You are on page 1of 5

Dynamic Spectrum Allocation in Wireless Cognitive Sensor

Networks: Improving Fairness and Energy Efficiency


Sang-Seon Byun
1
, Ilangko Balasingham
1, 2
, and Xuedong Liang
2, 3
1
Norwegian University of Science and
Technology, Trondheim, Norway
sang-seon.byun,@iet.ntnu.no
2
Rikshospitalet University Hospital,
Oslo, Norway
ilangkob@klinmed.uio.no
3
University of Oslo, Oslo, Norway
xuedong.liang@medisin.uio.no
Abstract- This paper considers the centralized spectrum allo-
cations in resource-constrained wireless sensor networks with the
following goals: (1) allocate spectrum as fairly as possible, (2) util-
ize spectrum resource maximally, (3) reflect the priority among
sensor data, and (4) reduce spectrum handoff. The problem is
formulated into a multi-objective problem, where we propose a
new approach to solve it using modified game theory (MGT). In
addition, cooperative game theory is adopted to obtain approxi-
mated solutions for MGT in reasonable time. The results ob-
tained from numerical experiments show that the proposed algo-
rithm allocates spectrum bands fairly with well observing each
sensors priority and nearly minimal spectrum handoffs.
I. INTRODUCTION
Cognitive radio exploits the underutilized spectrum re-
sources along time or frequency dimension and provides effi-
cient and intensive dynamic spectrum access through the de-
tection of spectrum opportunity and adaptive modulation [1,
2].
Such dynamic spectrum access schemes can be deployed
in the area of wireless sensor network (WSN). A typical WSN
is composed of resource-constrained sensors responsible for
monitoring physical phenomena and reporting data to sink
points. One of the primary objectives of WSNs is to transmit a
large amount of data timely and simultaneously, without using
large amount of network resources. Thus the dynamic spec-
trum access can become very vital to realize such transmis-
sions. For example, in the areas of healthcare and telemedi-
cine, a large number of sensors that are used to monitor the
vital signs of the patients should transmit their monitored re-
sults concurrently and timely. Furthermore, we can imagine
that there will be many wireless devices, which are operating
in the unlicensed ISM bands.
Considering the deployment of the dynamic spectrum ac-
cess in WSNs, the following major principles can be men-
tioned:
High utilization and fair allocation of idle spectrum bands
If too many sensors attempt to transmit their data simulta-
neously, current idle spectrum resources may be insufficient to
support all transmissions. In this situation, the spectrum re-
sources should be allocated as fairly as possible and maxi-
mally utilized; in addition, the transmissions should be priori-
tized. This can be achieved by maximizing the proportional
fairness with demand weights [9].
Minimum spectrum handoff
The other primary objective of WSNs is long time func-
tionality. However, the dynamic spectrum access leads to sup-
plemental energy consumption at each sensor due to the spec-
trum handoff. In order to verify this, we simulated ZigBee
standard radio using OMNET++ 3.3 [3] and measured the
amount of energy, which is consumed when a spectrum
handoff occurs. The simulation results show that a single
spectrum handoff consumes approximately 96.06% of the
average receiving energy or 110.75% of the average transmit-
ting energy. It means that the more instant spectrum handoffs
lead to the more power consumption. Hence unnecessary
spectrum handoff should be eliminated.
Centralized spectrum allocation
The problem of dynamic spectrum allocation in cellular
network has been studied widely, where various centralized or
distributed algorithms are summarized and compared in [4]. In
the centralized scheme, a centralized authority (e.g., base sta-
tion or dedicated coordinator) detects and identifies spectrum
opportunities and allocates the identified spectra to secondary
users in accordance with a predefined policy. Undoubtedly
there will be an overhead of message exchange since the cen-
tralized authority should propagate the results of spectrum
allocation into the sensor swarm whenever it receives trans-
mission requests.
In the distributed scheme, each secondary user competes
with one another to access the available spectrum resources.
Thus each user should have an ability to detect the spectrum
opportunities and determine an optimal strategy to maximize
its benefits (e.g., number of spectrum bands) over time. This
means that all sensors should have a fully operating cognitive
radio. However it is not feasible to deploy the full functions of
cognitive radio to such resource-constrained sensor nodes.
Thus, in a moderate size of WSN, where the sensors are not
distributed widely, for example, a healthcare system in an in-
tensive care unit, the centralized scheme is preferred to the
distributed one.
In this paper, we consider the problem of the centralized
spectrum allocations in WSN of moderate size, i.e., all sensors
are located within a cell or segment boundary. We assume that
sensors, which are going to transmit data, request spectrum
resources to the dedicated coordinator. Then the coordinator
allocates spectrum resources to the sensors: (1) as fairly as
possible, (2) achieving maximum spectrum utilization, (3) re-
flecting the priority among sensors, and (4) with infrequent
