Average Detection Probability Analysis for

Cooperative - MIMO Spectrum Sensing in

Cognitive Radio Networks
S Hariharan S Venkata Siva Prasad P Muthuchidambaranathan
Department of Electronics and Department of Electronics and Department of Electronics and
Communication Engineering Communication Engineering Communication Engineering
National Institute of Technology National Institute of Technology National Institute of Technology
Tiruchirapalli, India Tiruchirapalli, India Tiruchirapalli, India
Email: [email protected] Email: [email protected] Email: [email protected]

Abstract—In this paper, we consider the Multiple Input Mul- In this paper, we consider a cooperative spectrum sensing
tiple Output (MIMO) - cooperative spectrum sensing based on scheme with an energy detector, multiple antennas at each CR
Energy detection in cognitive radio networks. The primary user and fusion center (FC) , and imperfect reporting channels. The
data is forwarded to the fusion center with N receiving antennas
through the K cognitive radio user each with M transmitting major difference between this paper and [2] is as follows. In
antennas using Time division multiple access (TDMA). The [2], a single antenna based cooperative CR system with white
network is operating over independent and identically distributed Gaussian noise and Rayleigh channel between the PU-CR
(IID) Rayleigh fading channel. We obtain the probability density links and perfect reporting channels is considered, where as, in
function of soft combined signal from K cognitive users. We this paper, we consider a multiple antenna based cooperative
arrived an expression for average detection probability by using
an alternate of generalized Marcum Q-function and compared CR system where the reporting channel experience a Rayleigh
the performance of different soft combining schemes under fading with white gaussian noise. The MIMO concept was
various MIMO systems by simulation. adopted in this cognitive radio system, by utilizing the capa-
bility of MIMO in the fading channel environment, Due to the
I. I NTRODUCTION spatial diversity, this multiple antenna CR system delivers a
better detection performance than Single antenna CR system
Cognitive Radio (CR) is an intelligent wireless communi- in different channel conditions [5], [6].
cation system that is aware of its environment [1] and access The rest of this paper is organized as follows. In Section
unused radio frequency spectrum. The objective of the CR II, the system model and channel under consideration is de-
is to utilize the radio spectrum efficiently when the Primary scribed. Section III evaluates the probabilities of detection and
User (PU) who owns the spectrum access is idle. The user(s) of false alarm over Rayleigh fading channel. While Section IV
who access the idle/unused spectrum opportunistically are outlines the simulation results and Section V concludes the
Secondary User (SU). To access the free spectrum the SU has paper.
to detect the existence of the PU; For that, SU has to sense the
spectrum periodically. The challenge in sensing the spectrum is II. S YSTEM M ODEL
SU should minimize the interference to the licensed spectrum
and should efficiently sense the spectrum holes for required
1
throughput and Quality of Service. Hence the spectrum sensing Listening Channel SU-1 Reporting Channel
(SS) is one of the most important requirements of CR. M

In [2], the authors have proposed an optimal soft combining

method for cooperative spectrum sensing with conventional 1 1
energy detector based cognitive radio network for improving SU-2
M Fusion
reliability in detecting a spectrum hole and compared it with Centre
Primary
conventional diversity combining schemes like Equal gain Receiver N
Primary
combining (EGC) and Maximum ratio combining (MRC). In Transmitter

[3], the authors present an different analytical approach and 1

SU-K
obtain closed-form expression for the probability of detection
M
over Rayleigh fading and also analyzed the probability of
detection when EGC, selection combining (SC), and switch Figure 1. Cooperative - MIMO cognitive radio network
and stay combining (SSC) diversity schemes are used.

978-1-4673-6217-7/13/$31.00 2013
c IEEE 133
 We consider a CR network of single primary user, K
secondary user with single antenna to sense the primary user 
M
signal and M transmitting antenna to report to the FC and a Yj = (rij )2 (5)
FC with N receiving antenna as shown in the Figure 1 and i=1
both the sensing channel and reporting channel experiencing 
independent and identically distributed (IID) Rayleigh fading χ22MN ; H0
Yj ∼ 2
(6)
channels. The instantaneous SNR, denoted by γ, is distributed χ2MN (2γ) ; H1
according to an exponential distribution with Probability den-
The PDF of Yj can be written as,
sity function
 
1 γ ⎧ 1 (MN )
fγ (γ) = exp − (1) (2)
γ̄ γ̄ ⎪
⎪ MN −1 − y2
⎪
⎨ Γ(M N ) y e ; H0
where γ̄ is the average SNR. The SU sense the PU data and fYj (y) = (M N −1)

