Performance Evaluation of Cooperative

Spectrum Sensing Scheme with Censoring

of Cognitive Radios in Rayleigh Fading
Channel

Srinivas Nallagonda, Sanjay Dhar Roy &

Sumit Kundu

Wireless Personal Communications

An International Journal

ISSN 0929-6212
Volume 70
Number 4

Wireless Pers Commun (2013)

70:1409-1424
DOI 10.1007/s11277-012-0756-6

1 23
Your article is protected by copyright and
all rights are held exclusively by Springer
Science+Business Media, LLC.. This e-offprint
is for personal use only and shall not be self-
archived in electronic repositories. If you wish
to self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com”.

1 23
 Author's personal copy
Wireless Pers Commun (2013) 70:1409–1424
DOI 10.1007/s11277-012-0756-6

Performance Evaluation of Cooperative Spectrum

Sensing Scheme with Censoring of Cognitive Radios
in Rayleigh Fading Channel

Srinivas Nallagonda · Sanjay Dhar Roy · Sumit Kundu

Published online: 14 July 2012

© Springer Science+Business Media, LLC. 2012

Abstract In this paper, the performance of cooperative spectrum sensing (CSS) with cen-
soring of cognitive radio (CR) users in Rayleigh fading channel is analyzed. More precisely,
CR users which employ energy detectors are censored depending on the quality of radio
channels between them and a fusion center (FC). Each CR makes a hard decision about
primary user (PU) using energy detection and transmits the information to FC using BPSK
signaling if that CR user is selected to transmit. A training based channel estimator is used
at the FC to estimate the complex Gaussian fading coefficients characterizing the channels
between the CR users and the FC. This channel state information on fading coefficients is
used for censoring the CR users. Two fusion rules such as majority logic fusion and maximal
ratio combining (MRC) fusion rules are applied at the FC for estimating the performance in
terms of probability of missed detection (Pm ). We develop a simulation test bed for evaluat-
ing the performance of CSS scheme. Probability of missed detection has been evaluated for
both perfect and imperfect channel estimation for various probabilities of false alarm (Pf ),
reporting and sensing channel SNR values.

Keywords Cooperative spectrum sensing · Censoring · Energy detection ·

Probability of missed detection

S. Nallagonda (B) · S. D. Roy · S. Kundu

Department of Electronics and Communication Engineering, National Institute of Technology,
Durgapur, 713209 West Bengal, India
e-mail: [email protected]
S. D. Roy
e-mail: [email protected]
S. Kundu
e-mail: [email protected]

123
 Author's personal copy
1410 S. Nallagonda et al.

1 Introduction

Cognitive radio (CR) technique has been proposed to solve the conflicts between spectrum
scarcity and spectrum under utilization [1,2]. Spectrum sensing is an important issue of
cognitive radio technology since it needs to detect the presence of primary users (PUs) accu-
rately and quickly when the primary user (PU) signal is unknown. In such scenarios, one
appropriate choice of spectrum sensing schemes is energy detection based spectrum sensing.
An energy detector (ED) measures the energy in the received waveform over an observation
time window [3,4]. Spectrum sensing is a challenging task because of shadowing, fading and
time-varying nature of wireless channels [5]. The performance of a single cognitive radio
(CR) user using energy detector is sometimes limited due to severe fading or shadowing
of the channel between PU and CR. Detection performance can be improved by allowing
different cognitive radio users to cooperate by sharing their information which is the essence
of cooperative spectrum sensing (CSS) [6,7]. Cooperative spectrum sensing improves the
detection performance where all CR users sense the PU individually and send their sensing
information to fusion center (FC). The fusion center finally decides about the presence of PU
based on the information it receives from the CRs. The channels between primary user (PU)
and CRs are called sensing channels (S-channels) while the channels between the CRs and
FC are called reporting channels (R-channels).
Due to severe effects of noise and fading in the reporting channels, decision received from
some cooperative CR users may be received erroneously at fusion center (FC). In such cases
it is wise to stop transmission to FC from those cooperative CR users whose R-channels are
under severe fading so that energy is not wasted in transmitting a decision which is likely
to be erroneous. The cooperative CR users which are not participating in improving the
detection performance may be stopped so that system overhead and complexity is reduced as
well as detection performance can be improved. Therefore censoring of cognitive radios is
necessary for improving the performance of the CSS. The reporting channels are considered
as noisy and Rayleigh faded in [8,9] in context of a sensor network where sensors report their
decision to a FC. Censoring of sensors as proposed in [10,11] and channel aware censoring
of sensors, discussed in [12] can be well applied in context of CSS. More precisely, the goal
of censoring is to decide which CRs should transmit their observations to the FC to achieve
the best tradeoff between energy efficiency and detection probability. Censoring of cognitive
radios can be done on the basis of channel state information (CSI) on estimated R-channel
coefficients for both perfect and imperfect channel estimation, as discussed in [13–15]. The
performance of majority logic fusion has been evaluated in CSS scheme with censoring of
CR users[17]. Although complexities at the fusion centre (FC) for majority logic fusion rule
are less, the performance may not be good as in case of MRC based and likelihood ratio test
(LRT) based fusion rules.

