1

st
Intl Conf. on Recent Advances in Information Technology | RAIT-2012 |


978-1-4577-0697-4/12/$26.00 2012 IEEE

Performance of Cooperative Spectrum Sensing in
Fading Channels

Srinivas Nallagonda, Sanjay Dhar Roy, Sumit Kumdu
Dept. of Electronics & Communication Engineering,
National Institute of Technology Durgapur
Durgapur -713209, India
[email protected], [email protected]; [email protected]


Abstract: Detection is compromised when a cognitive radio (CR)
user experiences deep shadowing or fading effects. In order to
detect the primary user (PU) more accurately, we allow the CR
users to cooperate by sharing their information. In this paper we
investigate performance of cooperative spectrum sensing scheme
using energy detection (ED-CSS) to improve the sensing
performance in channels such as log-normal shadowing and
Nakagami fading channels. Hard decision combining rule (OR-
rule, AND-rule and MAJORITY-rule) is performed at fusion
center (FC) to make the final decision about primary user
present or not. Comparison among data fusion rules has been
investigated for a wide range of average SNR values. The
performance has been assessed in terms of miss detection (P
m
)
and false detection probabilities (P
f
). A simulation model has
been developed to evaluate performance of ED-CSS in different
fading environments. A comparative performance of ED-CSS has
been studied for various data fusion rules in Nakagami fading as
well as shadow faded channels.
Keywords: Cognitive radio, energy detection, fading channels,
cooperative spectrum sensing, fusion rules, detection probability
I. INTRODUCTION
Cognitive radio (CR) technique has been proposed to solve
the conflicts between spectrum scarcity and spectrum under
utilization [1]. It allows the CR users to share the spectrum
with primary users (PUs) by opportunistic accessing. The CR
user can use the spectrum only when it does not create any
disturbance or interference to primary users. Therefore,
spectrum sensing is an important issue of cognitive radio
technology since it needs to detect the presence of primary
users accurately and quickly. In many wireless applications, it
is of great interest to check the presence and availability of an
active communication link when the primary signal is
unknown. In such scenarios, one appropriate choice consists
of using an energy detector which measures the energy in the
received waveform over an observation time window [2].
Spectrum sensing is a hard task because of shadowing, fading
and time- varying nature of wireless channels [3]. Due to the
severe multipath fading, a cognitive radio may fail to notice
the presence of the PU. Cooperative spectrum sensing
improves the detection performance. All CR users sense the
PU individually and send their sensing information in the form
of 1-bit binary decisions (1 or 0) to Fusion center (FC).
The hard decision combining rule (OR, AND, and
MAJORITY rule) is performed at FC using a counting
rule (voting rule) to make the final decision regarding
whether the primary user present or not [4]-[7]. However,
the existing works only examined the collaborative
spectrum sensing, using energy detection in Log-normal
shadowing and the Rayleigh fading channel [5]-[6].
Comparison among hard decision fusion rules for the case
of cooperative spectrum sensing has been investigated in
a Suzuki fading channel [4]. We note that the analytical
expression for probability of detection in different fading
channels (Rayleigh and Nakagami) was given in [9] - [8].
In this paper, we have presented a new simulation
model to study the performance of cooperative spectrum
sensing, using energy detection (ED-CSS), over Log-
normal shadowing and Nakagami fading channels. Results
obtained via simulation test bed for the case of Log-
Normal shadowing channel match exactly with the results
obtained in the paper under same scenario [5]. In
particular, the performance of hard decision fusion rules in
Nakagami fading channel has been made.


Fig.1. Block diagram of an energy detector

The energy detection method is the common method
for detection of unknown signals in noise [2]. The block-
diagram of an energy detector as shown in Fig.1, the input
band pass filter (BPF) selects the center frequency
c
f , and
bandwidth of interest, W. The output of BPF filter is
passed to a squaring device to measure the received
energy and an integrator which determines the
observation interval, T. Finally, output of the integrator,
Y, is compared with a detection threshold, to decide
whether the signal is present or not. We assume that all
BPF
(.)
2

T
0
(.)

