You are on page 1of 6

2015 1st International Conference on Telematics and Future Generation Networks (TAFGEN)

Optimality of the HDC Rules in Cooperative


Spectrum Sensing for Cognitive Radio Network
Wasan Kadhim Saada,b,*, Mahamod Ismaila , Rosdiadee Nordina and Ayman A. El-Salehc
Department of Electronics, Electrical and System Engineering, Faculty of Engineering and Build Environment, Universiti
Kebangsaan Malaysia, 43600 Bangi, Selangor, Malaysia.
[e-mail: wasan, mahamod, adee@eng.ukm.my]
b
Department of Communications Engineering, Najaf Technical College, Foundation of Technical Education, Baghdad, Iraq.
c
Faculty of Engineering, Multimedia University, 63100 Cyberjaya, Selangor, Malaysia.
[e-mail: ayman.elsaleh@mmu.edu.my]
*Corresponding author: Wasan Kadhim Saad
a

essential function in enabling cognitive radio networks


(CRNs) required for detecting the probably-weak primary
radio signals with sufficient reliability. Hence, through an
environment-adaptive sensing operation and without causing
harmful interference to PUs, a CRN will be capable of filling
available spectrum holes, thus providing an opportunistic
radio access to the users it serve [1]. The most common major
local sensing techniques for spectrum sensing (SS) in CRN is
energy detection (ED) which does not require prior knowledge
of PU signal. Also, ED is classified as a simple detection
technique with a short sensing time requirement. However, it
performs poorly in low signal-to-noise ratio (SNR)
environment. Thus, ED owns fast processing capability, low
computational complexity and high flexibility [2]. To achieve
higher sensing performance, collaborative SS is required to
improve the detection probability, to shorten the detection
time and to attain improved data fusion results, thereby
improving the sensitivity of the cognitive receiver [3]. The
problem of combining multiple sensing decisions from
collaborated SUs, which is commonly known as the data
fusion problem, has also been considered in [4-5]. In [4], a
logic OR combining rule based HDC is presented for CSS to
combine decisions from several SUs. In [5], the authors
compared two simple schemes of HDC, namely AND rule and
OR rule detectors and they claimed that OR rule is better than
AND rule at low SNR. In this study, and based on above
observations, we adopt an ED to realize SS and the sensing
performance of ED is then compared with the LRT.
Furthermore, we consider the optimality problem of CSS by
using optimal decision combining rules with ED in more
detail. In addition, the CSS with HDC (e.g. AND, OR, and
Majority) rules is investigated in this work. This paper is
organized as follows: Section II introduces the concept of SS
in CRNs by using ED and LRT models for channel sensing.
Section III presents the derivations of detection and false
alarm probabilities for CSS with HDC rules. Section IV
presents the Bayes risk function for optimality of CSS by
using optimal combining rule. Section V conveys a discussion
on the obtained results. Finally, brief concluding remarks of
the study is drawn in Section VI.

AbstractCognitive Radio (CR) sensing has been widely


considered as a spectrum scanning mechanism that allows
secondary users (SUs) or cognitive radio users to use detected
spectrum holes caused by primary user (PU) absence. Hard
decision combining (HDC) schemes are proposed to combine the
sensing decisions of the collaborated users to come out with a
global binary decision on the presence or absence PUs. This
paper presents an analytical study on the optimality of HDC
rules at which the Bayes risk function is minimized. In this work,
the sensing performance of energy detection (ED) is also
evaluated in two cases; when the estimated noise power is
perfectly known at the SU receiver and when noise uncertainty is
present at the SU receiver. The sensing performance of the ED
and likelihood ratio test (LRT) of local spectrum sensing (SS) is
first compared. Then, the performance of cooperative spectrum
sensing (CSS) employing k-out-of-N combining rule has been
analyzed. A mathematical derivation of an optimal decision
combining rule under low Bayes risk has been formulated.
Computer results show that the sensing performance of the ED
method slightly outperforms the LRT method within the lower
range of probability of false alarm. However, the two methods
exhibit almost similar sensing performance within the higher
range of probability of false alarm. On the other hand, at lower
values of probability of detection, the OR combining rule exhibits
the best detection performance over the Majority and AND rules.
Finally, it has been found that the optimal decision combining
rule to achieve lower Bayes risk function is the Majority
distributed decision rule.
Keywordscognitive radio network; cooperative spectrum
sensing; hard decision combining rules; likelihood ratio test;
energy detection.

