Distributed Decision Making Algorithm for

Self-Healing Sensor Networks

Xiaojiang Du*, Ming Zhang*, Kendall Nygard*, Mohsen Guizani

, and Hsiao-Hwa Chen

*Dept. of Computer Science, North Dakota State University, Fargo, ND 58105, {Ming.Zhang, Xiaojiang.Du, Kendall.Nygard}@ndsu.edu

Dept. of Computer Science, Western Michigan University, [email protected]

Institute of Communications Engineering, National Sun Yat-Sen University, Taiwan, [email protected]

Abstract - Sensor nodes are unreliable small devices with
limited energy supply, and sensors may run out of power or fail
over time. A sensor network should have self-healing capability
to handle coverage holes, routing voids and disconnections
caused by random sensor locations and sensor failures. To
provide self-healing capabilities in sensor networks while keeping
the cost low, we propose to deploy a small number of mobile
sensors in addition to a large number of static sensors. A mobile
sensor may receive multiple requests from different areas. In this
paper, we present a distributed decision making algorithm based
on fuzzy logic. The distributed algorithm has low computation
requirement and is robust to incomplete information. Our
extensive simulations demonstrate the self-healing capabilities
provided by mobile sensors, and the good performance of the
distributed decision making algorithm.

I. INTRODUCTION

Previous studies mainly considered sensor networks where
either all sensors are static or all sensors are mobile [1, 2]. For
sensor networks with only static sensors, due to the large
number of sensor nodes and possible hostile environment
(e.g., battlefield, toxic regions), the exact sensor location may
not be controlled by typical deployment methods, e.g.,
distribution from an airplane. With these types of deployment,
some geographical areas may not be covered by any sensor
because of the randomness of sensor location. The coverage
holes potentially degrade sensing performance. Furthermore,
sensor nodes are unreliable small devices with limited power
supply, and are prone to fail. Sensor failure can cause
coverage holes, routing voids, and disconnections from the
BS, or even a partitioning of the network. In some previous
works [1, 2], researchers have considered sensor networks
where all sensors are mobile nodes and studied the problem of
how to move mobile sensors to maximize sensing coverage in
the network. However, a mobile sensor is much more
expensive than a static sensor with similar sensing capability
and deploying a large number of mobile sensors can be very
expensive.
To provide self-optimizing and self-healing capabilities in
sensor networks while keeping the cost low, we propose a
strategy in which a small number of mobile sensors are
purposely deployed to augment a large number of static
sensors. Mobile sensors can significantly improve network
performance by moving to locations where there is a coverage
hole or network disconnection. In [4], we have proposed
distributed algorithms for mobile sensors to cover coverage
holes and disconnected areas.
A mobile sensor may receive multiple requests from
different areas. Furthermore, a mobile sensor may not have
complete information of the network, since obtaining global
information will require excessive communication and energy
consumption. In this paper, we propose a fuzzy logic based
distributed decision making algorithm that has low
computation requirement and is robust to incomplete
information.

