Академический Документы
Профессиональный Документы
Культура Документы
Curve-based Deployment
Shibo He
, Xiaowen Gong
, Junshan Zhang
, Jiming Chen
School of Electrical, Computer, and Energy Engineering, Arizona State University, Tempe, AZ, 85287, USA
,
||
AB||
||
AGB||
2n+
21
2n
, where ||
AB||, ||
,
||
AB||
||
AGB||
2n+
21
2n
, where ||
AB||, ||
AGB|| and
n are constants related to the problem.
The rest of the paper is organized as follows. We present
in Sec. II the system model. In Sec. III, we give a sufcient
condition to characterize if a curve is the optimal deployment
curve, based on which sensor placement algorithms are treated
in depth in Sec. IV. We provide numerical results in Sec. V
to verify our analysis. We conclude our work in Sec. VI.
II. SYSTEM MODEL
We consider the deployment of a wireless sensor network
to monitor a region of interest of arbitrary shape, which is
a connected region without holes inside, and is enclosed by
four continuous curves: an entrance side
AB, a destination
side
CD, a left boundary side
AC and a right boundary
side
BD, as illustrated in Fig. 1(a). Intruders are assumed
to traverse region from any path beginning from the point
on the entrance side
AB and ending from the point on the
destination side
CD. In practice, can be regarded as the
intermediate zone that prevents intruders from accessing the
security zone adjacent to the destination side. We will use
XY
to denote a curve, and XY to denote a line segment, where X
and Y are the start and end points of the curve/line segment.
The length of curve
XY (for line segment XY ) is denoted
by ||
XY || (||XY ||).
There are n sensors available to detect intruders while
they are crossing . Denote the location of sensor i by T
i
,
i = 1, 2, , n. We use the practical SNR-based sensing
model, i.e., for a target located at X, the received SNR at
sensor i is given by SNR
i
(X) =
||TiX||
, where =
, its
vulnerability V under deployment T is dened as V
T
(
Y Y
) =
max
X
Y Y
V
T
(X). A candidate barrier
XX
is better than a
candidate barrier
Y Y
, denoted by V(
XX
) V(
Y Y
), if
B
A
C D
E F
(
.
Fig. 1. Region .
there exists a deployment T
such that V
T
(
XX
) V
T
(
Y Y
)
holds for any arbitrary deployment T.
A curve
XX
itself is
a candidate barrier. For a deployment scheme T on deployment
curve
XX
ZZ
), where
ZZ
, Z
AC, Z
BD, is an arbitrary candidate barrier. When the
deployment curve
XX
XX
), is dened as
the largest distance between any two points on the curve, i.e.,
D(
XX
) = max
P,P
XX
||PP
and a deployment
line segment Y Y
, V(Y Y
) V(
B
X
B
X
) under the condition
D(
XX
) ||Y Y
||.
Lemma 1 can be obtained directly from Theorem 1 in [11].
When ||L|| = ||EF||, i.e., the shortest line segment is also a
shortcut, for any deployment curve
XX
, D(
XX
) ||EF||.
According to Lemma 1, V(EF) V(
B
X
B
X
). Therefore,
under this condition, line-based deployment is optimal.
However, when ||L|| < ||EF||, line-based deployment may
not be optimal. We proceed to nd a sufcient condition under
which line-based deployment is suboptimal, i.e., there exists
a deployment curve
XX
such that V(
XX
XX
XX
|| ||Y Y
||,
then V(
XX
) V(Y Y
).
Suppose that Y Y
) <
V(EF). According to Lemma 2, V(C) V(Y Y
) < V(EF).
We have the following theorem.
Theorem 1: The line-based deployment is strictly subopti-
mal when ||C|| < ||EF||.
Theorem 1 indicates that curve-based deployment is more
preferable when ||C|| < ||EF||. Under this condition, a natural
question is how to nd the optimal deployment curve, which
is extremely difcult as region can be of arbitrary shape.
We give a sufcient condition to characterize the optimal
deployment curve in this paper. The necessary condition is
still an open issue.
Since C is the curve in with smallest length, it is of
great interest to study if C is the optimal deployment curve.
One observation is that part of C must intersect with the
entrance or destination side. We can verify this observation
by contradiction. Assume C lies within . Find two points
X and X
and
||XX
|| < ||
XX
||, where
XX
, where MM
2
3
' 3
<
' <
(a) (b)
M
N
' 0
' 1
3
' 3
'' 1
)
w)
Fig. 2. An Illustration of Lemma 3.
