
distance from the target increases. Thus for a sensor s, the
signal received from the target located at point p is given as
[5]
k
p s
p s S

=

) , ( (1)
Here the constant depends on the sensor calibration and
p s is the Euclidean distance between the sensor s and
the target p. The parameter k gives the signal decay factor and
for most types of sensors k = 2. Since each sensor requires
certain signal to noise ratio (SNR) to detect the signal, beyond
a certain noise figure F, the signal would not be detected. This
is because the signal strength would fall below the noise floor.
Thus if S(s, p) < F, the sensor would not detect the signal.
While the sensor models describe the strength of the signal
received at the sensor node, the actual detection criteria is
dependent on the exposure models used. In next subsection
we describe various exposure models.

B. Exposure Model

Depending upon the type of detection and the application,
several exposure models can be constructed. Note that the
exposure determines how the signal received from the target
is processed to make a decision. For example an acoustic
sensor can sense the target for a fixed period of time, integrate
the acoustic energy and if the energy exceeds the threshold, it
declared that the target is detected. This kind of exposure
model was first introduced in [5]. Another model is where the
sensor detects the change in the signal strength over time.
Thus depending upon the way a sensor detects we have two
different models:
1. Integrator model/capacitive model: This model
pertains to sensors that are energy detectors. Thus when the
total signal energy or exposure (which is the total signal
strength over the path) exceeds a threshold, the sensor
declared the target as detected. Mathematically we have:

=
2
1
) ( (
) , (
t
t
s
dt
dt
t p d
p s S E (2)
Here, p(t) is the path of the target from starting point a to
ending point b and the sensor model S(s,p) can either be
speed dependent or independent. Equation (1) gives the
sensor model for speed independent case.
2. Derivative/spike detector model: This models those
sensors that detect the change in the signal strength and if the
change is above a threshold a success/detection is declared.
For example Magnetic sensor detect a target if there is a
significant change in the magnetic field. Thus mathematically
exposure in this case can be defined as:
dt
p s S d
E
s
)) , ( (
= (3)

C. Detection Model

To detect an event each sensor, uses one of the above
exposure model and uses that exposure to make a decision.
Two different types of detection model can be used depending
on if each sensor individually makes the decision or sensors
mutually cooperate to decide about an event.
1. Individual Detection: In this case, detection occurs if
threhsold s
E E . This type of detection is referred to as
Individual detection because each sensor separately detects
and reports the event.
2. Cooperative Detection: To reduce the possibility of
false alarm and to improve the reliability of decisions, a
cooperative detection mechanism can be used. For
cooperative detection we define { }
N
s s
E
1
~
=
to be a set of binary
random variable such that
threshold s s
E E where E =1
~
and
otherwise E
s
0
~
= (4)
Then we say an event is detected if K E
s
s

~
where K is
the system parameter. This mechanism reduces the false alarm
probability by requiring an event to be at least detected by K
sensor nodes.

D. Target Characteristics:

Given the diverse set of applications envisioned for sensor
networks, variety of targets models can exist. For example in
a sensor network deployed to monitor wild brush fires, the
target or parameter of interest is occurrence of wild fire.
However in other cases, such as in sensor networks deployed
in a battlefield, a target can be a moving vehicle or personnel
movement. In such cases, we can characterize the target as an
object moving with constant speed v from point a to point
b located distance apart. In this paper we consider such
targets. We also assume that the target moves in a straight-
line path with constant speed for a distance of . The
straight-line path described in this paper is a special case of a
general path. Also any adversary unaware of exact sensor
locations would take the straight-line path since it is the
shortest distance to its destination. Given the target and sensor
characteristics, we can define regions of influence of the
1616

sensor. Below we define two radii associated with a given
sensor.
Definition: Radius of complete influence (
1
r ) for a
particular sensor (having given sensor and exposure model
and using a particular detection criterion) is defined as the
distance from the sensor such that all targets originating
within this radius are surely detected.
Definition: Radius of no Influence (
2
r ) for a particular
sensor (having given sensor and exposure model, and using a
particular detection criterion) is defined as the distance from
the sensor, such that the sensor cannot detect any target
originating beyond this radius.
Note that it is possible that
1
r =
2
r , however as we shall
see later in some case there might exist region where the
probability of detection is between 0 and 1, and hence
1
r is
different from
2
r .