 (7)
⎪ 21 ( 2γ
⎪
y
) 2 exp − y+2MN γ
forward it to the FC based on TDMA. Therefore the received ⎪
⎩ √2
signals at SU and the FC are given by ×IMN −1 2γy ; H1
√
rk = γpk Spk + npk (2) where Γ(.) is gamma function and Iv (.) is the v th -order
modified Bessel function of the first kind [3], [8].The decisions
√ are based on the weighted sum of the observed energies from
rij = γij rk + nij (3)
the combiner at FC.
where rk received signal at K th SU,1 ≤ k ≤ K , γpk channel

N
gain between PU and SU, Spk ∼ N (0, 1) is the PU signal Yf c = Yj wj (8)
received at K th SU which is assumed to be Gaussian with j=1
zero mean and unit variance, npk ∼ N (0, 1) is the sensing
channel noise assumed to be zero mean and unit variance. where Yf c is the weighted energy signal from the com-
rij is the signal received at FC from K th SU, 1 ≤ i ≤ M , biner and wj is the weight functions for different combining
1 ≤ j ≤ N , γij is channel gain between SU and FC.nij is schemes. We followed the weight functions used in [2] for
white noise which is a Gaussian random variable with zero different combining schemes for our proposed system model
mean and unit variance. and the weight functions are expressed as,

III. A NALYSIS OF D ETECTION PROBABILITY

wegcj = 1 (9)
In this section we define energy detection in fading environ- M
ment followed by the analysis of average detection probability wmrcj = γij (10)
for proposed system model. i=1
M
γij
A. Energy Detection in fading environment wocj = (11)
Energy detection is non-coherent detection method that i=1
(1 + γij )
detects the PU signal based on the sensed energy. In energy Where wegcj is the weight function for EGC, wmrcj is for
detection received signal is limited to the average noise power MRC and wocj is for OC [2].
and normalize the noise variance,then squared and integrated The average detection probability over fading can be ob-
over a time interval of sensing period to produce a measure of tained by
energy of the received signal. The output of the integrator Yf c 
is the detected energy and it is compared with the threshold λ
Pdf c = Pd f (γf c )dγf c (12)
to make the decision at FC. The objective of spectrum sensing γf c
is to decide between the two hypotheses,
 where fγf c (γ) is the PDF output SNR of the combiner at
n(t) ; H0 FC and it is the sum of SNR’s of all receiving antenna γf c =
r(t) = (4) N M 2
hs(t) + n(t) ; H1 i=1 i=1 γij adding all IID noncentral χ variate with 2MN
degrees of freedom and non-centrality parameter 2M N γ each
where H0 is the hypothesis for noise only situation while H1 results in another noncentral χ2 variate [3], [6].The probability
is the hypothesis for presence of PU.The FC is equipped with of detection(Pd) and probability of false alarm (Pf ) of the
energy detector.The Yj is sum of squares of Gaussian random diversity combiner is given by
variable with zero mean and unit variance and it follows
central chi-square (χ2 ) with 2M N degrees of freedom under
Γ(M N, λ2 )
H0 and noncentral χ2 with 2M N degrees of freedom and Pf = Pr(Yf c > λ|H0 ) = (13)
2γ non centrality parameter under H1 [8] and the observed Γ(M N )
 √
energy at the FC is given by, Pd = Pr(Yf c > λ|H1 ) = QMN ( 2M N γf c, λ) (14)

134 2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI)
where λ is the decision threshold, Γ(.) is gamma func-
1
tion, Γ(., .) denotes the incomplete gamma function [7] and OC
QMN (a, b) is the generalized Marcum Q-function [4]. 0.9 MRC
EGC
The PDF of γf c for IID Rayleigh branches[3], [8] is given 0.8
by 0.7

0.6

1 (MN −1)
f (γf c ) =

Pd
0.5
γ (15)
(M N − 1)!γ (MN ) f c
   0.4
γf c
× exp − 0.3
γ
0.2
The average detection probability at FC can be obtained by
 ∞ √
0.1
fc

Pd = QMN ( 2M N γf c, λ)f (γf c )dγf c (16) 0
−30 −25 −20 −15 −10 −5 0 5 10
0 Average SNR (dB)

The integration in the above expression can be solved by

using [[4],eqn. 28] while making the change of variable, x = Figure 2. Pd vs Avg.SNR(γ) for 1X2 system
√ 2 √ √
γf c , p2 = γ1 , a = 2N M , b = λ. Finally Pdf c expressed
as,
1
  OC
λ 1 0.9 MRC
Pdf c = exp −
2 (1 + N M γ) 0.8
EGC