1.1 Contribution

In this paper, we have investigated performance of CSS scheme with censoring of CR users
based on quality of their reporting channels to FC. Simple majority logic fusion and MRC
fusion are investigated under both perfect and imperfect reporting channel estimation case.
Both sensing (S) and reporting (R) channels have been considered as noisy and slow Rayleigh
faded.
The number of CR users with best estimated channel coefficients is selected from a set
of available CR users. The decisions about PU as obtained by using energy detectors at CRs
are sent by selected CRs to the fusion center (FC). The FC employs fusion rule to obtain

123
 Author's personal copy
Performance Evaluation of Cooperative Spectrum Sensing Scheme 1411

a final decision about PU. The performance of the CSS scheme has been assessed in terms
of probability of missed detection (Pm ) for both perfect and imperfect channel estimation.
Simulation has been carried out to assess the probabilities of missed detection under several
conditions of network such as various values of sensing channel SNR, number of CR users,
average reporting channel SNR and false alarm probabilities. Impact of estimation error on
missed detection performance and comparative performance between Majority logic fusion
and MRC fusion are also highlighted.
The rest of the paper is organized as follows. In Sect. 2, the system model is described
and important notations are listed. In Sect. 3, the considered simulation model is described.
In Sect. 4, simulation results are presented and commented. Finally, Sect. 5 concludes the
paper.

2 System Model

Following notations as given in Table 1 are used in this paper. We consider a network of
N cognitive radio (CR) users. A CR user is using an energy detector (ED) with a threshold
(λ) which makes hard binary decision (either 1 or 0) and transmit it using binary phase shift
keying (BPSK) as modulation format to fusion center (FC). All CR users relatively close to
each other form a cluster and uses an identical threshold (λ). The distance between any two
CRs is less than the distance between a primary user and a CR or the distance between a CR
and the fusion center (FC). Each CR user is having one energy detector as shown in Fig. 1.
An energy detector receives a signal x(t) as defined below at input and gives a binary
decision regarding the presence of the PU.
The received signal x(t) at k-th CR user can be represented as:

n(t) H0
xk (t) = (1)
h k ∗ s(t) + n(t) H1

Table 1
Description Symbol

Signal waveform s(t)

Noise waveform which is modeled as a zero-mean white Gaussian random process n(t)
One-sided noise power spectral density N01
Signal energy Es
Average S-channel Signal-to-noise ratio (SNR) γ̄ = NE s
01
One-sided bandwidth (Hz), i.e., positive bandwidth of low-pass (LP) signal W
Time-bandwidth product m = TW
A Gaussian variate with mean μ and variance σ 2 N (μ, σ 2 )

(). 2
T
BPF ∫ (. )
X(t) 0 Decide
H0 or H1

Fig. 1 Block diagram of an energy detector

123
 Author's personal copy
1412 S. Nallagonda et al.

where h k is the k-th CR user Rayleigh faded sensing channel coefficient, The noise n(t) is
modeled as a zero-mean white Gaussian random process, H1 and H0 are the two hypothesis
associated with presence and absence of a primary user (PU).
The noise energy can be approximated over the time interval (0, T ), as [3,16]:
T
1  2
2m
n 2 (t)dt = ni , (2)
2W
0 i=1

where n i ∼ N (0, N01 W ), for all i. (3)

The decision statistic at k-th CR,Yk , can be written as:


2m
ni
2
Yk = (4)
i=1

where n i = √ Nn i W ,
01
Yk can be viewed as the sum of the squares of 2m standard Gaussian variates with zero
mean and unit variance. Therefore, Yk follows a central Chi-square(χ 2 )distribution with
2m degrees of freedom. The same approach is applied when the signal s(t) is present with
the replacement of each n i by n i + si wheresi = s( 2Wi
). The decision statistic Yk in this
case will have a non-central Chi-squareχ (2γ ) distribution with 2m degrees of freedom and
2

a non centrality parameter 2γ [3,16]. We can describe the decision statistic in short-hand
notations as
 2
χ2m , H0 ,
Yk ∼ (5)
2 (2γ ),
χ2m H1 .
The probability density function (PDF) of Yk can then be written as [3–16]
⎧ y
⎨ 1 m−1 e− 2 ,
2m (m) y H0
f Yk (y) =  m−1
+y √ (6)
⎩1 y 2 2γ
e− 2 Im−1 2γ y , H1
2 2γ

where (.)is the gamma function [19, section 8.31] and Iv (.)is the vth-order modified Bes-
sel function of the first kind [19, section 8.43]. In non-fading environment (AWGN case
i.e., h = 1) the probabilities of detection and false alarm for a k-th CR user are given by the
following formulas [6,16]:

Pd,k = P(Yk > λ/H1 ) = Q m ( 2γ , λ) (7)

P f,k = P(Yk > λ/H0 ) = (m, λ/2)/ (m) (8)

where (., .)is the incomplete gamma function [19] and Q m (., .) is the generalized Marcum
Q-function [18]. The detection threshold λ can be set for a chosen false alarm probability P f
following equation (8). All the CR users are assumed to have identical detection threshold
(λ). The energy detector (ED) at k-th CR user compares decision statistic(Yk )with its detec-
tion threshold(λ)and takes hard binary decision about PU which is sent to FC using BPSK
over Rayleigh faded reporting channel (R-channel).
The signal from k-th CR received at FC is:

yk = sk h k + n k (9)