I. INTRODUCTION
The need for radio spectrum is critically increasing with
the significant increase of number of users using wireless
services. Therefore, a key technology that can help to mitigate
scarcity of spectrum is cognitive radio (CR). CR technology is
realized through the existence of CR users that work as
secondary users (SUs) to gain access to the underutilize
frequency bands which have been assigned to the licensed
user, i.e., the primary user (PU). Spectrum sensing is an
2015 IEEE

978-1-4799-7315-6/15/$31.00 2015 IEEE

22

2015 1st International Conference on Telematics and Future Generation Networks (TAFGEN)

consider real-valued Gaussian distribution for both of signal


si (k ), and noise ni (k ) if the number of samples is large

II. LOCAL SPECTRUM SENSING FOR COGNITIVE RADIO


NETWORK
At first, we consider the case of a single SU node only,
thus local sensing is taken into account. Local SS can be
realized by means of ED or LRT, and with or without the
existence of noise uncertainty.

enough (e.g., M 10 in practice) [7], CLT says that the


PDFs of T (Y ) under both hypotheses ( H 0 and H 1 ) can be
approximated by a Gaussian distributed with mean

( 0 = n2 and 1 = ( + 1) n2 ) and
variance
2
2
( 02 = n4 and 12 = ( + 1) 2 n4 ) , respectively.
M
M

A. Energy Detection Model for Local Channel Sensing in


Cognitive Radio Network
At the receiver of an individual SU in a CRN, the local SS
method is realized by means of ED. During the entire frame
duration, it is assumed that the PU can be either present or
absent. Thus, the SS process is a binary hypothesis testing
problem. The received signal composition at the SU under this
binary hypothesis can be given by:

H0
H1

Yi (k ) = ni (k )
Yi (k ) = hi (k ) s( k ) + ni (k )

:
:

There are many sensing schemes proposes in the literature [89] assuming that the noise that a perfectly-known additive
white Gaussian noise (AWGN) is used to derive the false
alarm and detection probabilities. However, practically, noise
is not perfectly Gaussian distributed because of the channel
fading. This makes it difficult for the SU to get full
information, and hence the noise uncertainty should be
considered in each local sensing decision at every SU receive.
Therefore, the goal of this section is to determine the sensing
performance of the ED for two cases: when the estimated
2
noise power at the SU node n is perfectly known at the
receiver and for the presence of noise uncertainty in the
system. Thus, for the first case and with the chosen decision
threshold i , the probability of the detection for a target false
alarm probability can be derived and given as follow:

(1)

where ni (k ) is a zero-mean additive white Gaussian noise


(AWGN) that is assumed to be independent and identically
distribution
(i.i.d)
random
process,
i.e.,
2

ni (k ) ~ C( 0, n ) . s (k ) is the PU signal and it is


presumed to be an i.i.d random Gaussian process with zero
2

mean and variance s , thus, the primary signal s (k ) is

1 1
M

Q PF
PD ,i = Q

2
( + 1)

independent of the noise ni (k ) . hi (k ) is the amplitude gain


of the channel between the PU and SU that is presumed to be
0 under hypothesis H 0 and 1 under hypothesis H1 , and

where Q ( .) is the complementary distribution function of


the standard Gaussian. Equation (3) explains that higher
average received SNRs and higher PD ,i are assumed if the SUs

Yi (k ), is the output of the ED by the i th SU. Thus, the output


of the ED is given by [6]:
M 1

T (Yi ) = Yi (k )

keep the same PF ,i and M. Besides, it can be used to derive

(2)

the receiver operating characteristic (ROC) curve for the


conventional ED which refers to the most important
performance measure for a hypothesis testing problem.
Whereas for the second case, with assuming that the actual
noise power changes randomly from one detection slot to
another with the uniform distribution which can be
2
2
2
represented as n , n and it is unvarying
during the duration of the detection [10-11], where denotes
the noise power uncertainty coefficient in dB and > 1 .
Thus, the probability of false alarm and probability of
detection is given as follows:

k =0

So we tend to compare the output of the ED, i.e., T (Y ) ,


which services as decision test static with a threshold, , to
decide the existence or absence of the primary signal, then
make a local decision H 1 if T Y > and H 0 if

T (Y ) < .

regime, i =

( )