Lemma 3: For any two points P and P
), M, N . Then we
have that ||PP
|| ||MN||.
Denote the region enclosed by line segment AB and curve
AB by
.
With Property 1, for any differentiable point P, P
,
draw its tangent line and then draw a normal line PM that
passes through P and is perpendicular to the tangent line; for
a point P from set of non-differentiable points, P
,
draw its left tangent and right tangent lines, and then draw
two normal lines passing through P, which are perpendicular
to these two tangent lines, respectively. See an illustration in
Fig. 1(b). For each normal line, we only consider the ray PM
which only intersects with
AB||;
G /
AC and H
BD, and the length of shortcut
between curves
AC and
HD is larger than ||
AB||;
G
AC and H /
BD, and the length of shortcut
between curves
GC and
BD is larger than ||
AB||;
G /
AC and H /
BD.
From Theorem 2, we can easily get the following corollary.
Corollary 1: For any curve
APB in the region , whose
start point is A and end point is B, we have V(
AB)
V(
APB
), where
APB
, let Z(P, d) = {P
|P
XX
, ||PP
P
1
P
2
||, Z(P, d
2
) =
XP
3
P
4
P
5
, and
||Z(P, d
2
)|| = ||
XP
3
|| + ||
P
4
P
5
||. Points in Z(P, d
1
) are
connected while points in Z(P, d
2
) are not.
Denition 1: A continuous curve
XX
is distance-
continuous if Z(P, d) is connected, and ||Z(P, d)|| is a con-
tinuous function of d, for any P
XX
and d [0, ).
Denition 2: For two points P and P
on the curve
XX
,
P is prior to P
(denoted by P < P
), if ||
XP|| < ||
XP
||;
P is a peer of P
(denoted by P = P
), if ||
XP|| = ||
XP
||.
1
G
2
G
1
3
2
3
3
3
4
3
5
3
; ' ;
3
(a) Z(P, d) and |Z(P, d)|.
;
' ;
3
' 3
'' 3
<
k
' ;;
k
' ;*;
(b) An illustration of Property 2
Fig. 3. An illustration of distance-continuous.
With the denitions, we have the following results.
Lemma 4: For any three points P, P
and P
on
XX
satisfying P < P
< P
,
XX
is distance-continuous if and
only if ||PP
|| < ||PP
|| holds.
From Lemma 4, the following Corollary can be easily
obtained.
Corollary 2: Assume that
XX
is distance-continuous. For
any four points P, P
, P
and P
on
XX
satisfying P <
P
< P
< P
, we have ||PP
|| > ||P
||.
With the concept of distance-continuous, we may wonder
if any curve is distance-continuous. Fortunately, with Lemma
4, we can identify a class of distance-continuous curves.
Property 2:
XX
is distance-continuous if
XX
is a semi-
circle with the start point X and end point X
. Furthermore,
denote by the region enclosed by
XX
. Let
XGX
is distance-continuous.
B. The Case When
AB is Distance-continuous
For a given deployment scheme T = {T
1
, T
2
, , T
n
},
draw line segments connecting T
1
T
2
, T
2
T
3
, , T
n1
T
n
, se-
quentially. For line segment T
i
T
i+1
, i = 1, 2, , n1, draw
its midperpendicular, which intersects with curve
T
i
T
i+1
at
point H
i
1
. Let H
0
= A and H
n
= B. For any deploy-
ment T, H
i
, i = 0, 1, , n, can be uniquely determined.
Denote the largest distance between T
i
and curve
H
i1
H
i
by
D(T
i
,
H
i1
H
i
), i = 1, , n.
With the distance-continuous property, i is the nearest
sensor to the points on curve
H
i1
H
i
. The vulnerability
of curve
H
i1
H
i
is independent of other sensor j, j =
1
The midperpendicular may intersect with
H
i1
T
i
, and ||T
i
H
i
|| ||T
i
P||, P
T
i
H
i
. Therefore,
V(
AB) = max
i
V(
H
i1
H
i
) = max
i
D(T
i
,
H
i1
H
i
)/, where
D(T
i
,
H
i1
H
i
) = max{||H
i1
T
i
||, ||T
i
H
i
||}. In view of this,
H
i1
H
i
is dened as sensor is effective detection curve
and
D(T
i
,
H
i1
H
i
) the effective detection range. Obviously,
sensor is effective detection curve and range are determined
by a specic deployment.
We design the Curve-based Sensor Deployment when the
deployment curve
AB is Distance-Continuous (CSD-DC) in
the following.