123
 Author's personal copy
Performance Evaluation of Cooperative Spectrum Sensing Scheme 1413

Fig. 2 Block diagram of Rayleigh faded

cooperative spectrum sensing FC
Reporting channels

CRn
CR1 CR2

Rayleigh faded
sensing channels
PU

− − − − − not selected to transmit

selected to transmit

√ √
where sk is BPSK signal ( E b and − E b ), corresponding to H1 and H0 respectively. R-chan-
nel coefficient h k is modeled as a zero-mean complex Gaussian random variable with variance
σn2 , i.e., h k ∼ C N (0, σ 2 ) as in [13] and n k ∼ C N (0, σn2 ), σn2 = E b /(2∗ R −channel S N R).
The complex Gaussian channel noise samples {n k } and Rayleigh fading coefficients {h k } are
mutually independent. We assume that the FC estimates the Rayleigh fading coefficients h k
according to minimum mean square error (MMSE) estimation strategy on the basis of the
observable yk as follows [13] (Fig. 2):
√
Eb
ĥ k = E[h k |yk,t ] = yk (10)
E b + σn2
We model the channel estimation error as the difference between the actual and the estimated
∼
channel coefficient i.e. h k = h k − ĥ k where h k is actual channel coefficient while ĥ k is its
estimate. We assume that the estimation error is a zero mean complex Gaussian random
variable and has the variance following [13–15]:
−1
Eb
σ∼2 = + 1 (11)
h σn2
Censoring is carried out only on the basis of the estimated channel coefficient in this case. The
FC selects the P (P ≤ N ) CRs associated with the best estimated channel coefficients and
informs the selected CRs via one-bit feedback. The best channel is the channel with highest
estimated channel coefficient amplitude |ĥ k |. The selected CRs observe the phenomenon (i.e.
presence or absence of PU) and take their local decisions in favor of one hypothesis (either
H0 or H1 ). The fading coefficients of a channel are assumed to be fixed over decision symbol
transmission time. The signals at the FC received from the k-th CR is

yk,d = m k h k + n k,d k ∈ {1, 2, . . . , N } (12)

123
 Author's personal copy
1414 S. Nallagonda et al.

√ √
where channel noise n k,d ∼ C N (0, σn2 ) and m k ∈ {+ E b , − E b } is the BPSK modulated
binary decisions.

2.1 Majority Logic Fusion Rule

Since the communication channel is noisy and affected by fading, a decision received by the
FC might differ from the one sent by the corresponding CR. The received decision denoted
by u k from the k-th selected CR, i.e,

1 i f the received decision is H1
uk = (13)
0 if the received decision is H0
where k ∈ {1, 2, . . . P}. Finally, FC finally makes a global decision u0 according to a fusion
rule u 0 =  (u 1 , . . . , u P ) following general majority-like expression [8–15]:
⎧
⎪ P
⎪
⎨ H1 i f u k ≥ P2 + 1
u 0 =  (u 1 , . . . , u P ) = k=1 (14)
⎪
⎪ P
⎩ H0 i f u k < P2 + 1.
k=1

In other words, if the number of decisions in favor of H1 is larger than the number of decisions
in favor of H0 , the FC takes a global decision in favor of H1 and vice versa. If the number of
decisions in favor H1 is equal to the number of decisions in favor of H0 FC flips a fair coin
and takes a decision in favor of either H1 or H0

2.2 Maximal Ratio Combining (MRC) Fusion Rule

The MRC fusion rule depends on the channel estimates, the cognitive radios’ performance
indices, and incorporates the effect of channel estimation error. Assuming that the CR users
have identical local performance indices and BPSK is the used modulation format, the MRC
fusion rule can be obtained by simplifying the likelihood ratio test (LRT) fusion. The LRT
fusion can be written as, following [13]:

P
f (yk,d |H1 )
= (15)
f (yk,d |H0 )
k=1

√
−4 Eb
Re(yk,d ĥ ∗k )

P
Pdk + (1 − Pdk )e 2
σw
= −4
√
Eb
(16)
k=1 Re(yk,d ĥ ∗k )
P fk + (1 − P fk )e 2
σw

E b σn2
where σw2 = E b σh̃2 + σn2 = + σn2 . (17)
E b + σn2
In case of perfect channel estimation σw2 is equal to σn2 when the estimation error variance
σ 2 is zero. At very low R-channel SNR both σn2 and σw2 tend to infinite. By taking a logarithm
h̃
of both sides in (16) and using the approximations e−x ≈ 1 − x and log(1 + x) ≈ x for small
values of x, we can simplify the LRT rule as:
√
2 Eb 
P
1 = log() = (Pdk − P f k )Re(yk,d ĥ ∗k ) (18)
σw2
k=1

123
 Author's personal copy
Performance Evaluation of Cooperative Spectrum Sensing Scheme 1415

We assume that the cognitive radios have identical local performance indices, 1 can be
simplified further as follows:

P ∧

∗
 M RC = Re yk,d h k (19)
k=1

where ĥ ∗k is the complex conjugate of the estimated channel coefficient ĥ k . Given ĥ k , one
can observe from (19) that  M RC is a linear combination of Gaussian random variables and
therefore has a Gaussian distribution. The FC can then take a decision in favor of H1 or H0
simply by comparing  M RC with the threshold zero.