IV. FINDING THE RADII OF INFLUENCE AND CRITICAL
DENSITY
In this section we derive analytical result for radii of
influence and critical density of nodes required for complete
detection. We begin with a simple case of a single sensor
located at origin. The sensor model is same as (1) with path
loss or decay factor k as 2. This is a typical case for many
sensors and hence represents a realistic scenario. The
exposure model taken is the integrator type model, which is
common for many sensors such as ultrasound, acoustic etc.
The target is an object that moves in a straight-line path with
speed v and travels a distance . This scenario represents a
battlefield case, where sensors (typically acoustic sensors) are
deployed to detect enemy movement. The enemy moves with
some particular speed and travels some distance. Since the
target/enemy is oblivious of sensor position, we require that
any straight-line path of distance should be detected.
Because the exposure is independent of orientation of the
axis, we assume the straight-line path is parallel to x-axis for
analysis purposes only. By symmetry of equations the results
would be valid for any randomly oriented path. We later
verify this via simulations. Each sensor has a noise figure F
and the threshold as
threshold
E . To evaluate the radii of
influence we evaluate the exposure at the sensor located at
origin. The target is initially located at ( )
0 0
, y x and moves
away from the sensor with speed v. Again we focus our
attention on the first quadrant, since by symmetry the results
hold for other quadrants also. This case is depicted in figure 1.
Let v T / = be the total time it takes for the target to
reach its final destination. Using the sensor model (1), we get
that the signal strength received at the sensor at any time t is
given as:
2 2
) ( ) (
))) ( ), ( ( ), 0 , 0 ((
t y t x
t y t x S
+
=

(5)
Since the target moves parallel to X-axis we have
0
) ( y t y =
and t v x t x * ) (
0
+ = . Using these in (5) we get
2
0
2
0
) ( ) * (
))) ( ), ( ( ), 0 , 0 ((
y t v x
t y t x S
+ +
=

(6)
Using the integrator model the total exposure is given as:

+ +
=
T
s
dt
y t v x
E
0
2
0
2
0
) ( ) * (

(7)
Using the substitution t v x x *
0
+ = and dt v dx * = we
transform the above equation into

+
+
=

0
0
2
0
2
x
x
s
v
dx
y x
E (8)
which can be easily evaluated as

+ +
=


0
2
0
2
0
0 1
0
tan
x y x
y
vy
E
s
(9)


Figure 1: Finding Region of Influence

Equation (9) gives the exposure of the path shown in figure 1.
Using polar coordinates the exposure can be written as

+
=

) cos(
) sin(
tan
) sin(
1

r vr
E
s
(10)
Here r is the initial distance of the target from the sensor and
is its orientation from X-axis. We first evaluate some
special cases of interest.
Case1: = 90
o
Substituting = 90
o
we get:

=

r vr
E
s

1
tan (11)
Case2: = 0
o
For this case we have indeterminate form. So
we use L Hospitals rule to find

+
=

) cos(
) sin(
tan
) sin(
1
0
lim

r vr
E
s
(12)
Differentiating numerator and denominator, and substituting
= 0
o
we get

+
=
r vr
E
s


(13)
1617

It can be shown that E
s
has a minimum at = 0
o
and has
maximum for = 90
0
.for a given value of r. Figure 3 shows a
plot of E
s
for some typical values ( = 0.1, = 10m, v =
2m/sec). Thus for all paths originating at radius r, the one
tangent to the circle at = 90
0
has maximum E
s
. For = 90
0
,
E
s
is given by (11).
To find the path of maximum exposure we consider the case
where = 90
0
. Since ) ( tan
1
x

is a concave function of x, the


exposure for case 1 in figure 4 is greater than that in case 2.
Note that for case 1 the exposure is sum of exposures for two
paths of length x and - x. The exposure is thus given as:

=

r
x
vr r
x
vr
x E
s

1 1
tan tan ) ( (14)


Figure 2: Calculating the exposure for different angles

Differentiating (15) and setting it to 0, we get that the
maximum E
s
occurs for
2

= x . The value of maximum E


s
is
thus given as:

=

r vr
E
s
2
tan
2
max
1

(15)

Figure 3: Exposure as a function of

Thus the maximum exposure for a path (with minimum
distance of r from the sensor) occurs for a path which is
tangent to circle at = 90
0
and is symmetric about Y-axis i.e.
2

= x ). To evaluate the
1
r , we equate the minimum E
s
to
E
threshold.
This ensures that all paths with their initial point
lying within this r would have exposure greater than the
threshold. The region of no influence is found by equating the
maximum E
s
to E
threshold.
. Thus we have the following lemma
Lemma 1. For a sensor located at origin and having a
sensor model as in (1) and integrator type exposure model, the
radius of complete influence (
1
r ) is given by the following
equation

threshold
E
r r v
=

+
1 1


(16)
and the radius of no influence is given as:
threshold
E
r r v
=

2
1
2
2
tan

2
(17)
Here E
threshold
is the threshold value used in individual
detection, v is the speed of the target and is the length of the
path.


Figure 4: Finding the Maximum Es
Note that
2
r represents the upper limit on the sensing
region of a given sensor. The sensor would not detect any
path lying outside this radius. On the other hand
1
r represents
the area where all paths would be detected. To cover an area A
the number of nodes required would be of the order

2
r
A
where r is the sensing radius used. Here r lies
between
1
r and
2
r . Note that using
1
r overestimates the total
number of sensors required, while using
2
r can underestimate
it. We use simulations to show that using
2
r is a good
estimate for finding the number of sensors required. In fact
simulations results show that for number of sensors calculated
using the value of
2
r , the probability of detection is 98% or
above. Thus we have the following result
Result 1: For random deployment of sensor having sensor
model of (1) and using exposure based individual detection,
1618

the number of sensors required to cover a given area A is of
the order

2
2
r
A
, where
2
r is given as in (17).
In next section we verify the above result using
simulations. We also extend our result for cooperative case.
Intuitively, for cooperative case, we would require more
number of sensors. As we shall see later, the number of
sensors increases linearly with K, where K is the number of
sensor detecting the target in cooperative case.

V. SIMULATIONS
We verify our result using simulations, which were
performed in MATLAB. We consider random uniform
deployment of nodes over a square area of length L = 150m.
The sensor model is as given in (1) with k=2. We later extend
our results to cases k=3 and k=4. Figure 5 gives the
probability of detection with varying number of nodes. For
this case we choose = 0.1 in (1). The target is an object
which moves with speed v=2m/sec and travels a distance =
10 m and it emits the signal of unity strength. The sensor have
threshold 005 . 0 =
threshold
E and noise figure F = 0.0001.
For these values of sensor and target characteristics we have
6 . 9
2
= r and number of sensors N = 244. The probability of
detection is average over 1000 different network topologies.
As can be seen from the figure the probability of detection
approaches 1, for N= 250. This verifies our claim that
2
r is a
good measure for estimating the number of sensors required.
The probability of detection is plotted for both randomly
oriented path and straight-line paths parallel to X-axis. As can
be seen from the figure the two match very closely. This
justifies our assumption of using path parallel to X-axis for
analysis purposes. Thus result 1 holds for any randomly
oriented path. All of the following simulations are done for
any randomly oriented straight-line path.


Figure 5: Probability of Detection Vs. Number of Nodes

We next verify our result for different target and sensor
characteristics. Figure 6 gives compares the theoretical and
simulation results for different target velocities. Three
different cases for target distance = 10m, = 10m and =
10m, are considered. The sensor characteristics in this case are
= 0.1 and 005 . 0 =
threshold
E . The simulation results
correspond to detection probability of 98% or more. As can
be seen from the figure the theoretical results match closely
with the simulation analysis. Thus given the target
characteristics we can estimate approximate number of sensor
required using result 1. The result is also intuitive since
increasing the target speed would decrease the exposure and
hence more sensors would be required. However if the target
travels more distance, it spends more time and hence its
exposure increases.