Algorithm 1: Deploy sensors on curve
AB such that
D(T
i
,
H
i1
H
i
) = ||H
i1
T
i
|| = ||T
i
H
i
||, (1)
D(T
i
,
H
i1
H
i
) =
D(T
j
,
H
j1
H
j
), (2)
for any i, j = 1, 2, , n.
The following result gives the existence of the sensor
deployment according to Algorithm 1.
Lemma 5: Algorithm 1 can always nd a feasible deploy-
ment T = {T
1
, T
2
, , T
n
}.
The proof (refer to [10]) of Lemma 5 provides a way to nd
the deployment designed by Algorithm 1. The performance of
Algorithm 1 is given in the following theorem.
Theorem 3: The algorithm 1 obtains the optimal deploy-
ment when
AB is distance-continuous.
C. The Case When
AB is not Distance-continuous
When
AB is not distance-continuous, points in Z(P, d),
P
AB, d 0, are probably not connected. Placing a sensor
on a point of the curve
AB can cover several disconnected
sub-curves. To obtain an optimal deployment, we have to try
every possible combination of locations of n sensors, which is
computationally impossible. Hence, in this section, we focus
on providing an approximation algorithm with theoretical
bound.
Algorithm 2: Deploy sensors on curve
AB such that
||
T
i1
T
i
|| = ||
T
j1
T
j
|| = 2||
T
0
T
1
|| = 2||
T
n
T
n+1
||, i, j =
2, , n, where A = T
0
< T
1
< < T
n+1
= B.
Before showing the performance of Algorithm 2, we give
the following Lemma 6.
Lemma 6: Assume XX
and X
= 90
. The vulnerability of
XX
is lower-bounded by V(
XX
X)
||
XX
||
(2n+
21)
,
where
XX
= XX
.
Draw a semicircle centered at the middle point of AB with
a radius of ||AB||/2. From Property 2,
is the region
enclosed by AB and
AB) can not be included by the region
enclosed by AB and the semicircle since
AB is not distance-
continuous. Then, one of the following conditions must hold:
i)
AB intersects with the semicircle, and ii)
contains the
semicircle. Hence, we can always nd a point G in the region
. Denote AG GB by
AGB, as
illustrated in Fig. 4. We have the following results.
473
Theorem 4: When curve
AB is not distance-continuous, the
approximation ratio of the vulnerability corresponding to the
deployment by the Algorithm 2 to that by the optimal one is
upper-bounded by
D
T
(
AB)
D
T
(
AB)
min
,
||
AB||
||
AGB||
2n+
21
2n
.
A B
1
7
2
7
1 Q
7
Q
7
1
' 7
2
' 7
1
'
Q
7
'
Q
7
; ' ;
<
' <
A B
<
' <
'' <
*
B
Fig. 4. An illustration of Theorem 4.
V. NUMERICAL RESULTS
In what follows, we give some numerical results to validate
our analysis.
We will use function f(x) = x
2
, x [100, 100], under
different values of to yield deployment curves: i) = 0.01
(denoted by curve 1), ii) = 0.012 (denoted by curve 2),
iii) = 0.014 (denoted by curve 3), iv) = 0.02 (denoted
by curve 4), v) = 0.022 (denoted by curve 5), and vi)
= 0.024 (denoted by curve 6) in our simulations. For any
x
1
[100, 0], let (x) = (x
1
x)
2
+(x
2
1
x
2
)
2
, x x
1
.
It is easy to verify that
(x)
x
> 0, for any x
1
[100, 0]
when = 0.01, 0.012, 0.014. Then, for any x
1
[100, 0],
and x
1
< x
2
< x
3
100, we have ||x
1
x
2
|| < ||x
1
x
3
||.
As curve 1 is symmetrical, the conclusion holds when x
1
[0, 100]. Therefore, curves 1, 2, and 3 are distance-continuous.
Similarly, we can show curves 4, 5, and 6 are not distance-
continuous.
We rst vary the available number of sensors n from 3 to
10, and apply algorithm 1 to deploy sensors on curves 1, 2,
and 3. We plot in the Fig. 5(a) the vulnerability of the curves
1, 2, and 3 under the deployment obtained by Algorithm 1.
The values of vulnerability decrease with the increase of the
number of sensors. We also use Algorithm 2 to deploy sensors
on non-distance-continuous curves 4, 5, and 6, and the results
are also plotted in Fig. 5(a).