3 Simulation Model

The simulation is developed in MATLAB. The simulation run is carried out according to the
following steps to verify the analytical framework introduced in the previous section.

1. Equally likely hypothesisH ∈ {H0 , H1 }is generated using a uniform random variable
generator.
2. The primary user (PU) signal s(t) with 2T W samples has been generated using uniform
random variable generator. Next we generate AWGN channel noise n(t) with zero
mean and variance N01 W using Gaussian random variable generator.
3. Rayleigh faded S-channel coefficients for k-th CR user h k is generated using Gaussian
random variables.
4. The received signal at k-th CR user is xk (t) = h k s(t) + n(t)for true hypothesis H1 and
xk (t) = n(t)for true hypothesisH0
5. The received signal at k-th CR user xk (t) is given as the input to band pass filter (BPF)
and its output is squared (i.e.xk2 (t)) using squaring device and passed through an inte-
2m 2
grator. The output of the integrator is obtained as: Yk = i=1 n i [from equation
(4)].
6. A predefined threshold λ is set at ED depending on chosen P f (from equation 8).
The ED of k-th CR user compares decision statistic at k-th CR Yk with its detection
thresholdλand takes a hard binary decision (1 or 0) about PU. The binary decision will
be sent to FC using BPSK signaling format via Rayleigh faded reporting channel.
7. The Rayleigh fading R-channel coefficients {h k } are generated as N complex Gaussian
random variables.
8. The channel estimation errors {h̃ k } are generated as ‘N ’ complex Gaussian random
variables with zero mean and variance σh̃ 2 .
9. ‘N ’ estimated channel coefficients {ĥ k } are generated using the channel coefficients
{h k } generated in step 7 and the channel estimation error {h̃ k } generated in step 8.
10. The additive white Gaussian noise samples {n k,d } are generated as ‘N ’ complex Gauss-
ian random variables with zero mean and variance σn 2 .
For majority logic fusion, steps 11–15 are carried out after step 10.
11. The P CRs with highest estimated channel coefficients ĥ k = |h k | are selected under
perfect channel estimation case and ĥ k = |h k − h̃ k | are selected under imperfect
channel estimation case.
12. The selected CRs (P-CRs) observe the phenomenon and transmit their decisions using
BPSK signaling to the FC. We use baseband transmission corresponding to BPSK in
the simulation.

123
 Author's personal copy
1416 S. Nallagonda et al.

13. The FC, upon reception of a decision from a selected CR, makes a decision on the
phenomenon status, i.e., H1 or H0 . Due to the channel impairments, the associated
decision might differ from the one sent by the corresponding CR.
14. The total numbers of decisions in favor of H0 and H1 are counted, and the FC makes
a final decision for H0 or H1 according to (14).
15. If the decided hypothesis is not equal to the “true” hypothesis generated at the first
step then a counter to track the detection errors is incremented by one.
16. Steps 1–15 are repeated a large number of times to reliably estimate (maintaining a
confidence interval) the probability of missed detection.
For MRC fusion, steps 17–21 are followed after step 10.
17. Absolute values of the estimated fading coefficients {ĥ k } are sorted to select the P
cognitive radios with highest estimated fading coefficients.
18. The selected CRs (P-CRs) observe the phenomenon and transmit their decisions using
BPSK signaling to the FC.
19. The complex conjugates {ĥ ∗k } of the estimated fading coefficients are calculated for
the ‘P’ selected CRs and the metric  M RC is evaluated according to equation (19).
20. The value of  M RC computed at the previous step is compared with the MRC threshold
(set to 0) to make a final decision in favor of a hypothesis (eitherH0 or H1 ).
21. Steps 1–10 and then 17–20 are repeated a sufficiently large number of times to reliably
estimate the probability of missed detection.

4 Results and Discussions

Table 2 shows the important network parameters used in the simulation study In Fig. 3, the
probability of missed detection is evaluated (Pm ) as a function of the number (P) of censored
cognitive radios (CRs) in presence of majority logic decision at the FC. The performance
with both perfect and imperfect channel estimation is analyzed for censoring cognitive radios
(CRs). In all cases, two values of R-channel SNR, namely −9 and −7 dB, are considered. As
expected, for a given value of the channel SNR, the probability of missed detection (Pm ) is
higher with imperfect channel estimation, as channel-based censoring leads to the selection
of a group of cognitive radios (CRs) which may not be the best ones. In particular, the prob-
ability of missed detection (Pm ) with imperfect channel estimation increases, with respect to
the case with perfect channel estimation, by 403.99 % when the channel SNR is −9 dB and
the number of selected sensors is 15. On the other hand, for a given number of selected cogni-
tive radios, the missed detection probability decreases with increasing values of the channel
SNR. This is obvious, as a higher channel SNR corresponds to a lower AWGN channel noise.
Furthermore, according to (11), an increase of the channel SNR leads to a decrease of the
estimation error variance σh̃ 2 and this, in turns, reduces the average estimation error, as the
estimation error |h̃ k | is Rayleigh distributed and its average value is directly proportional to
σh̃ . A reduced estimation error leads to a further reduction of the missed detection probabil-
ity. In particular, when the number of selected cognitive radios is 15 and the channel SNR
increases from −9 to −7 dB, the probability of missed detection decreases by 27.29 % in the
presence of imperfect channel estimation [curves (i, ii)] and by 32.41 % in the presence of
perfect channel [curves (iii, iv)]. It is seen that a floor appears in the probability of missed
detection when number of slected CR users exceeds 17 in case of perfect channel estimation
[curves(iii and iv)]. In case of perfect channel estimation, by censoring, fusion center selects
CRs with best reporting channel coefficients which means that decions sent by selected CRs
to FC have low probability of getting flipped. As FC is utilizing a majority logic fusion, it