Decide
H0 or H1
X(t)
1
st
Intl Conf. on Recent Advances in Information Technology | RAIT-2012 |



cooperative CR users employ same energy detector and use
the same threshold ( ).
A simulation study is carried out to evaluate the effects of
shadowing and fading on the performance of ED-CSS. The
rest of the paper is organized as follows. In Section II, the
system model under consideration is described and important
notations are listed. Section III briefly describes the
probabilities of detection and of false alarm over additive
white Gaussian noise (AWGN) channel, shadowing and
fading channels. Our simulation model is presented in section
IV. Results and discussions are presented in section V. Finally
we conclude in section VI.
II. SYSTEM MODEL
Before describing the system model, we list some notations
in a Table1 that are going to be used in this paper [2]-[9].

TABLE 1

Description Symbol
Signal waveform


) (t s
Noise waveform which is modeled as a zero-
mean white Gaussian random process
) ( t n

One-sided noise power spectral density
01
N

Signal energy,

s
E

=
T
s
dt t s E
0
2
. ) (

Signal-to-noise ratio (SNR)
01
N
E
s
=

Average SNR

One-sided bandwidth (Hz), i.e., positive
bandwidth of low-pass (LP) signal
W
Time-bandwidth product
TW m =
Carrier frequency
c
f

Probability of detection
d
P
Probability of false alarm
f
P
Probability of missed detection
d m
P P = 1
Hypothesis 0 corresponding to no signal
transmitted
0
H

Hypothesis 1 corresponding to signal transmitted
1
H

A Gaussian variate with mean and variance
2

) , (
2
N


The received signal
) ( t x
can be represented as

+
=
1
0
) ( ) ( *
) (
) (
H t n t s h
H t n
t x
(1)
According to the sampling theorem, the noise process can be
expressed as [10],






, ) 2 ( sin ) (

=
=
i
i
i Wt c n t n

(2)
where
x
x
x c

) sin(
) ( sin =
and ) (
2W
i
i
n n = . One can
easily check that
), , 0 ( ~
01
W N N n
i
for all i.
(3)
The noise energy can be approximated over the time
interval (0, T), as [2]-[12]:
,
2
1
) (
2
1
2
0
2

=
=
m
i
i
T
n
W
dt t n
(4)
If we define
,
01
W N
n
n
i
i
= then the decision statistic
Y can be written as [2] & [9]:



Fig.2. Block diagram of cooperative spectrum sensing

=
=
m
i
i
n Y
2
1
2
(5)
Y can be viewed as the sum of the squares of m 2
standard Gaussian variates with zero mean and unit
variance. Therefore, Y follows a central chi-square (
2

)
distribution with m 2 degrees of freedom. The same
approach is applied when the signal ) (t s is present with
the replacement of each
i
n by
i i
s n + where ) (
2W
i
i
s s = .
The decision statistic Y in this case will have a non-
central
2

distribution with m 2 degrees of freedom and

a non centrality parameter 2 [2]-[9]. We can describe
the decision statistic in short-hand notations as

.
,
), 2 (
,
~
1
0
2
2
2
2
H
H
Y
m
m

(6)


PU
CR1
CR3
CR2
FC
Primary user Fusion center
1
st
Intl Conf. on Recent Advances in Information Technology | RAIT-2012 |



III. DETECTION AND FALSE ALARM PROBABILITIES
A. Non-fading environment (AWGN channel)
In non-fading environment the probabilities of detection
and false alarm are given by the following formulas [2]- [9],
) , 2 ( ) / (
1

m d
Q H Y P P = > =
(7)

) ( / ) 2 / , ( ) / (
0
m m H Y P P
f
= > =

(8)
where
(.,.)
is the incomplete gamma function [12] and
(.,.)
m
Q is the generalized Marcum Q-function [11]. If the
signal power is unknown, we can first set the false alarm
probability
f
P to a specific constant. By equation (8), the
detection threshold can be determined. Then, for the fixed
number of samples TW 2 the detection probability
d
P can
be evaluated by substituting the in (7). As expected,
f
P is
independent of since under
0
H
there is no primary signal
present. When h is varying due to fading, equation (7) gives
the probability of detection as a function of the instantaneous
SNR, . In this case, the average probability of detection (
d
P
) may be derived by averaging (7) over fading statistics [5],

dx x f Q P
x
m d
) ( ) , 2 (

(9)
where ) ( x f

is the probability distribution function (PDF) of

SNR under fading.