II. DISTRIBUTED DECISION MAKING ALGORITHMS FOR
MOBILE SENSORS

A. The Sensor Network Model
We first describe the sensor network model. Denote the totally
number of static sensor nodes as M, and the totally number of
mobile sensor nodes as N. We assume that each sensor node is
aware of its own location. Location awareness is a basic
requirement for sensor nodes since most sensing data must be
associated with the location where data is generated. A sensor
network can use location services such as [5] and [6] to
estimate the locations of the individual nodes, and no GPS
receiver is required at each sensor.
B. Distributed Decision Making Algorithms
Mobile sensors can essentially provide self-healing and self-
optimizing capabilities for sensor networks. A mobile sensor
may receive multiple calls for movement. The mobile sensor
then must decide where it should move to achieve the
maximum improvement on network performance. We design
distributed decision making algorithms to determine the
location where a mobile sensor should move.
In the following, we briefly present the idea of the
distributed decision making algorithm for mobile sensors.
Mobile sensors consume energy while moving [3]. Thus, the
distance to a calling area is an important factor to consider.
The coverage gain is the coverage improvement from using
mobile sensors. The coverage gain depends on the size of the
coverage hole and the sensing capability of the mobile sensor.
The connectivity gain is more complex than the coverage gain,
and it depends on several factors, such as the decreases in hop
distance and delay (to BS), and the increase of throughput
after mobile sensors move to the new location. Thus, the
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.
1-4244-0355-3/06/$20.00 (c) 2006 IEEE
decision depends on several factors, including the distance, the
coverage gain, and the connectivity gain. A utility function is
defined to include all of these factors. A mobile sensor should
move to the location that maximizes the utility function.
Partitioning a sensor network can significantly degrade
network performance. Thus, a mobile sensor should always
move to a location that can connect a partitioned area. A very
large gain value should be assigned to a network partition. The
detection of a disconnected area can be based on a change of
traffic volume, or on active roaming of mobile sensors.
A static sensor can detect a coverage hole or loss of
connection in its nearby area, then the static sensor can send
out a Call message that requests mobile sensors to move to the
location. The Call message includes the location of the calling
node, so that a mobile sensor can calculate the distance. A
static sensor can estimate the size of a coverage hole [4], and
the size of the hole can be included in the Call message, then a
mobile sensor can calculate the coverage gain based on the
hole size and its sensing capability. The Call message can also
include the estimated reduction of hop numbers and delay to
the BS, and the estimated throughput gain after a mobile
sensor moves in, and this information can be used by a mobile
sensor to calculate the connectivity gain. The details of the
distributed decision making algorithms are presented in
Section III.
C. Decision Making with Partial Information
A typical sensor network has a large number of nodes and the
communication may be unreliable. Obtaining complete
network information for the BS or a sensor node could be very
expensive or even impossible, since this requires excessive
communication and energy consumption. Thus, a mobile
sensor may not have complete information of the network. We
design schemes for mobile sensor to incorporate uncertainty
into decision making with partial information.
Specifically, we utilize fuzzy logic [7] to allow mobile
sensor make good decisions with incomplete information.
Fuzzy logic deals with imprecision and incompleteness of
information by formalizing the degree to which observations
fit a classification. Decision trees can be built under fuzzy
logic for decision making with partial information. This
approach provides a straightforward way to arrive at definite
conclusions based upon vague, ambiguous, imprecise, noisy,
or missing input information. The fuzzy decision support
approach incorporates simple logic rules (e.g., IF X AND Y
THEN Z) to make decisions. Fuzzy logic procedures can be
easily implemented in sensor nodes because they require only
small amounts of storage and computational resources. To
apply fuzzy decision making to mobile sensor movement, we
view mobile sensors as agents who have goals to cover holes
in the geographical area, take responsibility for their actions,
and do so with consideration of all the information to which
they have access. As they move and increase their knowledge
base, the mobile sensors can reason and make local movement
choices that seek to optimize their goals. We develop decision
rules for fuzzy logic algorithms used in mobile sensor. The
details are given in Section III.
III. THE PROBLEM FORMULATION

In this section, we present the notations and problem
formulation of the mobile sensor movement problem. The goal
is to find out the best location where each mobile sensor
(given the number of mobile sensors) should move and
maximize the performance improvement in a sensor network.
The problem is formulated as an Integer Linear Programming
(ILP) problem in subsection III.C.
A. Notations
We present the notations used in the rest of the paper below.
} ,..., , {
2 1 m
s s s S is the set of static sensor nodes.
} ,..., , {
2 1 n
m m m M is the set of mobile sensor nodes.
} ,..., , {
2 1 k
h h h H is the set coverage holes. A hole
j
h is
defined by the boundary static nodes of the hole, and
j
h S

is a subset of S.