123
 Author's personal copy
Performance Evaluation of Cooperative Spectrum Sensing Scheme 1417

Table 2 Network Parameters

Parameter Values
used in the Simulation
Number of cognitive radios (N) 25
Average S-channel SNR γ̄ 15 dB, 20 dB, 25 dB
Average R-channel SNR −7 dB, −9 dB
Time-bandwidth product,m 5
Probability of false alarm (Pf ) 0.01, 0.0001

0
10
R-channel SNR = -9dB (Perfect Channel Estimation)
Probability of Missed detection (Pm)---->

R-channel SNR = -9dB (Imperfect Channel Estimation)

R-channel SNR = -7dB (Perfect Channel Estimation)
R-channel SNR = -7dB (Imperfect Channel Estimation)

(i)

(ii)

(iii)

-1 (iv)
10

5 10 15 20 25
Number of selected CRs---->
Fig. 3 Probability of missed detection (Pm ) as a function of the number of censored cognitive radios (CRs)
under perfect and imperfect channel estimation for various reporting channel (R-channel) SNRs (Majority
Logic fusion rule is considered at fusion center (FC), m = 5, Pf = 0.01, sensing channel (S-channel) SNR
(γ ) = 20 dB)

acheives a floor in the missed detection performance at a certain number of CR users i.e., no
further improvement in detection performance is obtained by increasing the number of CRs
beyond this.
In Fig. 4, the probability of missed detection (Pm ) is evaluated, as a function of the number
of selected cognitive radios (CRs), using MRC at the FC. Censoring of cognitive radios with
perfect and imperfect channel estimation is considered, setting the channel SNR to −9 and
−7 dB. As observed for the scenario considered in Fig. 3, the probability of missed detection
increases in the presence of imperfect channel estimation. More precisely, in the presence of
imperfect channel estimation, when the channel SNR is −9 dB and P = 15 the probability
of missed detection increases by 695.98 % as compared to the case with perfect channel esti-
mation. As the channel SNR increases, the probability of missed detection decreases with
both perfect and imperfect channel estimation. For example, at P = 15, as the channel SNR
increases from −9 to −7 dB, the missed detection probability decreases by 27.89 % with
perfect channel estimation and by 30.66 % with imperfect channel estimation.

123
 Author's personal copy
1418 S. Nallagonda et al.

0
10
R-channel SNR = -9dB (Perfect Channel Estimation)
R-channel SNR = -9dB (Imperfect Channel Estimation)
Probability of Missed detection (Pm)----> R-channel SNR = -7dB (Perfect Channel Estimation)
R-channel SNR = -7dB (Imperfect Channel Estimation)

(i)

(ii)
-1
10
(iii)

(iv)

5 10 15 20 25
Number of selected CRs---->
Fig. 4 Probability of missed detection (Pm ) as a function of the number of censored cognitive radios (CRs)
under perfect and imperfect channel estimation for various reporting channel (R-channel) SNRs (MRC fusion
rule is considered at fusion center (FC), m = 5, Pf = 0.01, sensing channel (S-channel) SNR = 20 dB

In Fig. 5, a direct comparison between majority logic and MRC fusion strategies is con-
sidered. The probability of missed detection (Pm ) is evaluated as a function of the number
of selected cognitive radios for various values of the channel SNR (−9 and −7 dB) under
perfect channel estimation. One can observe that, when channel SNR increases from −9 to
−7 dB and the number of selected cognitive radios is 15, the probability of missed detection
decreases by 27.89 % with MRC fusion and by 32.41 % with majority logic fusion. In partic-
ular, the performance with the MRC fusion rule is better than that with majority logic fusion.
For instance, when the channel SNR is −9 dB and the number of selected CRs is 15, the
probability of missed detection with MRC fusion is 41.18 % lower than that with majority
logic fusion.
In Fig. 6, the probability of missed detection is shown as a function of the number of
selected CRs for various values of the R-channel SNR. We compare the performance of
majority logic fusion strategy with the performance of MRC fusion strategy. As in the case
of perfect channel estimation considered in Fig. 5, for a fixed value of selected cognitive
radios, the probability of missed detection is a decreasing function of the channel SNR.
More precisely, when the number of selected cognitive radios is 15, the probability of missed
detection decreases by 30.66 % for MRC fusion when the R-channel SNR increases from
−9 to −7 dB. As in the perfect channel estimation case, MRC fusion guarantees a better
performance than majority logic fusion. In particular, for a channel SNR equal to −9 dB, the
probability of missed detection with MRC fusion is 33.31 % lower than the probability with
majority logic fusion [curves (i, ii)].
In Fig. 7, the probability of missed detection is shown as a function of the number of
selected CRs for various values of the probability of false alarm (Pf ) corresponds to set-
ting of different threshold levels (λ) in the energy detector. We observe that as Pf increases
from 0.0001 to 0.01, the probability of missed detection (Pm ) decreases [curves (i, ii) or