Figure 6: Variation of Number of Nodes with Target Speed

Similar results are also obtained for different sensor
characteristics. Figure 7 compares the theoretical and
simulation result for different values of sensor parameters =
0.05, = 0.1 and = 0.2 and for varying threshold. The target
speed is fixed at v=2m/sec and target travels a fixed distance
of 15m. Simulation results correspond to 98% or more
detection. Thus knowing sensor and target characteristics, we
can estimate the number of sensors required using result 1.
We also extended our result to cooperative case and also
for different decay parameters. Figure 8 gives number of
nodes required with varying K, which is the number of nodes
required to detect the target. As seen from the figure the
number of nodes required grows linearly with K. The results
were plotted for = 0.1, target speed v = 2m/sec,
005 . 0 =
threshold
E and for different values of target distance
= 10m, = 10m and = 10m. Similar results were also
obtained for varying threshold and target speed but are
omitted for brevity.
Figure 9 gives the number of nodes required for different
decay parameters k=3 and k=4. It is intuitive that for larger
decay factor the number of nodes required would be more
than in case of k=2. This is because the signal strength
1619

received at the sensor would decease rapidly and hence the
exposure would be small. This would reduce the effective
radius of influence and hence would increase the density of
nodes required.
VI. CONCLUSIONS
In this paper, we evaluated the critical number of nodes
required for target detection in a sensor network. We used
physical characteristics of sensors and target to derive an
equation for effective sensor radius. Using this effective
radius we estimated the critical density for coverage in sensor
network. We also studied the variation of density with
different sensor and target parameters, and also extended our
results to case of cooperative detection and also to different
signal decay factor. Our main contribution is incorporating
physical characteristics of sensor and target in evaluating the
sensing capacity of sensor networks. Such modeling enables
sensor network design, where the user can decide the density
of nodes to be used depending upon the target characteristics
it is trying to detect as well the nature of sensor deployed. As
part of our future work, we would extend this result to case of
more detailed sensor models, where the signal strength
received at the sensor is also function of target speed. The
straight-line case described in this paper is a special case of a
general path and we hope to extend the result to a general
target path. However the scenarios represented are realistic
since in most cases the target being oblivious of actual sensor
position would take the shortest path to destination that would
be a straight line. Thus, we believe that the above results are
practical and can be used in actual sensor network design.



Figure 7: Variation of Number of Nodes with Threshold

REFERENCES
[1] D.Estrin, R. Govindan, Next century challenges: scalable coordination
in sensor networks, Mobicom 1999, Seattle, WA, pp. 263-270, August
1999.
[2] K. Sohrabi, J Gao, V. Ailawadhi, G. Pottie, Protocols for self-
organization of a wireless sensor network, IEEE Personal
Communication Magazine, vol. 7, no. 5, pp. 16-27, Oct 2000.
[3] G. J. Pottie, W. J. Kaiser, Wireless integrated network sensors,
Communications of the ACM, vol. 43, no.5, pp. 551-558, May 2000
[4] S. Meguerdichain, F. Koushanfar, M. Potkonjak, M. B. Srivastava,
Coverage problems in wireless sensor Ad-hoc networks, IEEE
INFOCOM, 2001.
[5] S. Meguerdichain, F. Koushanfar, G. Qu, M. Potkonjak, Exposure in
wireless ad-hoc sensor networks, Mobicom 2001, Rome, Italy, July
2001.
[6] P. Gupta and P.R. Kumar, The Capacity of Wireless Networks, IEEE
Transaction on Information Theory, vol. 46, No. 2, March 2000.
[7] M. Franceschetti, M. Cook, J. Bruck, A Geometric theorem for
wireless network design optimization,
http://www.paradise.caltech.edu/ETR.html.


Figure 8: Variation of Number of nodes for different K Cooperative
Case


Figure 9: Probability of Detection Vs. Number of Node k =3 and k =4
cases

1620

También podría gustarte