An Optimal Data Fusion Rule

in Cluster-Based Cooperative Spectrum Sensing

Hiep-Vu Van and Insoo Koo
School of Electrical Engineering, University of Ulsan
680-749 San 29, Muger 2-dong, Ulsan, Republic of Korea
[email protected]
http://mcsl.ulsan.ac.kr

Abstract. In this paper, we consider a cluster-based cooperative spectrum sensing approach to improve the sensing performance of cognitive
radio (CR) network. In the cluster-based cooperative spectrum sensing,
CR users with the similar location are grouped into a cluster. In each
cluster, the most favorable user namely cluster header, will be chosen
to collect data from all CR users and send the cluster decision to common receiver who makes a nal decision on the presence of primary user.
In the cluster-based cooperative spectrum sensing, data fusion rule in
the cluster takes an important role to reduce the rate of reporting error. Subsequently we propose optimal fusion rule for each cluster header
with which we can minimize the sum of probability of false alarm and
probability of missed detection in each cluster header.
Keywords: Cognitive radio, Cooperative spectrum sensing, Optimal
data fusion, Improvement sensing performance.

Introduction

Nowadays, wireless communication is applied in more and more applications of

many elds in our modern life such as military, entertainment, communication
and so on. Actually, in wireless network the licensed devices often occupy almost range of frequency, but they use those frequency bands with under 100%
capability. Especially, in some cases this utilization is just few percentages [1].
Undoubtedly, frequency band is a limited resource. Thereby, all of frequency
bands should be used more eectively by increasing their utilization proportion.
Recently, CR technology is used as a useful tool for limited frequency bank [2],
[3]. By using this technology, the available frequency from PU can be detected
and used by CR users and otherwise CR users should vacate their occupied frequency when the presence of PU is detected. Therefore, sensing the status of PU
is a prerequisite of CR technology. The best sensing performance will let every
CR user know exactly whether PU is present or not and use free frequency band
from PU without any harmful inuence.


Corresponding author.

D.-S. Huang et al. (Eds.): ICIC 2009, LNAI 5755, pp. 708717, 2009.
c Springer-Verlag Berlin Heidelberg 2009


An Optimal Data Fusion Rule

709

In practice, the performance of individual sensing in CR network can be deteriorated by deep fading and shadowing. This problem can be solved by allowing
some CR users to perform cooperative spectrum sensing [4]-[6]. In cooperative
spectrum sensing, we rely on the variability of signal strength at various locations of CR users to obtain a better performance of detecting the PUs signal
in a large CR network with sensing information integrated between neighbors
compared to individual sensing [7]. Cooperative spectrum sensing process often
has 3 steps: sensing, reporting and making a decision. In sensing step, all CR
users perform spectrum sensing individually and later in reporting step, all local sensing observations are sent to common receiver. In nal step, making a
decision, common receiver uses a data fusion rule to fuse all local observations
together as a global decision about presence of PU.
When some CR users coordinate to perform cooperative spectrum sensing,
more accurate detection can be obtained. However, when the local observations
are forwarded to a common receiver through fading channels, the sensing performance can be severely degraded. In order to overcome this problem, Sun et
al. have proposed a cluster-based cooperative sensing method[8]. In this method,
few CR users with similar location are collected into a cluster. In each cluster,
a favorable user is selected to be cluster header. Cluster header will receive local sensing information from all CR users to make a cluster decision and later
report to common receiver. This approach really improves the sensing performance in comparison with conventional method. However, in the reference paper
[8] OR-rule is used in both cluster header and common receiver, which is not
optimal.
In order to improve the performance of cluster-based cooperative sensing, in
the paper we propose optimal data fusion rule for cluster header with which
we can nd optimal thresholds for both CR users in cluster and cluster header
such that we can minimize the sum of probability of false alarm and probability
of missed detection in each cluster header. We also consider three data fusion
rules at common receiver: haft-voting rule, AND-rule and OR-rule to prove the
eciency of the proposed optimal fusion rule for cluster header.