123
 Author's personal copy
Performance Evaluation of Cooperative Spectrum Sensing Scheme 1419

0
10
R-channel SNR = -9dB (Majority Logic)
R-channel SNR = -9dB (MRC)
Probability of Missed detection (Pm)----> R-channel SNR = -7dB (Majority Logic)
R-channel SNR = -7dB (MRC)

(i)
-1
10
(ii)

(iii)

(iv)
5 10 15 20 25
Number of selected CRs---->
Fig. 5 Performance comparison between Majority Logic and MRC fusion rules is evaluated for various
reporting channel (R-channel) SNRs under perfect channel estimation, m = 5, Pf = 0.01, sensing channel
(S-channel) SNR (γ ) = 20 dB

100
R-channel SNR = -9dB (Majority Logic)
Probability of Missed detection (Pm)---->

R-channel SNR = -9dB (MRC)

R-channel SNR = -7dB (Majority Logic)
R-channel SNR = -7dB (MRC)

(i)

(ii)

10-1 (iii)

(iv)

5 10 15 20 25
Number of selected CRs---->
Fig. 6 Performance comparison between Majority Logic and MRC fusion rules is evaluated for various
reporting channel (R-channel) SNRs under imperfect channel estimation, m = 5, Pf = 0.01, sensing channel
(S-channel) SNR (γ ) = 20 dB

curves (iii, iv)]. For example, when the probability of false alarm increases from 0.0001 to
0.01 and the number of selected cognitive radios is 15, the probability of missed detection
decreases by 22.33 % with MRC fusion and by 13 % with majority logic fusion. In particular,

123
 Author's personal copy
1420 S. Nallagonda et al.

0
10
Pf=0.01(Majority logic)
Pf=0.0001(Majority Logic)
Probability of Missed detection (Pm)----> Pf=0.01(MRC)
Pf=0.0001(MRC)

(i)

(ii)
-1
10 (iii)

(iv)

5 10 15 20 25
Number of selected CRs---->
Fig. 7 Performance comparison between Majority Logic and MRC fusion rules is evaluated for various
fixed probability of false alarm under perfect channel estimation (m = 5, reporting channel (R-channel)
SNR = −9 dB, sensing channel (S-channel) SNR (γ ) = 20 dB

the performance with the MRC fusion rule is better than that with majority logic fusion. For
instance, when the probability of false alarm is 0.01 and the number of selected sensors is
15, the probability of missed detection with MRC fusion is 38.20 % lower than that with
majority logic fusion [curves (ii, iv)].
In Fig. 8, a direct comparison between majority logic and MRC fusion strategies, in terms
of probability of missed detection (Pm ) as a function of the number of selected cognitive
radios (CRs), is carried out considering various values of the probability of false alarm and
with imperfect channel estimation. As in the case of perfect channel estimation considered
in Fig. 7, for a fixed value of selected cognitive radios, the probability of missed detection
is a decreasing function of the probability of false alarm. For example, when the number of
selected cognitive radios is 15, the probability of missed detection decreases by 13.4 % for
MRC fusion when the probability of false alarm increases from 0.0001 to 0.01 As in the
perfect channel estimation case, MRC fusion guarantees a better performance than majority
logic fusion. In particular, for a probability of false alarm equal to 0.01 the probability of
missed detection with MRC fusion is 22.38 % lower than the probability with majority logic
fusion [curves (ii. iv)].
Figure 9 depicts the performance comparison between majority logic and MRC fusion
strategies in terms of probability of missed detection (Pm ) under perfect channel estimation
case for various values of the S-channel SNR. The probability of missed detection decreases
with increase in S-channel SNR when number of selected CR users is fixed. More precisely,
when the S-channel SNR increases from 15 to 25 dB and the number of selected CR users
is 9, the probability of missed detection decreases by 39.04 % with majority logic fusion,
by 55.35 % with MRC fusion. For S-channel SNR equal to 20 dB, the probability of missed
detection (Pm ) with MRC fusion is 33 % lower than the probability of missed detection with
majority logic fusion [curves (iii, iv)].