B. Log-normal shadowing
The linear channel gain may be modeled by a Log-normal
random variable,
X
e where X is a zero-mean Gaussian
random variable with variance
2
. Log-normal shadowing is
usually characterized in terms of its dB-spread,
dB

which is
related to by
dB
) 10 ln( 1 . 0 = [5].
C. Rayleigh fading channels
If the signal amplitude follows a Rayleigh distribution,
then the SNR follows an exponential PDF given by [9]:
( ) , 0 exp ) (
1
=

f (10)
The average
d
P in this case, dRay P , can be evaluated by
substituting (10 )in (9), here ) (x f

= ) ( f .
1 1
2
0
2 !
1
) ( ) (
2
+

+ =

m
m
k
k
k
dRay e P

\
|

=
+

+
2
0
) 1 ( 2
!
1
) (
2
) 1 ( 2
m
k
k
k
e e

(11)
D. Nakagami fading channels
If the signal amplitude follows a Nakagami
distribution, then PDF of follows a gamma PDF given
by [9]:

( ) ( ) , 0 , exp ) (
1
) (
1
=

M
M
M
M
M
f

(12)
where M is the Nakagami parameter. The average
d
P in
the case of Nakagami channels dNak P can be evaluated
by substituting (12) in (9), here ) (x f

= ) ( f .
( )
(

+ + =

=
+
1
1
2 1 1 ! 2
2 /
1
) ; 1 ; (
m
n
M
n
dNak n M F G P
n

(13)
where (.;.;.)
1 1
F is the confluent hyper geometric
function (.,.;.)) ( [12, section 9.2],
( )
M
M
M
M

1
2 ) (
1

= , (14)
( ) , ) (
2 / 2

+
= e M
M
M
(15)
and
( )
( )( ) [
1 ! 1 2
1
1
2
1

+
|

\
|

+ =
+

M
M
M M
M
M M
M
M
M
M
e G

( ) ( ) ( )

=
+ +

+

M
M
n
n
n
M
M
M
M
L L
2
2
0
2 1

(16)
where (.)
n
L is the Laguerre polynomial of degree n [12,
section 8.970]. We can obtain an alternative expression
for dRay P when setting M=1 in (13) and this expression
is numerically equivalent to the one obtained in (11).
E. Cooperative spectrum sensing in fading channels
Detection performance can be improved by allowing
different cognitive radio users to cooperate by sharing
their information.
Let N denote the number of users sensing the PU. Each
CR user makes its own decision regarding whether the
primary user present or not, and forwards the binary
decision (1 or 0) to fusion center (FC) for data fusion as
shown in Fig.2. The PU is located far away from all CRs.
All the CR users receive the primary signal with same
local mean signal power, i.e. all CRs form a cluster with
distance between any two CRs negligible compared to the
distance from the PU to a CR. For simplicity we have
assumed that the noise, fading statistics and average SNR
are the same for each CR user. We consider that the
channels between CRs and FC are ideal channels
(noiseless). Assuming independent decisions, the fusion
problem where k out of N CR users are needed for
decision can be described by binomial distribution based
1
st
Intl Conf. on Recent Advances in Information Technology | RAIT-2012 |



on Bernoulli trials where each trial represents the decision
process of each CR user. The generalized formula for overall
probability of detection,
d
Q for the k out of N rule is given by
[4]:
( )
l N
d
l
d
N
k l
d
P P
l
N
Q

=

|
|

\
|
=

1
(17)
where P
d
is the probability of detection for each individual CR
user as defined by (7) & (9).
The OR-fusion rule (i.e. 1 out of N rule) can be evaluated by
setting k=1 in equation (17):

( ) ( )
N
d
l
l N
d
l
d
l N
d
l
d
N
l
OR d
P P P
l
N
P P
l
N
Q ) 1 ( 1 1 1 1
0
1
,
=
|
|

\
|
=
|
|

\
|
=
=

=

(18)
The AND-rule (i.e. N out of N rule) can be evaluated by
setting k=N in equation (17):
( )
N
d
l N
d
l
d
N
N l
AND d
P P P
l
N
Q ) ( 1
,
=
|
|