System Model

This paper considers a CR network included K clusters with nj (j = 1, 2 . . . K)

CR users for each cluster and a common receiver. In a cluster, the position of all
CR users is close together. Therefore, we assume that all CR users in the same
cluster have the similar channel communicated with PU (same SNR, j ) and
common receiver (same SNR, j ) as shown in F igure 1. The common receiver
functions as a base station (BS) which manages the cognitive radio network and
all associated cognitive radio.
This network focuses to sense the presence of PU by performing cooperative
spectrum sensing. In the network, each CR users should sense individually by
using one of detection methods such as matched lter detection, energy detection, feature detection, and so on [9], [10]. In those detection methods, if the CR

710

H.-V. Van and I. Koo

Fig. 1. System Model

user has limited information about signals of the PU, then the energy detection
is optimal [10]. In the energy detection, the radio frequency energy in the sensing
channel is collected in a xed bandwidth W over an observation time window T
to decide whether the channel is utilized or not. We assume that each CR user
performs local spectrum sensing using energy detector independently and the
sensing channel is time-invariant during the sensing process.
In the i th (i = 1, 2 . . . nj ) CR user, the local spectrum sensing is to decide
between two following hypotheses.


H0 : xi (t) = ni (t)
H1 : xi (t) = hi s(t) + ni (t)

(1)

where xi (t) is the observation signal at the i th CR user, s(t) is the signal of
PU, ni (t) is the additive white Gaussian noise (AWGN), and hi is the complex
channel gain of the sensing channel between the PU and the i th CR user.
In energy detection, the collected energy in the frequency domain is denoted
by Ei , which serves as a decision statistic and has the following distribution [11].


H0 : Ei = 22u
H1 : Ei = 22u (2i )

(2)

where 22u denotes a central chi-square distribution with 2u degrees of freedom

and 22u (2i )denotes a noncentral chi-square distribution with 2u degrees of freedom and a non-centrality parameter 2i . The instantaneous SNR of the received
signal at the i-th CR user is i and u = T W is the time-bandwidth product. In
this paper, we use u = 10.

An Optimal Data Fusion Rule

711

Cluster-Based Cooperative Spectrum Sensing

For the aim of enhancing the reliability ratio of sensing performance, we consider a cluster-based cooperative spectrum sensing method which is recognized
through following steps [1].
Step 1: All CR users in each cluster perform local spectrum sensing individually using energy detector and sends their local observations to the cluster
header.
Step 2: The cluster header receives those local observations and later makes
a cluster decision.
Step 3: The cluster decisions of each cluster are reported to the common
receiver by their cluster header. After that a global decision will be made.
3.1

Local Spectrum Sensing with Energy Detection

For using the energy detector, the average probability of false alarm (Pf,i,j ), the
average probability of detection (Pd,i,j ) and the average probability of missed
detection (Pm,i,j ) of local decision are given respectively, by [11].
Pf,i,j = Prob{Ei > j |H0 } =

(u, 2j )
(u)

(3)



Pd,i,j = Prob{Ei > j |H1 } = Qu ( 2j , j )

(4)

Pm,i,j = 1 Pd,i,j

(5)

and
where j and j denote the energy threshold and the instantaneous SNR of CR
users in the j-th cluster respectively,
(a, x) is the incomplete gamma function

which is given by (a, x) = x ta1 et dt, (a) is the gamma function, and
Qu (a, b) is the generalized Marcum Q-function which is given by: Qu (a, x) =
 u t2 +a2
1
2
Iu1 (at)dt, Iu1 (.) is the modied Bessel functions of the rst
au1 x t e
kind and order u-1.
3.2

Cluster Header Decision

All 1-bit decisions from CR users in a cluster are fused together according to
following logic rule.

nj

Gj,i thj
Bj = 1,
i=1
where thj [1, 2 . . . nj ]
(6)
nj

Gj,i < thj

Bj = 0,
i=1

where thj is a optimal threshold of the j th cluster header. Gj,i = {0, 1} is the
local decision of the i th user in the j th cluster.