123
 Author's personal copy
Performance Evaluation of Cooperative Spectrum Sensing Scheme 1421

0
10
Pf=0.01(Majority logic)
Pf=0.0001(Majority Logic)
Probability of Missed detection (Pm)----> Pf=0.01(MRC)
Pf=0.0001(MRC)

(i)

(ii)
(iii)
(iv)

5 10 15 20 25
Number of selected CRs---->
Fig. 8 Performance comparison between Majority Logic and MRC fusion rules is evaluated for various
fixed probability of false alarm under imperfect channel estimation (m = 5, reporting channel (R-channel)
SNR = −9 dB, sensing channel (S-channel) SNR (γ ) = 20d B

0
10
S-channel SNR = 15dB (Majority Logic)
Probability of Missed detection (Pm)---->

S-channel SNR = 20dB (Majority Logic)

S-channel SNR = 25dB (Majority Logic)
S-channel SNR = 15dB (MRC)
S-channel SNR = 20dB (MRC)
S-channel SNR = 25dB (MRC)

(i)

(ii)

(iii)
-1
10 (iv)

(v)

(vi)

5 10 15 20 25
Number of selected CRs---->
Fig. 9 Probability of missed detection (Pm ) as a function of the number of censored cognitive radios (CRs)
under perfect channel estimation for various sensing channel (S-channel) SNRs. Majority Logic and MRC
fusion rules are considered at fusion center (FC) and comparison is evaluated, m = 5, Pf = 0.01, reporting
channel (R-channel) SNR) = −9 dB)

123
 Author's personal copy
1422 S. Nallagonda et al.

0
10
S-channel SNR = 15dB (Majority Logic)

Probability of Missed detection (Pm)---->

S-channel SNR = 20dB (Majority Logic)
S-channel SNR = 25dB (Majority Logic)
S-channel SNR = 15dB (MRC)
S-channel SNR = 20dB (MRC)
S-channel SNR = 25dB (MRC)

(i)

(ii)
(iii)

(iv)

(v)

(vi)
-1
10
5 10 15 20 25
Number of selected CRs---->
Fig. 10 Probability of missed detection (Pm ) as a function of the number of censored cognitive radios (CRs)
under imperfect channel estimation for various sensing channel (S-channel) SNR. Majority Logic and MRC
fusion rules are considered at fusion center (FC) and comparison is evaluated, m = 5, Pf = 0.01, reporting
channel (R-channel) SNR) = −9 dB)

In Fig. 10, a direct comparison between majority logic and MRC fusion strategies, in terms
of probability of missed detection (Pm ) as a function of the number of selected cognitive radios
(CRs), is carried out, considering various values of the sensing channel (S-channel) SNR and
with imperfect channel estimation. As in the case of perfect channel estimation considered
in Fig. 9, for a fixed value of selected cognitive radios, the probability of missed detection is
a decreasing function of the sensing channel (S-channel) SNR. When the number of selected
cognitive radios is 15 and the sensing channel (S-channel) SNR increases from 15 to 25 dB,
the probability of missed detection decreases by 41.73 % with MRC fusion. As in the perfect
channel estimation case, MRC fusion guarantees a better performance than majority logic
fusion. In particular, for a sensing channel (S-channel) SNR is equal to 20 dB, the probability
of missed detection with MRC fusion is 22.38 % lower than the probability with majority
logic fusion [curves(iv, vi)].

5 Conclusions

We have studied the performance of CSS scheme with cognitive radio users censored on the
basis of the quality of reporting channel connecting CR users to FC in presence of Rayleigh
fading. Both the cases of perfect channel estimation and imperfect channel estimation have
been evaluated and compared under majority logic and MRC fusion rules. The probability
of missed detection reduces with increase in number of CR users. We observe that no further
improvement in missed detection performance is obtained by increasing the number of CRs
beyond a certain limit in case of perfect channel estimation based censoring. Channel estima-
tion error has significant impact on detection performance as probability of missed detection

123
 Author's personal copy
Performance Evaluation of Cooperative Spectrum Sensing Scheme 1423

increases with increase in estimation error. Probability of missed detection decreases with
increase in reporting channel SNRs, sensing channel SNRs and false alarm probabilities for
both perfect and imperfect channel estimation. Maximal Ratio combining (MRC) fusion rule
outperforms majority logic fusion under all cases of R-channel and S-channel SNR with
perfect or imperfect channel estimation. The above study is useful in designing a cooperative
sensing scheme.