N
M−1  n
1 λ 1 0.7
× (17)
n=1
n! 2 (1 + N M γ) 0.6
Pd

0.5
This is the final expression for average detection probability
of Cooperative MIMO spectrum sensing in cognitive radio net- 0.4

work of single SU. For K SU’s the overall average detection 0.3
probability can be obtained by averaging the average Pd of K
0.2
SU’s.
0.1
IV. S IMULATION R ESULTS
0
−30 −25 −20 −15 −10 −5 0 5 10
The simulation results shown in the figures 2 and 3 are cor- Average SNR (dB)
responding detection probability curves of SIMO, figures 4 and
5 corresponding to MISO and figures 6 and 7 corresponding to Figure 3. Pd vs Avg.SNR(γ) for 1X4 system
MIMO systems with different soft combining schemes under
IID Rayleigh fading channel when the given false alarm prob-
ability is 10−2 .The thresholds for these schemes are obtained 1
from (13) to meet the given false alarm probability.The number OC
0.9
of realizations used for calculating the simulated probability MRC
EGC
of detection is 10000. These curves correspond to a scenario 0.8

in which the number of secondary users is K = 2 experience 0.7

different SNRs on their Sensing and reporting channel. The 0.6
results show the curve plot between average SNR at the output
Pd

0.5
of the combiner against probability of detection. The plots
show that the detection probability improves at lower SNR 0.4

level with increasing number of antennas at both the SU and 0.3

FC and also having better performance compared to the results
0.2
in [2].
0.1
V. C ONCLUSION
0
−30 −25 −20 −15 −10 −5 0 5 10
In this paper we have discussed MIMO-cooperative spec- Average SNR (dB)
trum sensing based on energy detection in CR networks and
we have studied the average detection probability for MIMO- Figure 4. Pd vs Avg.SNR(γ) for 2X1 system
cooperative spectrum sensing in CR network operating over

2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI) 135
 IID Rayleigh fading channels at both sensing and reporting
1
channel. We obtained the CDF and PDF statistics on received
0.9 OC
MRC signal SNR from the diversity combiner at FC with different
0.8 EGC combining schemes. We used an alternative representation of
0.7
generalized marcum Q-Function to obtain a closed form ex-
pression of average detection probability.Finally we analyzed
0.6
the proposed system performance by simulation.
Pd

0.5
R EFERENCES
0.4
[1] S.Haykin, Cognitive Radio: Brain-Empowered Wireless Communication,
0.3 IEEE Journal on selected areas in communs. Vol.23, No.2, pp.201-
220,Feb. 2005.
0.2
[2] Jun Ma, G.Zhao and Y.Li, Soft Combination and Detection for coop-
0.1
erative Spectrum Sensing in Cognitive Radio Networks, IEEE trans on
wireless communs. Vol.7, No.11, pp.4502-4507, Nov. 2008 .
0 [3] F.F.Digham, M.S Alouini and M.K.Simon On the Energy Detection of
−30 −25 −20 −15 −10 −5 0 5 10
Unknown signals Over fading Channels, IEEE Trans on Commun. Vol.
Average SNR (dB)
55, No. 1, pp 21-25, 2007.
[4] A.H.Nuttall, Some Integrals Involving the QM function, IEEE trans on
Figure 5. Pd vs Avg.SNR(γ) for 4X1 system Information Theory. Vol.21, No.1, pp. 95-96,Jan. 1975.
[5] Jian Li and Petre, MIMO Radar Signal Processing, John Wiley and
Sons, 2009.
[6] Yong Soo Cho, Jaekwon Kim, Won Young Yang and Chung G. Kang,
1 MIMO-OFDM wireless communications with MATLAB, John Wiley
MRC and Sons, 2010.
0.9 [7] I.S. Gradshteyn and I.M. Ryzhik, Table of Integrals , Series, and Prod-
EGC
OC ucts,, 6th ed. San Diego, CA:Academic 2000.
0.8
[8] A.Papoulis and S. Unnikrishna Pillai, Probability, Random variables and
0.7 Stochastic Process, 4th Ed., New work: McGraw-Hill, 2002.

0.6
Pd

0.5

0.4

0.3

0.2

0.1

0
−30 −25 −20 −15 −10 −5 0 5 10
Average SNR (dB)

Figure 6. Pd vs Avg.SNR(γ) for 2X2 system

0.9

0.8

0.7

0.6
Pd

0.5

0.4

0.3

0.2
OC
0.1 MRC
EGC
0
−30 −25 −20 −15 −10 −5 0 5 10
Average SNR (dB)

Figure 7. Pd vs Avg.SNR(γ) for 4X4 system

136 2013 International Conference on Advances in Computing, Communications and Informatics (ICACCI)