j
h
: the size of hole of
j
h
- in terms of the number of
required mobile sensors.
) , (
j i
h m d : The Euclidean distance between a mobile
sensor
i
m and a hole
j
h
.
E: The available energy of a newly deployed mobile node.
) , (
j i c
h m g : The coverage gain if
i
m moves to hole
j
h
.
) , (
j i l
h m g : The connectivity gain if
i
m moves to hole
j
h
.
( , ) ( , )
i j i j
p m h k d m h : The required energy for sensor
i
m
to move to hole
j
h
. Here k is a constant.
( )
p j
g h : The priority of hole
j
h
.
B. The Utility Function
The utility function includes three factors: the movement cost,
the coverage gain, and the connectivity gain. The three factors
are defined below.
Movement Cost of Mobile Sensors. The movement cost
of a mobile sensor is in proportion to the distance it travels.
E.g., in [3], the mobile robot consumes 0.210 Joule per inch
movement. For simplicity, we define the movement cost to be
the distance. The actual coefficient will be included in the
utility function presented below. The movement cost is:
( , ) ( , )
i j i j
c m h d m h .
Coverage Gain. Given two sensor nodes s and
i
m , the
overlapping coverage between sensor s and
i
m is denoted as
( )
i
v s m I , and it is given below: If ( , ) 2
i
d d s m r ,
2
2
2
1
( ) 2 cos 1
2 4
i
d d d
v s m r ar
r r r

, ] j \
, ] , (
, (
, ]
( , ]
I

Otherwise, ( ) 0.
i
v s m I
The coverage gain of moving mobile sensor
i
m to a hole
j
h
is defined as:
2
( , ) ( )
j
c i j i
s h
g m h r v s m

I
U
.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.
I.e., ) , (
j i c
h m g is the net area covered by the mobile sensor
i
m (after it moves in).
Connectivity Gain. The connectivity gain is defined as the
fraction of new links (after a mobile sensor moves in) to the
existing links. I.e., the number of new links (after the mobile
node
i
m moves to the hole
j
h
) divided by the total number of
links among existing sensors.
For simplicity, assume the transmission ranges of all static
sensors and all mobile sensors are the same. First, we denote
the link between two sensor nodes m and s as:
,
1 ( , )
0
m s
d m s r
l
otherwise
,

, where r is the transmission range of a

sensor node. Then we denote
,
i i
j
m m s
s h
l l

as the total
number of new links between a mobile sensor
i
m and
boundary nodes of a hole
j
h
. The number of new links can be
determined by using the location information of the mobile
sensor (after movement) and each boundary node of the hole.
Now we define the connection gain as:
( , )
l i j
g m h /
i
j
m s
s h
l l

where
n
l is the number of links of node n.
The Utility Function. For a given mobile sensor
i
m and
a hole
j
h
, the utility function is:
1 2
( , ) ( , ) ( , ) ( , )
i j i j c i j l i j
g m h c m h g m h g m h + + ) (*)
where
1
and
2
are coefficients.
C. The Integer Linear Programming Formulation
The optimal sensor movement problem in the entire network
can be formulated as an Integer Linear Programming (ILP)
problem. The formulation seeks the global optimal solution of
the mobile sensor movement problem. The ILP formulation is
presented in the following. We define the variable of the ILP
as
ij
x , where 1
ij
x , if mobile sensor
i
m moves to hole
j
h
;
otherwise
0
ij
x
.
The objective function of the ILP problem is to:
Maximize ( , )
ij i j
i M j H
x g m h

where
( , )
i j
g m h
is given in (*).
Subject to:
1. The energy spent on movement should be less than a given
portion of the initial energy of a mobile sensor. I.e.,
( , )
ij i j
j H
x c m h E

for each
i
m , ( 1 0 ) (1)
2. Each mobile node only moves once.
1
i j
j
x

for each mobile sensors

i
m (2)
3. The number of mobile nodes for a hole should be less than
the size of the hole.
i j j
i
x h

for each hole

j
h
(3)
Note that constrain (2) may be removed depending on the
network design. The computation requirement of solving the
above ILP problem exceeds the capability of typical sensor
nodes. Thus, we will not discuss how to solving the ILP.
Instead, we propose a fuzzy logic based distributed decision
making algorithm that has low computation requirement and
are suitable to sensor nodes. The detail is given in Section IV.
D. The Distributed Algorithm at Each Mobile Sensor
After a mobile sensor receives the first request from a hole,
the mobile sensor sets up a timer T, and wait for more
requests. When the timer expires, the mobile sensor uses the
fuzzy logic algorithm to determine the best location that it
should move to. At each mobile sensor, a threshold-based
algorithm is used. I.e., when the value of the utility function is
less than a pre-defined threshold, the mobile sensor will reject
the request and wait for new requests. The threshold-based
algorithm is used to avoid unnecessary movement of a mobile
sensor. The distributed algorithm running at each mobile
sensor is given in Figure 1.