3 4 5 6 7 8 9 10
20
40
60
80
100
The number of sensor (n)
V
u
l
n
e
r
a
b
i
l
i
t
y
1
= 0.01
2
= 0.012
3
= 0.014
4
= 0.02
5
= 0.022
6
= 0.024
(a) Vulnerability of curve 1-6 under
different n.
2 4 6 8 10
10
20
30
40
50
60
The number of sensor (n)
V
u
l
n
e
r
a
b
i
l
i
t
y
Optimal deployment curve
Deployment line 1
Deployment line 2
Deployment line 3
(b) Comparisons of curve-based
approach with line-based approach.
Fig. 5. Numerical results.
Finally, we discuss the advantages of curve-based approach
over line-based approach. As discussed in the Sec. III, line-
based deployment is suboptimal when ||C|| < ||EF||. This
means the existing line-based deployment will nd a line
segment with larger length than what we nd in the curve-
based deployment. Assume curve 1 is the optimal deployment
curve found by curve-based approach, whose length is 295.78,
and the length of the shortest line segments found by different
line-based approaches are 305.78 (denoted by deployment line
1), 315.78 (denoted by deployment line 2) and 325.78 (denoted
by deployment line 3), respectively. We show the values of
vulnerability of curve 1 under the Algorithm 1 and deployment
lines 1-3 by line-based approach in the Fig. 5(b). Algorithm
1 has better performance than that by line-based approach.
The larger ||EF|| ||C||, the better performance curve-based
approach would obtain.
VI. CONCLUSION
This work studied sensor deployment for barrier coverage
in wireless sensor networks. We are interested in answering
three fundamental questions: 1) Is line-based approach always
optimal? 2) If not, can we nd the optimal deployment curve?
3) how to design sensors on the optimal deployment curve
to enhance barrier coverage performance? We rst showed
the suboptimality of line-based deployment when the length
of shortest deployment line segment is larger than that of
shortest deployment curve, and for the rst time quantied
the need of curve-based deployment. Finding the optimal
deployment curve is highly non-trivial due to the arbitrary
deployment region. We obtained a sufcient condition for the
deployment curve to be optimal, which gives an answer to
second question. To address the third question, we introduced
a concept of distance-continuous function, and provided an
algorithm to obtain the optimal sensor deployment when the
deployment curve is distance-continuous, and an algorithm
which can attain close-to-optimal sensor deployment when
the deployment curve is not. Numerical results have been
performed to validate our conclusions.
REFERENCES
[1] S. Kumar, T. Lai, and A. Arora. Barrier coverage with wireless sensors.
In Proceedings of ACM MobiCom, 2005.
[2] G. Yang and D. Qiao. Barrier information coverage with wireless
sensors. In Proceedings of IEEE INFOCOM, 2009.
[3] S. He, J. Chen, X. Li, X. Shen, and Y. Sun. Cost-effective barrier cov-
erage by mobile sensor networks. In Proceedings of IEEE INFOCOM,
2012.
[4] B. Liu, O. Dousse, J. Wang, and A. Saipulla. Strong barrier coverage
of wireless sensor networks. In Proceedings of ACM MobiHoc, 2008.
[5] A. Chen, S. Kumar, and T. Lai. Designing localized algorithms for
barrier coverage. In Proceedings of ACM MobiCom, 2007.
[6] Y. Wang and G. Cao. Barrier coverage in camera sensor networks. In
Proceedings of ACM MobiHoc, 2011.
[7] A. Saipulla, C. Westphal, B. Liu, and J. Wang. Barrier coverage of
line-based deployed wireless sensor networks. In Proceedings of IEEE
INFOCOM, 2009.
[8] S. Meguerdichian, F. Koushanfar, M. Potkonjak, and M. B. Srivastava.
Coverage problems in wireless ad-hoc sensor networks. In Proceedings
of IEEE INFOCOM, 2001.
[9] W. Ge, J. Zhang, and G. Xue. Cooperative geographic routing in wireless
sensor networks. In Proceedings of Milcom, 2006.
[10] S. He, X. Gong, J. Zhang, J. Chen, and Y. Sun. Barri-
er coverage in wireless sensor networks: From lined-based to
curve-based deployment. Technical report, 2013. Available at:
www.public.asu.edu/shibohe/technical report.pdf.
[11] X. Gong, J. Zhang, and D. Cochran. A coverage theory of bistatic radar
networks: Worst-case intrusion path and optimal deployment. Technical
report, 2012. Available at http://arxiv.org/abs/1206.1355.
474