spectrum handoff. This means that it is necessary to cast this
problem into a multi-objective problem. With the considera-
tion of system efficiency and complexity, we formulate the
multi-objective problem into a single-objective problem using
978-1-4244-1722-3/08/$25.00 2008 IEEE. 1
modified game theory (MGT) [7]. Furthermore we adopt a
cooperative game theory to solve the MGT. A significant
point for the proposed algorithm is that the complexity is
O(N
2
M), where N and M are the number of sensors and the
number of idle spectrum resources, respectively. From the
numerical experiments, the proposed algorithms allocate
spectrum resources fairly while observing each sensors prior-
ity and nearly minimal spectrum handoffs.
The rest of this paper is organized as follows. Section II
presents related work. Section III provides the system model
and problem formulations. Section IV converts the multi-
objective problem into a single-objective one using MGT, and
Section V proposes a cooperative game-based algorithm to
solve the MGT. Section VI describes the results of numerical
experiments. Finally conclusions are given in Section VII.
II. RELATED WORK
To our best knowledge, there are only a few papers that
mention the deployment of cognitive radio in the context of
WSNs. In [5], it was proposed that by adjusting the constella-
tion size, different data rate can be achieved. This will directly
influence the power consumption of each node and, in turn,
will affect the lifetime of the whole sensor network. Further-
more a distributed spectrum allocation strategy was adopted
with the assumption that each sensor node had a fully func-
tional cognitive radio i.e., detection, identification and ex-
ploitation of spectrum opportunities. In [6], the authors ap-
plied the dynamic spectrum access in the time domain by ex-
ploiting white space between bursty transmissions of multi-
access communication channels. It was mentioned that, if sen-
sors communicate sporadically or perhaps at a low rate, the
system can efficiently reuse remaining white space.
With regard to the centralized spectrum allocation, a fair
scheme to allocate subcarrier, rate, and power for multiuser
OFDM system using a cooperative game approach was pro-
posed in [8]. In [10-13], various non-cooperative games were
introduced for the distributed dynamic spectrum sharing for
situations, where all users are selfish and do not reveal their
private information.
III. SYSTEM MODEL AND PROBLEM FORMULATION
We formulate the centralized spectrum allocation problem
as a multi-objective nonlinear programming. Prior to describe
the formulation, the following assumptions are made:
1. For the simplicity, the available spectrum resources are
expressed as a multiple of predefined spectrum unit.
2. A spectrum unit can be part of a white space in time do-
main [6, 14, 15] or a frequency spectrum band [5].
3. In case of dynamic spectrum access in the frequency do-
main, we assume interleaved FDMA [16-18].
4. All the sensors are distributed within the range of a base
station (or AP) with which primary nodes are associated.
5. The detection of spectrum opportunities, identification and
exploitation are always correct, and a dedicated coordinator
does all these processes.
6. With regard to the transmission power control of each sen-
sor, we consider both cases: 1) the case that all sensors
have ability to control their transmission power in order not
to interfere with other sensors beyond their targeted range,
and 2) the case that none of sensors have ability to control
their transmission power. Therefore a spectrum unit is al-
lowed to be assigned at only one sensor.
7. A simple model for interference is based on the binary ge-
ometry metric [10].
8. The process of spectrum allocation is performed in every
start of predefined epoch, where the coordinator detects idle
spectrum units and sensors that requested spectrum re-
sources. The epoch may also start whenever the coordinator
receives demands of spectrum resources, or periodically.
9. Routing has been done and the route between any
source/destination pair is given.
10. We consider minimizing the number of spectrum handoffs
at only transmitter.
11. The transmissions may be performed in the way of one-to-
many, many-to-one, or one-to-one.
Figure 1. An example illustrates interference. If sensor is target lies
within the transmission range of j, i and j cannot share the same
spectrum bands.
Based on the above assumptions, the multi-objective
nonlinear programming can be described as follows:
Parameters
V: Set of sensors that request spectrum bands.
S: Set of idle spectrum units.
L
is
(binary): Currently, sensor i is synchronized with the
spectrum unit s.
w
i
: The weight in accordance with the priority of sensor i.
The priority reflects both the demand of spectrum resources
and the degree of urgency of the sensor data.
R
ij
(binary): It indicates that sensor i transmits data to a
destination that is in the transmission range of sensor j.
Thus the spectrum units that are occupied by sensor i
should be orthogonal with those occupied by sensor j. Fig-
ure 1 illustrates an example of interference. Also this con-
straint is described in Equation (3). Let R
ij
= 0 if i = j, or
sensor i and j transmit data to the same destination, and R
ij
= R
ji
.
Variable
x
is
(binary): It indicates that sensor i uses the spectrum units.
Objectives
maximize wi ln xis
sS