1. A mobile sensor m receives the first request from a hole.
And m sets a timer T.
2. While time t < T:
3. Wait for new requests.
4. Calculate the movement cost (based on the distance),
the coverage gain and connection gain.
5. If the utility function <= the threshold
6. Reject the request and wait for new requests;
7. else
8. Mobile sensor stores the request.
9. end if
10.End While.
11.Apply fuzzy logic algorithm and find out the best
location.
12.Move to the location.

Figure 1: The Distributed Algorithm in Each Mobile Sensor

IV. DISTRIBUTED DECISION MAKING USING FUZZY LOGIC

We first define the fuzzy variables as follows. Let
) , (
h
m
h
m
h
m
b w represent a mobile node ms decision on the
request from a hole h. Where
h
m
h
m
b , denote the belief of
moving to the hole and waiting for further request
respectively. The believes satisfy 1 +
h
m
h
m
b .
A. The Membership Function
The membership functions of the fuzzy logic algorithm are
presented in Figure 2. The outputs of the membership
functions are classified into three categories: small, medium
and large, depending on the distance, the size of the hole and
the number of new links created by the mobile sensor,
respectively.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.


Figure 2(a): The Input is the Distance


Figure 2(b): The Input is the Hole Size

Figure 2(c): The Input is the Number of New Links from Mobile Sensors

Figure 2: The Membership Functions of the Fuzzy Logic Algorithm

B. Fuzzy Rules
We define fuzzy rules based on the following principles: short
distance has higher priority; large hole has higher priority;
large increase of new connections has higher priority. The
fuzzy rules are listed in Table I.

TABLE I: FUZZY RULES

Distance Size Number of New
Connections
Decision
L - - N
M M/S S N
M M/S L Y-
M L L Y
S L S Y+
S M S Y
S S S N
S L M Y+
S M M Y+
S S M Y
S L/M L Y
S S L Y-
The notations in Table I are explained below: L Large;
M Medium; S Small; N No; Y Yes.
C. Defuzzification
We use the centre of area (COA) method to do the
defuzzification. It calculates the centre of gravity of the
resultant fuzzy set.
V. EXPERIMENTAL RESULTS

With the help of the fuzzy algorithm described in Section IV,
a mobile sensor uses a fuzzy controller to decide it action. The
fuzzy controller has three inputs: distance, coverage gain and
connectivity gain. Based on the fuzzy rules in Table I, the
mobile sensor can make the moving decision. The fuzzy
controller has four possible outputs: positive move, move,
negative move and not move.
We design simulation experiments to evaluate the
performance of the fuzzy-logical-based distributed decision
making algorithm. The performance of the fuzzy algorithm is
compared with two heuristic algorithms - First Call First Go
(FCFG) algorithm and Weighted Sum (WS) algorithm, which
are described below. FCFG algorithm: Once a mobile sensor
receives a request from a hole, it moves to the hole
immediately. WS algorithm: A mobile sensor waits for
requests for a pre-defined time and then moves to the hole that
has the maximum weighted sum of three factors: distance,
coverage gain and connectivity gain. The weight of each
factor is pre-defined (obtained from extensive simulation
experiments). In the following experiments, we use two sets of
weights: Distance:Cov:Conn = 4:3:3 and 2:5:3.
We conduct two sets of experiments. In the first set of
experiments, the total area of coverage holes and the number
of mobile nodes are fixed. The network size varies from
400m*400m to 1000m*1000m. In the second set of
experiments, the network size and the number of mobile nodes
are fixed, while the total area of coverage holes changes.
A. Results from Experiment Set One
In the experiments, sensor nodes are randomly deployed in a
square. The side length of the square varies from 400m to
1000m with an increasing step of 100m. The schemes in [4]
are used to detect coverage holes and estimate the size of
holes, the coverage gain and connectivity gain. In each
experiment, the total number of required mobile sensors is set
to 60, and 30 mobile nodes are deployed in random locations
in the network. The maximal acceptable moving distance of a
mobile sensor is set to 100m. The experiment is repeated 10
times for each configuration and the presented result is the
average of the 10 experiments.
The experimental results are shown in Figure 3~5, which
plot the total moving distances, the total coverage gains and
the total connectivity gains of mobile sensors under different
algorithms, when the network size varies from 400m*400m to
1000m*1000m. From Figure 3, we can see that FCFG
algorithm always has shorter moving distances than other
algorithms. The main reason is that mobile sensors using
FCFG always move to the first calling hole, which usually is a
close hole. However, Figure 4 and 5 show that FCFG has less
coverage and connectivity gains than WS and fuzzy
algorithms. From Figure 4 and 5, we find out that WS
algorithm and fuzzy logic algorithm have similar coverage and
connectivity gains for all the tested network sizes. In addition,
Figure 3 shows that sensors under fuzzy algorithm have
shorter moving distance than those under WS algorithm.
Small Large Medium
3 6
Number of
New links
Relative
Hole Size
Small Large Medium
0.33 0.67
Distance
Small Medium
50 m 100m
Large
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.
400 500 600 700 800 900 1000
6
7
8
9
10
11
12
13
Network Size (m)
T
o
t
a
l