Since

we

s(k ) h
k =1

are
2

interested

in

low

SNR

M n2 , we also need to use

large number of samples, where signifies the received


average SNR of the PUs signal measured at the ith secondary
receiver of interest under hypothesis H 1 . We assume that all
users utilize ED as well as the same decision rule (i.e. same
threshold ). These assumptions render our scheme
suboptimal. However, they facilitate analysis and make
practical application easier. Hence, to approximate the
decision statistic as Gaussian distribution, central limit
theorem (CLT) can be used. According to the CLT, we

978-1-4799-7315-6/15/$31.00 2015 IEEE

(3)

i u
1

PF ,i ( ) =
Q
du
2 2 2
n2 2 n
u

n
M

n2

23

(4)

2015 1st International Conference on Telematics and Future Generation Networks (TAFGEN)

( + 1)u
1
du
PD ,i ( ) =
Q i
2
(5)

2
2 2
n
n2 2
( + 1) u
n

2 02 + 02 (m1 m0 )
PF ,i ( ) = Q i 1
.....
0 12 02

Therefore, for a targeted false alarm probability, the


probability of the detection is given by solving the integration
in the expressions above and can be written as follows:

i 12 02 + 12 (m1 m0 )
PD ,i ( ) = Q
....
1 12 02

n2

M
1
1
1
Q ( PF ) +
M
2
PD ,i ( ) = Q

(6)

n2

(Y) =

hypothesis

n2

(h
2

1
2
s

+u

n2

1
M 2

2
n

(9)

Y TY
exp
2 h 2 2 +u
s

du

(10)

Y Y
du
exp
2u
T

III. COOPERATIVE SPECTRUM SENSING WITH HARD DECISION


COMBINING (HDC) RULES FOR COGNITIVE RADIO NETWORKS

2
1

The hidden terminal problem is one of the most


challenging issues of applying SS when the CR is shadowed,
being in deep multipath fade or inside buildings with high
prevalence loss while a PU in the neighboring region. CR may
fail to observe the existence of the PU leading to the hidden
terminal problem where a SU accesses a PU assigned channel
causing harmful interference to it. In fact, multiple CR users
can be coordinated to address this issue and perform spectrum
sensing whereby users only share their last 1-bit decisions
( H 0 or H 1 ) rather than their decision statistics, in order to
reduce the communication overhead [12]. Thus, each ith SU
performs local spectrum measurements independently, makes
a binary decision Z i { 0 ,1 } for all i =1, ....., N and then
all the SUs forward their binary decisions to a common
receiver. Besides, all 1-bit decisions are combined to make a
last decision H 0 or H1 on the absence or existence of the PU
within the observed frequency band according to the following
logic rule:

the H 1 and H 0 of

Y = (Y1 , ...., YN ) , respectively. Thus, the local decision of


hypothesis H 1 is made when (Y ) > while the local
decision of hypothesis H 0 is made when (Y ) < . In this

E=

< k

H0

H1

T (Y ) k
i =1

(11)

It is worth mentioning that (k-out-of-N) fusion rules can be


realized in terms of logic functions and they are widely
utilized due to their simplicity of implementation. Commonly
used HDC rules such as AND (N-out-of-N), OR (1-out-of-N)
and Majority (N/2-out-of-N) rules can be found in [13]. The
cooperative probability of detection and probability of false

case, the probability of false alarm and probability of detection


is given by:

978-1-4799-7315-6/15/$31.00 2015 IEEE

M 2

(u )

f (Y H1 ) and f (Y H 0 ) is the probability density


under

(8)

Without loss of generality, with the existence of noise


uncertainties in the system, the LRT is given as follows:

(Yi 0 )2

2 2 ....
M
f (Y H1 ) 0
0


(7)
(Y ) =
=
exp
2
M
f (Y H 0 ) 1

i =1

(
Y
)

i 1

2 2

i =1
1

(PDF)

the product of the individual probability density since the


noise samples are statistically independent. Therefore, the
LRT is given by:

function

2 02 12 (m1 m0 )
+ Q i 1

1 12 02

( 0 , 1 ) and variance ( , ) for


each hypothesis H 0 and H 1 , respectively. The joint
probability density function of the M samples for Yi (k )
under each hypothesis H 0 and H 1 results can be found from

where

We consider a source whose output under hypothesis H 1 is


a constant voltage of value m, while its output under
hypothesis H 0 is zero for the problem of detecting of a known
signal in AWGA at each sensor. Thus, if we assume that we
have an observation Y ( k ) , which is a Gaussian
distribution with mean