iV
(1)
2
maximize Lisxis
iV

sS

(2)
Subject to
xis + xjs ( )Rij 1 for all i, j V and s S. (3)
Using Equation (1), the idle spectrum units can be allo-
cated, as fairly as possible, achieving maximum spectrum
utilization, and reflecting the priority, while using Equation
(2), we can eliminate unnecessary spectrum handoffs.
IV. MODIFIED GAME THEORY APPROACH
Due to the computational complexities involved with the
original game theory, a modification to the method was sug-
gested [7]. We describe the algorithm for the MGT in this
section by deploying it to the problem of spectrum allocations.
a) Convert the maximization problem into minimization
problem as follows:
Objectives
minimize U = -1 wi ln xis
sS
_
(
\
|
\
!
|
iV
_ (4)
minimize H = -1 Lisxis
iV

sS

.
(5)
b) Normalize U and H as
Un =
U U
*
Uw U
*
,
Hn =
H H
*
Hw H
*
(6)
where U
*
is the optimal value when only U is minimized, H
*
is the optimal value when only H is minimized, U
w
is the
value of U when only H is minimized, and H
w
is the value of
H when only U is minimized.
c) Compute the normalized super-criterion as
Sn = (1Un)(1 Hn)
. (7)
d) Formulate a Pareto optimal objective function FC as fol-
lows:
FC = (Un)
q
+ (Hn)
q
{ }
1 q
(8)
where q 2.
e) Then the multi-objective problem is encapsulated into a
single objective problem as
Objective
Minimize F(X) = FC Sn (9)
Subject to
U
*
U Uw
(10)
H
*
H Hw
(11)
xis + xjs ( )Rij 1 for all i, j V and s S. (12)
V. COMPUTATION OF MGT
In Section IV, we convert the multi-objective problem
into a nonlinear single-objective problem with a single design
vector. However it is still not easy to solve this by using the
nonlinear programming in reasonable time. Moreover, prior to
solving the MGT, we should determine U
w
,

and H
w
.
Finding optimal <H
*
, U
w
> can be done easily using the
algorithm in Table I with the complexity of O(N
2
M), where N
and M are the number of sensors and the number of idle spec-
trum units, respectively. In Step 1, allocate idle spectrum units
to sensors according to array L. In Step 2, negotiate the spec-
trum ownership not to interfere with each other and towards
improving U
w
. Since U
w
acts as an upper bound of U during
the procedure of MGT, it is also important to improve U
w
in
order to obtain better U while not deteriorate H
*
.
TABLE I
THE ALGORITHM FOR DETERMINING OPTIMAL UW AND H
*
.
1. Initialization for each iV and sS
x
is
= L
is
2. Negotiation for each i, jV and sS {
if ((x
is
+ x
js
)R
ij
> 1)
Negotiate(i, j, s) }
3. U
w
and H
*
Compute U
w
and H
*
using Equations (4) and (5).
Subroutine Negotiate(a, b, p) {
x
is
= 1, x
js
= 0
Compute U and store it in temp
1
.
x
is
= 0, x
js
= 1
Compute U and store it in temp
2
.
if temp
1
< temp
2
x
is
= 1, x
js
= 0
}
TABLE II
THE COOPERATIVE GAME-BASED ALGORITHM WITHOUT TRANSMISSION
POWER CONTROL.
1. Initialization For the minimum F(X),
- (X) = F(X)
- U
*
=
iv
w
i
|S|, H
*
=
iv