\
|
=

=

(19)
Finally, for the case of MAJORITY-rule (i.e. N/2 out of N) the

MAJ d
Q
,
is evaluated by setting

2 / N k = in (17).
Similarly, the overall probability of false alarm,
f
Q for the
case of OR, AND, and MAJORITY rule can be evaluated by
replacing P
d
with P
f
in equations (17), (18), and (19)
respectively.
IV. SIMULATION MODEL
The simulation is developed in MATLAB using the
following system parameters: Time-bandwidth product, m = 5,
average SNR,

=10dB and 1 . 0 =
f
Q . To obtain the fading
channel power distribution one can rely on the
amplitude/envelope distribution. Let us assume that each
multipath component (MPC) obeys an instantaneous fading
amplitude/envelope of h a = with PDF P
a
(a). The
instantaneous power of the said fading channel is thus given as
2
a g = with PDF ) ( g P
g
; its average } { } {
2
a E g E g = =
is often normalized to unity, i.e. , 1 = g

using a simple PDF
transformation, one can relate the PDF of the channel power
with the one of the envelope which is given by [13]:
) (
2
1
) ( g P
g
g P
a g
=
(20)
And inversely
) ( . 2 ) (
2
a P a a P
g a
=
(21)
A. Probability of detection simulation in non-fading (pure
AWGN) channel
i. Generate a BPSK signal s(t) with 1, -1 up to
2TWsamples and generate AWGN signal n(t) with
zero mean, variance
W N
01
i.e.,
), , 0 (
01
W N N


here
. /
01

s
E N =

ii. Received signal is x (t) = s (t) + n (t).
iii. Now x (t) is the input to BPF and output of
squaring device is x
2
(t) and passes through
integrator. Then the output of integrator Y is

=
=
m
i
i
n Y
2
1
2

iv. Detection threshold ( ) can be obtained for an
each specific value of P
f
(from equation (8)).
v. Compare Y with detection threshold ( ).
vi. If Y is greater than , binary decision 1 which
indicates PU is present otherwise binary decision
0 which indicates PU is absent.
vii. The steps (i) to (vi) have been repeated for N
number of CRs.
viii. Now each CR user has its own 1-bit binary
decision, then the OR-(P
d,CR1
|| P
d,CR2
|| P
d,CR3
. ||
P
d, CRN
), AND-(P
d,CR1
&& P
d,CR2
&& P
d,CR3
. &
& P
d, CRN
), and MAJORITY-(H=P
d,CR1
+ P
d,CR2
+
P
d,CR3
. + P
d, CRN,
if (H>N/2) , then
d
Q =1)-
rules are performed at FC, and then get H
1
for
one time.
ix. The steps (i) to (viii) have been repeated a large
number of times and then the average value of
d
Q and
m
Q

have been estimated.
B. Probability of detection simulation in Log-normal
shadowing
i. To generate log-normal shadowing channel gain
h, the procedure mentioned as in B of section III,
is followed.
ii. Now the received signal is x (t) = h * s (t) + n
(t).
iii. Then the steps from (iii) to (ix) as in subsection
A of IV are followed.
C. Probability of detection simulation in Nakagami
fading channel
i. Envelope/amplitude of channel h follows a
Nakagami distribution. Nakagami distribution
can be generated from Gamma distribution. To
find the Nakagami parameters (M, w), second
moment of Nakagami distribution is set to unity.
Fix the value of M as 1,2and 3 and find other
parameter w.
ii. If the random variable Y ~ gamma (u, v) then we
get h ~ Nakagami (M, w) by setting u=M, v = w/
M in Y ~ gamma (u, v) and h= square root (Y).
iii. Now follow the steps (ii) and (iii) as in
subsection B of IV.
V. RESULTS AND DISCUSSIONS
In this section, simulation results are shown.
Fig.3 and Fig.4 show complementary ROC
(
m
Q vs.
f
Q ) curves for different cooperative CR users
under Log-normal shadowing (
dB
=2dB) and Nakagami
1
st
Intl Conf. on Recent Advances in Information Technology | RAIT-2012 |