M
o
v
i
n
g

D
i
s
t
a
n
c
e

(
x
1
0
0
m
) FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 3: The Total Moving Distance under Different Algorithms
400 500 600 700 800 900 1000
6
8
10
12
14
16
18
20
Network Size (m)
T
o
t
a
l

C
o
v
e
r
a
g
e

G
a
i
n
FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 4: The Total Coverage Gain under Different Algorithms
400 500 600 700 800 900 1000
40
60
80
100
120
140
Network Size (m)
T
o
t
a
l

C
o
n
n
e
c
t
i
o
n

G
a
i
n
FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 5: The Total Connectivity Gain under Different Algorithms

We also calculate the utility function (given in equation
(*)), for different weights
1
and
2
. Of course the utility
function depends on the values of
1
and
2
. Determining the
proper values for
1
and
2
is not a trivial task. One approach
is to run extensive simulations with different values of
1
and
2
, and compare the utility function with the actual lifetime of
the sensor network. The values that provide a good match
between the utility function and the network lifetime are
proper choices for
1
and
2
. The simulation-based approach
will be our future work. In the current work, we compute the
utility function for several different pairs of
1
and
2
. Most
results show that the fuzzy-logic-based distributed algorithm
achieves the largest utility function among all the algorithms.
We plot the utility function for two pairs of
1
and
2
in
Figure 6, where (
1
,
2
) are (0.7, 0.02) and (0.8, 0.01) in
Figure 6(a) and 6(b) respectively.
400 500 600 700 800 900 1000
0
1
2
3
4
5
6
Network Size (m)
T
h
e

U
t
i
l
i
t
y

F
u
n
c
t
i
o
n
FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 6(a): (
1

,
2

) = (0.7, 0.02)
400 500 600 700 800 900 1000
1
0
1
2
3
4
5
Network Size (m)
T
h
e

U
t
i
l
i
t
y

F
u
n
c
t
i
o
n
FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 6(b): (
1

,
2

) = (0.8, 0.01)