sS
L
is
For finding <U
*
, H
w
>,
- (X) = -1
iv
w
i
ln(
sS
x
is
).
For all,
- min(X) =
- Assign all idle spectrum units to sensors.
2. Winner selec-
tion on each
spectrum
for each sS {
for each iV {
for each j ( i) V {
x
js
= 0 }
x
is
= 1
Compute (X).
if ((X) < min(X)) {
min(X) = (X)
winner = i }
x
is
= 0; }
Assign spectrum unit s to winner. }
3. U
*
and H
w
For finding <U
*
, H
w
>, U
*
= (X) and H
w
= -1

iv

sS
x
is
L
is
.
On the contrary, it is difficult to determine optimal F(X)
and <U
*
, H
w
> in reasonable time
1
. Therefore we adopt a coo p-
erative game to find minimum F(X) and <U
*
, H
w
> approxi-
mately in reasonable time. The cooperative game-based algo-
rithm can be described separately whether each sensor can
control its transmission power or not. Table II describes the
algorithm where none of sensors can control their transmission
power; Table III describes the algorithm where each sensor
has an ability to control its transmission power. The time
complexities of both algorithms are O(N
2
M), where N and M
are the number of sensors and the number of idle spectrum

1 The problem of determining <U
*
, H
w
> is similar with the graph coloring problem.
3
bands, respectively. However, in this algorithm, it is impossi-
ble to improve H
w
without deteriorating U
*
. Thus we only
compute H
w
from the results of U
*
.
In both the algorithms, as described in the Step 1, we only
use their lower bounds in order to avoid computing the opti-
mal U
*
and H
*
in every steps of coalition and negotiation.
Nevertheless, we observe that the obtained solutions (i.e., <U,
H>) are satisfactory.
If we assume that none of sensors can control their trans-
mission power, then it means that every spectrum unit can be
occupied by only one sensor. Thus, we select a winner among
all the competitors for each spectrum unit while improving the
objective: minimum F(X) or maximum proportional fairness.
TABLE III
THE COOPERATIVE GAME-BASED ALGORITHM WITH TRANSMISSION POWER
CONTROL.
1. Initialization For the minimum F(X),
- (X) = F(X)
- U
*
=
iv
w
i
|S|, H
*
=
iv

sS
L
is
For finding <U
*
, H
w
>,
- (X) = -1
iv
w
i
ln(
sS
x
is
).
For all,
- Assign all idle spectrum units to sensors.
- negotiated[i] = 0 for all iV.
2. Forming coali-
tions
Copy X to Y.
Sort iV in accordance with w
i
and arrange the
index k=1, , |V| from the largest to smallest.
for k=1, , |V| {
if (!nogotiated[k]) {
min(Y) =
for each j ( i) V {
if (!negotiated[j]) {
Negotiate(k, j, Y) and compute (Y).
if ((Y) < min(Y)) {
min(Y) = (Y)
coalition[k] = j } } }
negotiated[k] = 1
negotiated[coalition[k]] = 1 } }
3. Negotiation in
each coalition
for each i V {
Negotiate(i, coalition[i], X) }
if X satisfies equation (12) goto Step 4; else
goto Step 2.
4. U
*
and H
w
For finding <U
*
, H
w
>, U
*
= (X) and H
w
= -1