712

H.-V. Van and I. Koo

It can be easy seen that if thj = 1 or thj = nj the data fusion rule will be
OR-rule or AND-rule respectively.
From the assumption that all CR users in the j th cluster has the same
SNR, we have the same probability of false alarm (Pf,j ), probability of detection
(Pd,j ) and also probability of missed detection Pm,j = 1 Pd,j in all CR users
of the j th (j = 1, 2 . . . K) cluster.
Therefore, the probability of false alarm in the j th cluster header (Qf,j ) is
given by [12]
Qf,j = Prob {H1 |H0 }
nj

(7)
=
C l P l (1 P )nj l
l=thj

nj

f,j

f,j

Similarly, the probability of missed detection in the j th cluster header (Qm,j )

is given by
Qm,j = Prob {H0 |H1 }
nj

(8)
=1
C l P l (1 P )nj l
l=thj

nj

d,j

d,j

In this paper, one of our objectives is to nd the optimal energy threshold j for
all CR users in the j th cluster (j = 1, 2 . . . K) as well as the optimal threshold
thj of the j th cluster header. Here, we dene the optimal thresholds as opt
j
and thopt
that
can
minimize
(Q
+
Q
)
of
the
j

th
cluster
header,
such
that
f,j
m,j
j
we have
opt
arg min(Qf,j + Qm,j )
(9)
[opt
j , thj ] =
opt
,thopt
j
j

and thopt
can be found by numerical method.
opt
j
j
3.3

Global Decision

In common receiver, the global decision will be created by integrating all received
cluster decisions according the below logic rule.

H = H1 :
Bj thg
(10)
j=1

H = H0 : otherwise
where Bj = {0, 1} is the decision of j th cluster and thg is the threshold of
common receiver.
For the sake to prove the eciency of optimal rule in cluster, we consider
three values of threshold thg as follows.
No. thg Fusion Rule
1
1
OR-rule
2 K
+1
Half-Voting
rule
2
3
K
AND-rule

An Optimal Data Fusion Rule

713

Here, we let Qej be the error probability when cluster decision Bj is reported
to the common receiver but the decision Bj is obtained.
For the case of BPSK and a given j , the error probability of the j th cluster
header over Rayleigh fading channels can be given by [8]

Qej = 0 Qej|max,j f (max,j )dmax,j



n1
m
j
n
1 nm+
=
Cn1 (1)nm1 2(nm)
j

(11)

m=0

where Qej is error probability of the channel between the j th cluster header
and common receiver.
Commonly, the CR user with the highest SNR will be chosen to be the cluster
header. However, as the assumption that all CR users in a cluster have the
similar SNR of the channel with common receiver, the cluster header will be
chosen randomly within all CR users in cluster.
In common receiver, the global decision will be made by integrating all cluster
decisions with their channel errors according to respective data fusion rule. Those
are assumed as below:
Data fusion rule is OR-rule (thg = 1)
By using this rule, the global false alarm probability (Qf ) and global missed
detection probability (Qm ) in common receiver are given by following, respectively.
Qf,or = 1

((1 Qf,j (1 Qe,j ) + Qf,j Qe,j )

(12)

(Qm,j (1 Qe,j ) + (1 Qm,j )Qe,j )

(13)

j=1

and
Qm,or =

K

j=1

Data fusion rule is Half-Voting rule (thg = K

2 + 1)
Actually, when we use half-voting rule to make global decision, we will confront with a dicult problem of calculation the global probabilities of false
alarm (Qf,half ) and missed detection (Qm,half ) for the general case of K
clusters. Therefore, we consider the case K = 4 and derive (Qf,half ) and
(Qm,half ) for K = 4 as follows.
Qf,half = Qf e,1 .Qf e,2 .Qf e,3 .(1 Qf e,4 )
+Qf e,1 .Qf e,2 .(1 Qf e,3 ).Qf e,4
+Qf e,1 .(1 Qf e,2 ).Qf e,3 .Qf e,4
+(1 Qf e,1 ).Qf e,2 .Qf e,3 .Qf e,4
+Qf e,1 .Qf e,2 .Qf e,3 .Qf e,4

(14)

714

H.-V. Van and I. Koo

and

where

Qm,half = Qme,1 .Qme,2 .(1 Qme,3 ).(1 Qme,4 )