fading respectively. Non- fading AWGN curve is shown for
comparison in both cases (AWGN and N=1curves matched
with curves in [5]-[9]). We can observe in these figures, fusing
the decisions of different CR users cancels the effect of
shadowing or fading on the detection performance effectively.
Moreover, with increase in N (curves (vi) & (iv) to (i) in both
figures), cooperative spectrum sensing out performs than
AWGN local sensing (curve (v) in both figures). This is due to
the fact that for larger N, with high probability there will be a
user with a channel better than that of the non- fading AWGN
case.
Fig.5 and 6 shows the probability of detection (
d
Q ) vs.

under Log-normal shadowing and Nakagami fading
scenarios for different number of cooperative CR users. We
have chosen
f
Q
as 0.1 and m=5 for each curve in these
figures. We observe that there is an excellent improvement in
performance of ED-CSS with increase in N and average SNR
in both the figures (curves (i) to (v)). In particular, for a
probability of detection equal to 0.9, local spectrum sensing
requires 12dB while collaborative sensing with N=8
only needs approximately 5dB for individual CR users.
Fig.7 shows the performance of hard decision fusion rules
and their comparison based on complementary receiver
operating characteristics (ROC) curves i.e. probability of
missed detection ) (
m
Q vs. ,
f
Q for 5 collaborative users
under Nakagami fading channel (M=3), m=5 and average
SNR =10dB. We observe that for a particular value of
, 1 . 0 =
f
Q probability of missed detection ) (
m
Q is 0, 0.1
and above 0.8 for OR-rule, MAJORITY and AND-rules
respectively. We can say that OR-rule performs better than
MAJORITY and AND-rules (curves (i), (ii) & (iv)
respectively). The curve (iii) for no-collaboration case (N=1)
is provided for comparison purpose.
Fig.8 shows the performance of hard decision rules and
their comparison based on
d
Q vs. average SNR for 5
collaborative users under Nakagami fading channel (M=3), m
=5 and . 1 . 0 =
f
Q We observe that for a particular value
of average SNR 6dB probability of detection is above 0.8,
0.15 and 0 for the OR-rule, MAJORITY and AND-rules
respectively. We can say that OR-rule performs better than
MAJORITY and AND-rules (curves (iv), (ii) & (i)
respectively). The curve (iii) for no-collaboration case (N=1)
is provided for comparison purpose.
VI. CONCLUSIONS
We have studied the performance of cooperative spectrum
sensing scheme using energy detection (ED-CSS) under Log-
normal shadowing and Nakagami fading channel. We
develop a simulation model for the evaluating the
performance in terms of miss detection and false alarm
probabilities. The performance of ED-CSS also has been
investigated via probability of detection versus different
average SNR values in Log-normal shadowing and
Nakagami fading channels. The performance of ED-CSS
has been investigated for different data fusion rules (OR,
MAJORITY and AND-rules) using our simulation test
bed and compared with each other through
complementary ROC. Finally we have shown that
cooperative spectrum sensing, using energy detection
performs better for OR- data fusion rule compared to
MAJORITY and AND-rules under same average SNR
conditions in Nakagami fading channel.
REFERENCES
[1] S. Haykin, Cognitive radio: brain-empowered wireless
communications, IEEE J. Select. Areas Commun, vol. 23, pp.
201-220, Feb. 2005.
[2] H. Urkowitz, Energy detection of unknown deterministic signals,
Proceedings of IEEE, vol. 55, pp. 523231, April 1967.
[3] S D. Cabric, S. M. Mishra, and R. W. Brodersen, Implementation
issues in spectrum sensing for cognitive radios, in Proc. of
Asilomar Conf. on Signals, Systems, and Computers, Nov. 7-10,
2004, vol. 1, pp. 772776.
[4] Spyros Kyperountas, Neiyer Correal, Qicai Shi and Zhuan Ye,
Performance analysis of cooperative spectrum sensing in Suzuki
fading channels, in Proc. of IEEE International Conference on
Cognitive Radio Oriented Wireless Networks and
Communications (CrownCom07), pp. 428-432, June 2008.
[5] A. Ghasemi and E. S. Sousa, Collaborative spectrum sensing for
opportunistic access in fading environments, in Proc. of 1st IEEE
Symp. New Frontiers in Dynamic Spectrum Access Networks,
Baltimore, USA, Nov. 8-11, 2005, pp. 131-136.
[6] A. Ghasemi and E. S. Sousa, Opportunistic spectrum access in
fading channels through collaborative sensing, IEEE Journal on
selected Areas in Communications, vol. 2, no. 2, pp. 71-82, March
2007.
[7] A. Ghasemi and E. S. Sousa, Impact of user collaboration on the
performance of opportunistic spectrum access, in Proc. IEEE
Vehicular Technology Conference (VTC Fall'06), Montreal,
September 2006.
[8] Jiaqi Duan and Yong Li, Performance analysis of cooperative
spectrum sensing in different fading channels, in Proc. IEEE
Interantional conference on Computer Engineering and
Technology (ICCET10), pp. v3-64-v3-68, June 2010.
[9] F. F. Digham, M.-S. Alouini and M. K. Simon, On the energy
detection of unknown signals over fading channels, in Proc. of
IEEE International Conference on Communications (ICC03), pp.
35753579, May 2003.
[10] C.E. Shannon, Communication in the presence of noise, proc.
IRE, vol. 37, pp. 10-21, January 1949.
[11] A. H. Nuttall, Some integrals involving the QM function, IEEE
Transactions on Information Theory, vol. 21, no. 1, pp. 9596,
January1975.
[12] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
Products, 5th ed. Academic Press, 1994.
[13] Yonghui Li, Mischa Dohler, Cooperative communications:
hardware, channel, PHY, Jhon Wiley & Sons, 2010.