iv

sS
x
is
L
is
.
Subroutine Negotiate(a, b, Z) {
for each sS {
if ((z
as
+ z
bs
)R
ij
> 1)
z
as
= 1, z
bs
= 0
Compute (Z) and store it in
1
(Z).
z
as
= 0, z
bs
= 1
Compute (Z) and store it in
2
(Z).
if (
1
(Z) <
2
(Z))
z
as
= 1, z
bs
= 0 } }
If each sensor can control its transmission power so that it
does not interfere with other sensors beyond its targeted range,
then we can form a coalition at first (Step 2). The sequence of
forming coalition depends on the weight of each sensor. It
means that a sensor having the largest weight finds its coali-
tion partner first. Then the sensor negotiates the spectrum
ownership with its coalition partner for improving the main
objective (Step 3). These processes will continue until the
constraint in Equation (12) is not violated.
VI. NUMERICAL EXPERIMENTS
Ideally evaluating the qualities of MGT solutions - <U,
H> and U
*
, which are determined by the cooperative game -
should be compared with the optimal solutions, i.e., the opti-
mal U
*
and H
*
. We can find optimal H
*
easily but it is impos-
sible to extract the optimal U
*
in a reasonable time. Thus, we
compare U and U
*
determined by the cooperative game with a
crude upper bound obtained by LP-relaxation. In order to find
the upper bound, the nonlinear programming for optimal U
*
should be approximated in a linear form. Then we use GLPK
[19] to solve the approximated linear program.
For the experiment, we generate random topologies 40
sensors, and define <source, destination> pairs randomly. The
main experimental parameters are listed in Table IV.
TABLE IV
MAIN EXPERIMENTAL PARAMETERS.
Experimental parameter Value
Topology size (N) 40 sensors
Number of idle spectrum
units
271
Transmission power
control
With or without
P
hold
0.1
Weight
Assign a random weight between 0.1 and
100 on each sensor
Q 2
P
hold
: probability that each sensor holds a spectrum unit at its previous transmission phase.
TABLE V
EXPERIMENTAL RESULTS.
<-U
*
, -H
w
> <-U
w
, -H
*
> <-U, -H>
LP bound
of -U
*
WO/T.P.
<3784.31,
26>
<4133.67,
274>
<4080.36,
272>
8982.25
W/T.P.
<7679.41,
152>
<6220.94,
684>
<7622.12,
678>
8982.25
WO/T.P.: without transmission power control
W/T.P.: with transmission power control
Table V lists the results as a tuple of U and H. The col-
umn titled as <-U
*
, -H
w
> shows the results when we optimize
only U using the algorithms in Tables II and III. The column
<-U
w
, -H
*
> shows the results when we optimize only H using
the algorithm in Table 1. The column <-U, -H> shows the
results of MGT using the algorithms in Tables II and III. The
last column lists the results of LP-relaxation. The results show
that the improvement obtained in one objective penalizes the
other, where the result of MGT becomes balanced between the
different objectives. In addition, the solutions of U are very
close to the LP bound of U with a factor of approximately 2
for without transmission power control, where it is less than
1 for with transmission power control.
In order to show that the spectrum bands are allocated
fairly in accordance with the weights, we need to count the
number of assigned spectrum units at each sensor according to
the three proposed algorithms. The results are shown in Figure
2. Figure 2(a) shows the results without transmission power
4
control, while Figure 2(b) shows the results with transmission
power control. Figure 2(a) shows that the MGT with the coop-
erative game allocates the spectrum bands fairly according to
the sensors weight. However, in Figure 2(b), we see that the
number of assigned spectrum units does not correspond with
their respective weights in some sensors. This is due to the
fact that these sensors interfere with fewer sensors than other
sensors. Therefore, such sensors have been allowed to occupy
more spectrum units than others, irrespective of their weights.
(a) Without transmission power control.
(b) With transmission power control
Figure 2. The number of assigned spectrum units at each sensor in
accordance with the algorithms.