+Qme,1 .(1 Qme,2 ).Qme,3 .(1 Qme,4 )
+Qme,1 .(1 Qme,2 ).(1 Qme,3 ).Qme,4
+(1 Qme,1 ).Qme,2 .Qme,3 .(1 Qme,4 )
+(1 Qme,1 ).Qme,2 .(1 Qme,3 ).Qme,4
+(1 Qme,1 ).(1 Qme,2 ).Qme,3 .Qme,4
+Qme,1 .Qme,2 .Qme,3 .(1 Qme,4 )
+Qme,1 .Qme,2 .(1 Qme,3 ).Qme,4
+Qme,1 .(1 Qme,2 ).Qme,3 .Qme,4
+(1 Qme,1 ).Qme,2 .Qme,3 .Qme,4
+Qme,1 .Qme,2 .Qme,3 .Qme,4

(15)

Qf e,j = (1 Qf,j )Qej + (1 Qej )Qf,j

Qme,j = (1 Qm,j )Qej + (1 Qej )Qm,j
(j = 1, 2, 3, 4)

(16)

Data fusion rule is AND-rule (thg = K)

In this case, the global probabilities of false alarm (Qf,or ) and global probabilities of missed detection (Qm,or ) in common receiver are given by following.
Qf,and =

(Qf,j (1 Qe,j ) + (1 Qf,j )Qe,j )

(17)

j=1

and
Qm,and = 1

((1 Qm,j )(1 Qe,j ) + Qm,j Qe,j )

(18)

j=1

Simulation Results

In this simulations, our concern is the eciency of sensing performance in a

cluster as well as in common receiver. Therefore, we consider the j th cluster
with ten CR users (n = 10) and assume SNR uniformly within the range of
5dB to 10dB at each CR users. In the j th cluster header, optimal data fusion
rule will be used to create cluster decision. For the sake of comparison, we also
provide the sensing performance with the AND-rule and OR-rule.
Figure 2 shows the sensing performance of the cluster corresponding Optimalrule, OR-rule and AND-rule. In which, we can see that the Optimal-rule can
signicantly reduce the error probability in cluster Qf,j +Qm,j which is expressed
by eqn. (7) and (8), when compared with OR-rule or AND-rule in each value of
SNR. When the SNR is about 10dB the reporting error can achieve an acceptable
value - less than 0.002 with Optimal-rule, but it achieves the value bigger than
0.015 with both AND-rule and OR-rule.

An Optimal Data Fusion Rule

715

10

Qfj+Qmj

10

10

Optimalrule
ORrule
ANDrule

10

10

SNR

Fig. 2. Reporting error Qf,j +Qm,j in the j th cluster corresponding dierent decision
fusion rules and dierent values of SNR

10

Q +Q

10

Optimalrule
ORrule
ANDrule

10

6
8
Average SNR

10

12

14

Fig. 3. Reporting error Qf + Qm in common receiver versus dierent values of average

SNR and dierent decision fusion rules in Cluster-Case 1

Qf+Qm

10

10

Optimalrule
ORrule
ANDrule

10

6
8
Average SNR

10

12

14

Fig. 4. Reporting error Qf + Qm in common receiver versus dierent values of average

SNR and dierent decision fusion rules in Cluster-Case 2

716

H.-V. Van and I. Koo

0

10

Optimalrule
ORrule
ANDrule

Qf+Qm

10

10

10

6
8
Average SNR

10

12

14

Fig. 5. Reporting error Qf + Qm in common receiver versus dierent values of average