i 12 02 02 (m1 m0 )
+ Q

0 12 02

B. Likelihood Detection Model for Local Channel Sensing in


Cognitive Radio Network

2
0

24

2015 1st International Conference on Telematics and Future Generation Networks (TAFGEN)

alarm performance for the ORrule without presence of noise


uncertainty in the CRN is given by,

C F = P 0 (C10 C00 ) , C D = (1 P0 ) (C01 C11 ) ,


P0 the probability is associated with the hypothesis H 0 , and
Cij refers to the cost of global decision being H i when H j is

where

N
1 1
M
Q PF
(12)
QD =1 1 Q

(
)
+
1
2

i =1

M
Q F = 1 1 Q ( + 1)Q 1 PD +
(13)

2
i =1

(C10 > C00 ) and (C01 > C11 ) ,


this implies that (C D > 0) and (C F > 0) and therefore,
C = C D C F = (1 P0 ) C01 + P0 C00 . Thus, in the case
when QF and QD are equiprobable, we may express B as
a function of k as follow,
present. Using the fact that

On the other hand, the cooperative detection performance in


CRN wise probabilities for the ANDrule is given by,

i
N i
N CF ,i (PF ,i ) (1 PF ,i ) ...
B (k ) = C +
(21)
i
N i
i = k i C D , i (PD ,i ) (1 PD , i )

1 1
M
Q PF

QD = Q

(
)
+
1
2

i =1


N
M
Q F = Q ( + 1) Q 1 PD +

2
i =1

N

(14)

The system is designed for obtaining decision rules that


minimizes B . Thus, in this section, we attempt to obtain the
optimal value of k that minimizes the B for k-out-of-N
combining rule. Therefore, we let ( PD ,i = PD ) , ( PF ,i = PF ) ,

(15)

And, the final decision for the cooperative detection


performance in CRN wise probabilities for the Majority rule
are given by:

QD =

N
N
2

j =0

QF =

N
N
2

j =0

(C F ,i = CF ) and (C D ,i = C D ) for (i = 1,...., N ) . Using this


notation, and by differentiating Equation (21), setting the
result equal to zero, rearranging terms, and by taking the
natural logarithm of both sides, we can get:

N
N
N

N j
+ j
N (1 PD , 0 ) 2 (PD , 0 ) 2 (16)
j
+
2

N
N
N

N j
2 + j (17)
2
1

P
P

N

F ,0
F ,0
2 + j

P
1 PD
1 PD
C
k ln
ln F (22)
k ln D + N ln
PF
1 PF
1 PF
CD

The value of k for the decision combining rules (k-out-of-N)


is given by rearranging the terms as:

While the generalized formula for overall detection probability


and false alarm probability at FC with the presence of the
noise uncertainty in the CRN are,
N

QD = CNj (1 PD ( ) )

N j

j =
N

QF = CNj (1 PF ( ))
j =

N j

(P

( ) )

(PF ( )) j

C 1 P N
F
ln F

C
1
P

D
D
=
k
P 1 PF

ln D
P
1
P

F
D

(18)

(19)

where, the optimal value of

where, PD ( ) and PF ( ) are the probabilities conditioned on


the noise power uncertainty ( ) in the CRN.

978-1-4799-7315-6/15/$31.00 2015 IEEE

if 0
otherwise

(24)

where, . denotes the standard ceiling function and the


commonly used fusion rules are AND (N-out-of-N), OR (1out-of-N), and Majority ( N + 1 2 -out-of-N) rules,

It is important to investigate the optimality of CSS by


using optimal fusion rule (k-out-of-N) for minimizing B
function. Let us define B as a function that we aim to
minimize and as in [14]:

B = C + CF QF CDQD

kopt is equal to:


k opt =
0

IV. BAYES RISK FUNCTION FOR OPTIMALITY OF COOPERATIVE


SPECTRUM SENSING

(23)

respectively. Identifying the optimum value of k for which


the detection errors are reduced is highly important. It can be
shown that the optimum value of k is based on the detection
threshold. The best rule is the AND rule, i.e., ( k = N ) for a

(20)