VII. CONCLUSIONS AND FUTURE WORK
In this paper, we use the MGT to allocate spectrum bands
in resource-constrained wireless sensor network: (1) as fairly
as possible, (2) achieving maximum spectrum utilization, (3)
reflecting the priority, and (4) with infrequent spectrum
handoff. We adopt a cooperative game-based algorithm with
the complexity of O(N
2
M) to solve the MGT in a reasonable
time. The experimental results of MGT are well balanced be-
tween the two objectives: maximizing the proportional fair-
ness and minimizing the number of spectrum handoffs. We
also compare the solutions of MGT with LP bounds, and ob-
serve that the algorithm provides near optimal solutions.
Although the proposed scheme operates in a centralized
manner, it may also be feasible to allocate spectrum bands in a
distributive manner by combining with a non-cooperative
game algorithm. Since the non-cooperative game algorithm
operates on a resource-constrained sensor, it should be imple-
mented as less computationally complex as possible with low
message overhead, which may become part of our future re-
search work.
ACKNOWLEDGEMENT
This work was carried out during the tenure of an ERCIM
Alain Benesoussan Fellowship Programme.
REFERENCES
[1] Zhao, Q., and Sadler, B.: A survey of Dynamic Spectrum Access. IEEE
Signal Processing Magazine, vol. 79, no. 3, pp. 79-89 (2007).
[2] Haykin, S.: Cognitive Radio: Brain-Empowered Wireless Communica-
tions. IEEE Journal on Selected Areas in Communications, vol. 23, no.
2 (2005).
[3] OMNET++ 3.3, http://www.omnetpp.org.
[4] Katzela, I., and Naghshineh, M.: Channel Assignment Schemes for
Cellular Mobile Telecommunication Systems: A Comprehensive Sur-
vey. IEEE Personal Communications, vol. 3, pp. 10-31 (1996).
[5] Gao, S., Qian, L., Vaman, D., and Qu, Q.: Energy Efficient Adaptive
Modulation in Wireless Cognitive Radio Sensor Networks. Proc. IEEE
ICC (2007).
[6] Geirhofer, S., Tong, L., and Sadler, B.: A Measurement-based Model for
Dynamic Spectrum Access. Proc. IEEE MILCOM (2006).
[7] Sunar, M., Kahraman, R.: A Comparative Study of Multiobjective Op-
timization Methods in Structural Design. Turk J. Engin. Environ. Sci.,
vol. 25 (2001).
[8] Han, Z., Ji, Z., and Liu, R.: Fair Multiuser Channel Allocation for
OFDMA Networks Using Nash Bargaining Solutions and Coalitions.
IEEE Trans. on Communications, vol. 53, no. 8, pp. 1366-1376 (2005).
[9] Pioro, M., and Medhi, D.: Routing, Flow, and Capacity Design in
Communication and Computer Networks. Morgan Kaufmann (2004).
[10] Cao, L., and Zheng, H.: Distributed Spectrum Allocation via Local
Bargaining. Proc. IEEE SECON (2005).
[11] Etkin, R., Parekh, A., and Tse, D.: Spectrum Sharing for Unlicensed
Bands. Proc. IEEE DySPAN (2005).
[12] Huang, J., Berry, R., and Honig, M.: Auction-based Spectrum Sharing.
ACM/Springer Mobile Networks and Apps., pp. 405-418 (2006).
[13] Ji, Z., and Liu, K.: Collusion-Resistant Dynamic Spectrum Allocation
for Wireless Networks via Pricing. Proc. IEEE DySPAN (2007).
[14] Zhao, Q., Tong, and L., Swami, A.: Decentralized Cognitive MAC for
Dynamic Spectrum Access. Proc. IEEE Symposium on New Frontier in
Dynamic Spectrum Access Networks (2005).
[15] Geirhofer, S., Tong, L., and Sadler, B.: Dynamic Spectrum Access in
WLAN Channels: Empirical Model and Its Stochastic Analysis. Proc.
International Workshop on Technology and Policy for Accessing Spec-
trum (2006).
[16] Weiss, T., and Jondral, F.: Spectrum Pooling: An Innovative Strategy
for Enhancement of Spectrum Efficiency. IEEE Communication Maga-
zine, vol. 42, pp. 8-14 (2004).
[17] Berthold, U., Jondral, F.: Guidelines for Designing OFDM Overlay
Systems. Proc. IEEE Symposium on New Frontier in Dynamic Spectrum
Access Networks (2005).
[18] Tang, H.: Some Physical Layer Issues of Wide-band Cognitive Radio
Systems. Proc. IEEE Symposium on New Frontier in Dynamic Spectrum
Access Networks (2005).
[19] GLPK: GNU Li n e a r Pr o g r a mmi n g To o l k i t ,
http://www.gnu.org/software/glpk.
5

You might also like