SNR and dierent decision fusion rules in Cluster-Case 3

In the other hand, for our concern of sensing performance in common receiver,
we consider network with four clusters which composed ten CR users for each
cluster and the SNR of channels between PU and four clusters are 7.5dB, 8.0dB,
8.5dB and 9.0dB respectively. Moreover, SNR of channels between four clusters
and common receiver will be changed by following table.
Channel 1
2
3
4
SNR (dB) 0.5*m 0.75*m 1*m 1.25*m
where m has value in frame within 1 and 15.
In this step, we consider three cases of applied data fusion rules in cluster as
well as in common receiver, which can be determined as like below table.
Case In Cluster In Common Receiver
Optimal-Rule
1
OR-Rule
OR-Rule
AND-Rule
Optimal-Rule
2
OR-Rule
AND-Rule
AND-Rule
Optimal-Rule
3
OR-Rule
Half-Voting Rule
AND-Rule
In all cases of data fusion rule, cooperative spectrum sensing performance, which
are shown in F igure 3 5, can achieve better performance with optimal-rule in
cluster. Specially, from Figure 5 we can get the best performance with optimalrule and half-voting rule in cluster and common receiver respectively.

An Optimal Data Fusion Rule

717

Conclusion

In this paper, we consider cluster-based cooperative spectrum sensing as well as

data fusion rule to improve sensing performance. In cluster, we nd an optimalrule which is proved to be the best rule with the smallest value of reporting error.
Moreover, the optimal-rule, which is utilized in cluster, and the half-voting rule,
which is utilized in common receiver, are the best combination with the minimum
ratio of reporting error.

Acknowledgement
This work was supported by the Korea Research Foundation Grant funded by
the Korean Government(MOEHRD) (KRF-2009-0063958).

References
1. Federal Communications Commission: Spectrum Policy Task Force. Rep. ET
Docket, 02135 (2002)
2. Mitola, J., Maguire, G.Q.: Cognitive Radio: Making Software Radios More Personal. IEEE Pers. Commun. 6, 138 (1999)
3. Haykin, S.: Cognitive Radio: Brain-empowered Wireless Communications. IEEE
J. Select. Areas Commun. 23, 201220 (2005)
4. Ganesan, G., Y. Li, G.: Cooperative Spectrum Sensing in Cognitive Radio Networks. In: Proc. IEEE Symp. New Frontiers in Dynamic Spectrum Access Networks
(DySPAN5), Baltimore, USA (2005)
5. Ghasemi, A., Sousa, E.S.: Collaborative Spectrum Sensing for Opportunistic Access in Fading Environments. In: Proc. IEEE Symp. New Frontiers in Dynamic
Spectrum Access Networks (DySPAN5), Baltimore, USA, vol. 81, pp. 131136
(2005)
6. Mishra, S.M., Sahai, A., Brodersen, R.: Cooperative Sensing Among Cognitive
Radios. In: Proc. IEEE Int. Conf. Commun., Turkey, vol. 4, pp. 16581663 (2006)
7. Cabric, D., Mishra, S.M., Brodersen, R.W.: Implementation Issues in Spectrum
Sensing for Cognitive Radios. In: Proc. of Asilomar Conf. on Signals, Systems and
Computers, Pacic Grove, CA, USA, pp. 772776 (2004)
8. Sun, C., Zhang, W., Letaief, K.B.: Cluster-based Cooperative Spectrum Sensing for
Cognitive Radio Systems. In: Proc. IEEE Int. Conf. Commun., Glasgow, Scotland,
UK, pp. 25112515 (2007)
9. Hur, Y., Park, J., Woo, W., Lim, K., Lee, C.H., Kim, H.S., Laskar, J.: A Wideband Analog Multi-resolution Spectrum Sensing (MRSS) Technique for Cognitive Radio (CR) Systems. In: Proc. IEEE Int. Symp. Circuit and System, Greece,
pp. 40904093 (2006)
10. Sahai, A., Hoven, N., Tandra, R.: Some Fundamental Limits on Cognitive Radio.
In: Proc. Allerton Conf. on Communications, control, and computing, Monticello
(2004)
11. Digham, F.F., Alouini, M.S., Simon, M.K.: In the Energy Detection of Unknown
Signals Over Fading Channels. In: Proc. IEEE Int. Conf. Commun., Anchorage,
AK, USA, pp. 35753579 (2003)
12. Zhang, W., Mallik, R.K., Letaief, K.B.: Cooperative Spectrum Sensing Optimization in Cognitive Radio Networks. In: Proc. IEEE Int. Conf. on Commun., Beijin,
pp. 34113415 (2008)