References

1. Haykin, S. (2005). Cognitive radio: Brain-empowered wireless communications. IEEE Journal of

Selected Areas Communications, 23, 201–220.
2. Wang, B., & RayLiu, K. J. (2011). Advances in cognitive radio networks: A survey. IEEE Journal
of Selected Topics in Signal Processing, 5(1), 5–23.
3. Urkowitz, H. (1967). Energy detection of unknown deterministic signals. Proceedings of IEEE, 55,
523–531.
4. Akyildiz, I. F., Lo, B. F., & Balakrishnan, R. (2011). Cooperative spectrum sensing in cognitive radio
networks: A survey. Physical Communication (Elsevier) Journal, 4(1), 40–62.
5. Cabric, S. D., Mishra, S. M., & Brodersen, R. W. (2004). Implementation issues in spectrum sensing
for cognitive radios. In Proceedings of asilomar conference on signals, systems, and computers
(Vol. 1, pp. 772–776) November 7–10.
6. Duan, J., & Li, Y. (2010). Performance analysis of cooperative spectrum sensing in different fading
channels. In Proceedings on IEEE interantional conference on computer engineering and technology
(ICCET’10) (pp. v3-64–v3-68).
7. El-Saleh Ayman, A., Ismail, M., MohdAli Mohd, A., Kamarudin, M. R., & Rahman, T. A. (2011). Ana-
lytical simulation and performance optimization for spectrum sensing in cognitive radio networks.
International Journal on Computer Science and Engineering (IJCSE), 3(2), 554–568.
8. Ferrari G., & Pagliari R. (2006). Decentralized detection in sensor networks with noisy communication
links. In F. Devoli, S. Pelezzo, & S. Zappetore (Eds.), Distibuted cooperative laboratories: Networking,
instrumentation, and measurements (pp. 233–249). New York: Springer.
9. Chen, B., Jiang, R., Kasetkasem, T., & Varshney, P. (2004). Channel aware decision fusion in wireless
sensor networks. IEEE Transactions on Signal Processing, 52, 3454–3458.
10. Rago, C., Willett, P., & Bar-Shalom, Y. (1996). Censoring sensors: A low-communication-rate scheme
for distributed detection. IEEE Transcations on Aerospace and Electronic System, 17, 554–567.
11. Appadwedula, S., Veeravalli, V. V., & Jones, D. L. (2005). Energy-efficient detection in sensor
networks. IEEE Journal on Selected Areas in Communications, 23(4), 693–702.
12. Ahmadi, H. R., & Vosoughi, A. (2009). Channel aware sensor selection in distributed detection
systems. In IEEE 10th workshop on volume signal processing advances in wireless communications
2009, SPAWC.2009 21–24 (pp. 71–75).
13. Ahmadi, H. R., & Vosoughi, A. (2008). Impact of channel estimation error on decentralized detec-
tion in bandwidth constrained wireless sensor networks. In Proceedings on IEEE MILCOM 2008
(pp. 1–7).
14. Kundu, C., Kundu, S., Ferrari, G., & Raheli, R. (2011). Distributed detection using MRC with censored
sensors and Rayleigh faded communications. In 2011 IEEE international symposium on information
theory (ISIT2011) (pp. 2163–2167). Russia: Saint-Petersburg.
15. Kundu, C., Kundu, S., Ferrari, G., & Raheli, R. (2011). Distributed detection with censoring of sensors
in Rayleigh faded channel. Third international conference on communication systems and networks
(COMSNETS), Bangalore, India.
16. Digham, F. F., Alouini, M. S., & Simon, M. K. (2003). On the energy detection of unknown signals
over fading channels. In Proceedings of IEEE international conference on communications (ICC’03)
(pp. 3575–3579).
17. Nallagonda, S., Roy, S. D., & Kundu, S. (2011). Performance of cooperative spectrum sensing with
censoring of cognitive radios in Rayleigh Fading Channel. In Proceedings of IEEE INDICON 2011.
18. Nuttall, A. H. (1975). Some integrals involving the QM function. IEEE Transactions on Information
Theory, 21(1), 95–96.
19. Gradshteyn, I. S., & Ryzhik, I. M. (1994). Table of integrals, series, and products (5th ed.).
London: Academic Press.

123
 Author's personal copy
1424 S. Nallagonda et al.

Author Biographies

Srinivas Nallagonda received his B.E. degree in Electronics and Com-

munication Engineering in 2006 from Osmani University, Hyderabad,
India and M.Tech. degree in Telecommunication Engineering from NIT
Durgapur, India in 2009. He joined as Ph.D. Scholar in NIT Durga-
pur in 2010. His research interests include Cognitive Radio Networks.
As of today, he has published eight (8) research papers in various
conferences.

Sanjay Dhar Roy received his B.E. (Hons.) degree in Electronics

and Telecommunication Engineering in 1997 from Jadavpur Univer-
sity, Kolkata, India and M.Tech. degree in Telecommunication Engi-
neering in 2008 from NIT Durgapur. He received his Ph.D. degree from
NIT Durgapur in 2011. He worked for Koshika Telecom Ltd. from
1997 to 2000. After that he joined the Department of Electronics and
Communication Engineering, National Institute of Technology Durga-
pur as a Lecturer in 2000 and is currently an Assistant Professor there.
His research interests include Radio Resource Management, Handoff,
and Cognitive Radio Networks. As of today, he has published fifty
(50) research papers in various journals and conferences. Dr. Dhar Roy
is a member of IEEE (Communication Society) and is a reviewer of
IET Communications, Electronics Letters and Journal of PIER, IJCS,
Wiley, International Journal of Electronics, Taylor & Francis.

Sumit Kundu received his B.E. (Hons.) degree in Electronics and

Communication Engineering in 1991 from NIT, Durgapur, India and
M.Tech. degree in Telecommunication Systems Engineering and Ph.D.
in Wireless Communication Engineering from IIT Kharagpur, India,
respectively. He has been a faculty in the department of ECE, National
Institute of Technology, Durgapur since 1995 and is currently an Asso-
ciate Professor there. His research interests include radio resource man-
agement in wireless networks, Wireless Ad Hoc and sensor networks,
and Cognitive Radio Networks. As of today, he has published hun-
dred (100) research papers in various journals and conferences. He is a
member of IEEE (Communication Society) and is a reviewer of several
IEEE journals.

123