Figure 6: The Utility Function under Different Algorithms

B. Results from Experiment Set Two
In the 2
nd
set of experiments, we fix the network size to be
1000m*1000m, and 200 mobile sensors are randomly
deployed in the network. The maximal acceptable moving
distance for mobile node is set to 100m. The number of holes
varies from 20 to 100 in the tests.
We measure the total moving distances, the total coverage
gains and the total connectivity gains of mobile sensors under
different algorithms, which are plotted in Figure 7~10. From
Figure 7, we can see that FCFG algorithm always has shorter
total moving distance than the other two algorithms. The
reason is already stated in subsection V.A. Changing the
weight of different factors have impact on the decision. In
Figure 8, the WS algorithm with coverage weight 0.5 has
better coverage gain than other algorithms. Figure 8 also
shows that fuzzy logic algorithm has a good coverage gain.
FCFG algorithm has the lowest coverage gain and
connectivity gain. On the other hand, the response time of
FCFG algorithm is much shorter than other algorithms. Figure
9 shows that the WS algorithm and fuzzy logic algorithm have
similar results. When adding the three factors together, the
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.
fuzzy logic algorithm has the largest value of the utility
function among all the algorithms. We also compute the utility
function for different values of
1
and
2
. Most results show
that the fuzzy-logic-based distributed algorithm achieves the
largest utility function among all the algorithms. The utility
functions for two pairs of
1
and
2
are shown in Figure 10,
where
1
and
2
are (0.5, 0.2) and (0.8, 0.1) in Figure 10(a)
and 10(b) respectively.
20 40 60 80 100
20
40
60
80
100
120
140
Number of Holes
T
o
t
a
l

M
o
v
i
n
g

D
i
s
t
a
n
c
e

(
x
1
0
0
m
) FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 7: The Total Moving Distance under Different Algorithms
20 40 60 80 100
20
40
60
80
100
Number of Holes
T
o
t
a
l

C
o
v
e
r
a
g
e

G
a
i
n
FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 8: The Total Coverage Gain under Different Algorithms
20 40 60 80 100
0
200
400
600
800
1000
Number of Holes
T
o
t
a
l

C
o
n
n
e
c
t
i
o
n

G
a
i
n
FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 9: The Total Connectivity Gain under Different Algorithms

VI. CONCLUSIONS

To provide self-healing capabilities for sensor networks, we
propose to deploy a few mobile sensors in a sensor network. A
mobile sensor may receive multiple requests from different
areas. In this paper, we formulated the optimal sensor
movement problem as an Integer Linear Programming (ILP)
problem. The computation requirement of solving the ILP
problem exceeds the capability of typical sensor nodes. To
overcome the problem, we proposed a fuzzy logic based
distributed decision making algorithm that has low
computation requirements and is robust to incomplete
information. Our simulations demonstrated that the distributed
algorithm can determine a good location where a mobile
sensor should move.
20 40 60 80 100
20
40
60
80
100
120
Number of Holes
T
h
e

U
t
i
l
i
t
y

F
u
n
c
t
i
o
n
FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 10(a): (
1

,
2

) = (0.5, 0.2)
20 40 60 80 100
10
20
30
40
50
60
Number of Holes
T
h
e

U
t
i
l
i
t
y

F
u
n
c
t
i
o
n
FCFG
WS(4:3:3)
Fuzzy
WS(2:5:3)

Figure 10(b): (
1

,
2

) = (0.8, 0.1)

Figure 10: The Utility Function under Different Algorithms

REFERENCES

[1] Y. Zou and K. Chakrabarty, Sensor deployment and target localization
based on virtual forces, Proc. of the IEEE INFOCOM 2003.
[2] G. Wang, G. Cao, T. La Porta, Movement-Assisted Sensor Deployment,
Proc. of the IEEE INFOCOM 2004.
[3] M. Rahimi, H. Shah, G. Sukhatme, et al., Studying the Feasibility of
Energy Harvesting in a Mobile Sensor Network, Proc. of the IEEE
International Conference on Robotics and Automation, pp. 19-24. Taipai,
Taiwan, May, 2003.
[4] M. Zhang, X. Du, and K. Nygard, Improving Coverage Performance in
Sensor Networks by Using Mobile Sensors, Proc. of the IEEE Military
Communication (MILCOM) 2005, Atlantic City, NJ, Oct. 2005.
[5] J. Bruck, J Gao, A. Jiang, Localization and Routing in Sensor Networks
by Local Angle Information, Proc. of the 6th ACM MobiHoc, UIUC, May,
2005.
[6] L. Doherty, L. El Ghaoui, and K. S. J. Pister, Convex position estimation
in wireless sensor networks, Proc. of IEEE INFOCOM 2001, Anchorage,
AK, Apr. 2001.
[7] D. Driankov, H. Hellendorf, and M. Reinfrank, An introduction to fuzzy
control, Springer-Verlag, 1993.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2006 proceedings.