25

2015 1st International Conference on Telematics and Future Generation Networks (TAFGEN)

small fixed threshold. On the other hand, the OR rule ( k =1)


is said to be optimal for a fixed large threshold. When the
threshold for determining H1 equals k, the (k-out-of-N) rule is
equivalent to the Majority rule, i.e., (k = N 2) .
V. RESULTS AND DISCUSSION
The performance of the cooperative spectrum sensing is
evaluated using the complementary ROC curve, spectrum
usage and SNR requirements for different situations of interest
of the CR system. Fig. 1 explains that the complementary
ROC curves for the comparison between the sensing
performance of the ED and LRT in the case of no noise
uncertainty as well as in the case of noise uncertainty for
different noise power uncertainty (i.e., 1.02, and 1.04 dB). It
can be noticed from Fig. 1that the two methods have almost
similar performance where the curve of ED almost overlaps
with the LRT one. Besides, the ED method slightly exceeds
the LRT in the lower range of the false alarm probability.
Also, the performance for each ED and LRT deteriorates with
the increase of the noise power uncertainty. Fig. 2 explains the
ROC curves for comparing the detection performance of the
ED without and with noise uncertainty for different values of
noise uncertainty coefficient values (i.e., 1.02, 1.04 and 1.06),
SNR= -14 dB and M= 500. In Fig. 2, it is found that ED is very
sensitive to noise uncertainty where the performance gradually
reduces as the noise factor increases. In fact, the noise
estimation error significantly impedes reliable SS and with an
increase of the noise level, a serious degradation in the
detector's performance occurs. Therefore, SUs may need to
predict the spectrum band of interest regardless if the PUs is
absent or present. As a result, SU interference is harmful to
licensed users if the PUs are present, making low PU quality
of service. This case often happens in CR systems especially
in an environment of lower SNR. Therefore, choosing an
appropriate threshold is very important to ensure a good
performance.

Fig. 1. Complementary ROC curve for ED and LRT with different


noise power uncertainty.

Fig. 2. ROC curves for the ED with and without noise uncertainty.

Fig. 3 describes the sensing performance for the


probability of cooperative detection under (k-out-of-N) rule
and for the cases of AND, OR and Majority rules. We assume
6 SUs are cooperated to detect a PU signal. From Fig. 3, we
notice that through increasing k from 1 to N, the performance
of (k-out-of-N) rule moves down from the performance of OR
rule to AND rule. In particular, we can see that OR rule has
the best probability of detection than AND and Majority rules.
When comparing the performances of all combining rules, the
performance of the Majority rule is much better than the
performance of the AND rule which shows the worst
performance among all fusion rules. It is clearly seen that the
performance of majority fusion rule lies between OR fusion
rule and AND fusion rule.

Fig. 3. Cooperative spectrum sensing of (k-out-of-N) HDC rules for SUs = 6

Some intuitive results are stated in Fig. 5 and this justifies


the B function versus the detection threshold for various
fusion rules from ( k = 1) to ( k = 6) in a CRN of 6 users.
Thus, from Fig. 5, the best value of the detection threshold is
achieved with the OR rule ( k = 1) while the AND rule and

Fig. 4 shows the ROC curves of different combination rules


over different SNR values for the targeted probability of false
alarm of 0.1. As shown in Fig. 4, the detection probability for
OR rule is much higher than AND rule and k-out-of-N rule,
especially under the lower values of SNR (i.e., -14 dB). For
instance, with SNR= -18 dB, the probability of detection for
OR rule is 88%, while that for AND rule is 3.4%.

978-1-4799-7315-6/15/$31.00 2015 IEEE

majority rule, ( k = 6) and ( k = 3) , respectively, yield the


same value of detection threshold. Also, the optimal
combining rule over all the examined range of detection

26

2015 1st International Conference on Telematics and Future Generation Networks (TAFGEN)

threshold is majority rule (k

= 3) where the optimal selection

optimal combining rule over all the examined ranges of the


detection thresholds whereas for a fixed very small and very
large detection thresholds, the AND rule and OR rule tends to
be optimal, respectively.

k is ( N 2) . Also, for a fixed very small detection


threshold, the AND rule (k = 6) is the optimal rule. On the
other hand, the OR rule, i.e., (k = 1) tends to be optimal for a

of

ACKNOWLEDGMENT

fixed and very large detection threshold.