1
st
Intl Conf. on Recent Advances in Information Technology | RAIT-2012 |



10
-4
10
-3
10
-2
10
-1
10
0
10
-4
10
-3
10
-2
10
-1
10
0
Qf
Q
m


(i)
(ii)
(iii)
(iv)
(v)
(vi)
AWGN
N=1(no collaborat ion)
N=2
N=3
N=6
N=8

Fig.3. Qm

vs. Qf under log-normal shadowing (dB=2dB) for
different number of cooperative CR users (

=10dB, m=5), OR
rule.
10
-4
10
-3
10
-2
10
-1
10
0
10
-4
10
-3
10
-2
10
-1
10
0
Qf
Q
m


(i)
(ii)
(iii)
(iv)
(v)
(vi)
AWGN
N=1(no collaborat ion)
N=2
N=3
N=6
N=8

Fig.4. Qm vs. Qf

under Nakagami fading (M=3) for different
number of cooperative CR users (

=10dB, m=5), OR rule.

0 2 4 6 8 10 12 14 16 18 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SNR (dB)
Q
d


(i)
(ii)
(iii)
(iv)
(v)
N=1 (no collaborat ion)
N=2
N=3
N=6
N=8

Fig.5. Qd

vs.

under Log-normal shadowing (dB=2dB) for

different number of cooperative CR users (Qf =0.1 , m=5) , OR
rule.
0 2 4 6 8 10 12 14 16 18 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SNR (dB)
Q
d


(i)
(ii)
(iii)
(iv)
(v)
N=1 (no collaboration)
N=2
N=3
N=6
N=8

Fig.6. Qd vs.

under Nakagami fading channel (M=3) for different
number of cooperative CR users (Qf =0.1 , m=5) , OR rule.

10
-4
10
-3
10
-2
10
-1
10
0
10
-4
10
-3
10
-2
10
-1
10
0
Qf
Q
m


(i)
(ii)
(iii)
(iv)
N=1(no collab)
OR-rule
AND-rule
Majorit y-rule

Fig.7. Performance of hard decision fusion rules via Qm vs. Qf under
Nakagami fading channel (M=3) for N=5 CR users.

0 2 4 6 8 10 12 14 16 18 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SNR (dB)
Q
d


(i)
(ii)
(iii)
(iv)
N=1(no collab)
OR-rule
AND-rule
Majority-rule

Fig.8.

Performance of hard decision fusion rules via Qd vs.

under Nakagami fading channel (M=3) for N=5 CR users.