The author would like to thank Malaysias Ministry of


Education (MoE) for their financial support, under grant
scheme number ERGS/1/2013/ICT03/UKM/02/1. Also, I
extend my sincere thanks and gratitude to the Ministry of
Higher Education & Scientific Research-Foundation of
Technical Education in Iraq for providing a doctoral
scholarship to complete study in Universiti Kebangsaan
Malaysia (UKM).
REFERENCES
[1]

[2]

[3]

Fig. 4. ROC curves for the ED versus SNR for different combining rules

[4]

[5]

[6]

[7]

[8]

[9]
Fig. 5. Bayes risk function in terms of detection threshold for different
combining rules when k = 1, 2, 3,, 6, no. of SUs (N=6), and SNR= 10 dB.

[10]

VI. CONCLUSION

[11]

In this paper, we have investigated the sensing performance


of ED and compare it with that of LRT. From the results, it is
demonstrated that the LRT outperformed the ED at low false
alarm probability. Also, CSS is studied where several SUs
cooperate together to detect the spectrum holes through
operation of ED with HDC rules. In addition, it has been
found that the overall cooperative sensing performance of the
OR rule under lower values of SNR and for all values of
probability of detection is superior to the other combination
rules. Furthermore, the optimality of CSS is investigated
through obtaining the optimal decision rule that minimizes the
Bayes risk function. Therefore, the majority fusion rule is the

978-1-4799-7315-6/15/$31.00 2015 IEEE

[12]

[13]

[14]

27

E. Waleed, U. H Najam., and S. K Hyung., SNR-based adaptive


spectrum sensing for cognitive radio networks, Intern. Journ. of Innov.
Comput., Inform. and Cont. (ICIC), vol. 8, no. 9, pp. 6095-6105, 2012.
D. Cabric, A. Tkachenko, and R. W. Brodersen, Spectrum sensing
measurements of pilot, energy and collaborative detection, IEEE Milit.
Comm. Conf. ( MILCOM), pp. 1-7, 2006.
L. Ying-Chang, Z. Yonghong, E.C.Y Peh and T. H. Anh, Sensingthroughput tradeoff for cognitive radio networks, IEEE Trans. on wire.
Comm., vol. 7, no. 4, pp.1326-1337, 2008.
G. Amir and S. S. Elvino, Collaborative spectrum sensing for
opportunistic access in fading environments, IEEE New Front. in Dyna.
Spect. Acce. Netw. (DySPAN), pp. 131-136, 2005.
L. Xiao, K. Liu, and L. Ma, A weighted cooperative spectrum sensing
in cognitive radio networks, IEEE Intern. Conf. on Inform., Networ.
and Autom. (ICINA), pp. V2-45-V2-48, 2010.
E. S. Hassan, Spectrum Sensing and Power Efficiency Trade-off in
Cognitive Radio Networks over Fading Channels, Wseas Transactions
on Systems, vol. 12, no. 1, pp. 32-41, 2013.
Z. Quan, S. Cui, and A. H. Sayed, Optimal Linear Cooperation for
Spectrum Sensing in Cognitive Radio Networks, IEEE Journal of
Selected Topics in Signal Processing, vol. 2, no. 1, 28-40, 2008.
Y.-C Liang, Y. Zeng, E. Peh, and A. T. Hoang, Sensing-throughput
tradeoff for cognitive radio networks, IEEE Transactions on Wireless
Communications, vol. 7, no. 4, pp. 13261337, April 2008.
E. Axell, G. Leus, E. G. Larsson, and H. V. Poor, Spectrum sensing for
cognitive radio, IEEE Signal processing magazine, May 2012.
Z. Yonghong and L. Ying-Chang, Eigenvalue-based spectrum sensing
algorithms for cognitive radio, IEEE Transaction on Communication,
vol. 57, no. 6, pp. 1784 1793, 2009.
S. Shellhammer and R. Tandra, performance of the power detector with
noise uncertainty, IEEE 802.22-06/0134r0, 2006.
W. Zhang, R. K. Mallik and K. B. Letaief, Cooperative Spectrum
Sensing Optimization in Cognitive Radio Networks, IEEE
International Conference on Communications (ICC), pp. 3411-3415,
2008.
W. K. Saad, M. Ismail, R. Nordin and A. A. El-Saleh, On the
Performance of Cooperative Spectrum Sensing of Cognitive Radio
Networks in AWGN and Rayleigh Fading Environments, KSII
Transactions on Internet and Information Systems, vol. 7, no. 8, pp.
1754-1769, 2013..
K. V. Pramod, Distributed Detection and Data Fusion, New York:
Springer-Verlag, 1997.

You might also like