Sie sind auf Seite 1von 4

Optimization of Parameters for Spectrum Sensing in

Cognitive Radios
Yue Wang, Chunyan Feng, Caili Guo, Fangfang Liu
School of Information and Communication Engineering
Beijing University of Posts and Telecommunications, Beijing, China
Email: wangyuebupt@gmail.com
Abstract In cognitive radios, the first and crucial task is to
detect the spectrum holes or the presence of primary user by
employing spectrum sensing. However, there exist two kinds of
detection errors (i.e., miss-detection error and false-alarm error),
which degrade the sensing performance severely. Aiming at
minimizing the probability of total detection errors, based on the
law of total probability we propose a novel target function (which
also denotes the so-called comprehensive sensing performance
metric) for optimization of sensing parameters. According to
such established target function, we then apply the method of
extremum seeking to develop the optimization of time-bandwidth
product and sensing threshold jointly for local spectrum sensing
with energy detection. Moreover, we further derive the closedform expression of the optimal fusion rule for cooperative
spectrum sensing by performing the discrete difference operation
on the target function. Finally, the optimal area of joint
parameters selection for local spectrum sensing and the effect of
different detecting channel conditions on the optimization for
cooperative spectrum sensing are evaluated numerically.
Keywords cognitive radio; spectrum sensing; target function;
optimization; parameter

I.

INTRODUCTION

In the scenario of cognitive radio (CR), to discover the


potential spectrum holes defined as the licensed spectrum but
unused by primary user (PU) and to be aware of the PU
reappearance, spectrum sensing is the first and crucial task in
CR technology. In particular, local spectrum sensing
techniques employed by a secondary user (SU) include energy
detection, matched filter detection, cyclostationary feature
detection, etc [1]. Furthermore, in order to efficiently overcome
the fading of the detecting channel between PU and SU as well
as the problem of hidden terminal which happens when a SU is
shadowed, cooperative spectrum sensing participated by a
group of spatially distributed SUs is usually conducted based
upon their local spectrum sensing results [2].
However, during both local and cooperative spectrum
sensing processes, there inevitably exist two kinds of detection
errors, i.e., miss-detection and false-alarm errors, which
respectively reflect the interference level to PU and the waste
level of available spectrum holes. Thus, for achieving better
sensing performance, such two kinds of detection errors should
be avoided as much as possible in terms of lower probabilities
of miss-detection and false-alarm. In [3], given the assumption
of known received signal to noise ratio (SNR) of PU signal and
variance of such received SNR at every SU, the optimal

number of SUs participating in cooperative spectrum sensing is


evaluated for unilaterally pursuing either a targeted probability
of miss-detection or that of false-alarm. In [4], the half voting
rule is declared as the optimal fusion rule for cooperative
spectrum sensing. However, the optimization in [4] is
performed based on an oversimple target function which is lack
of taking the spectrum occupancy statistics of PU into account.
Moreover, such declaration indeed disregards the impact of
different detecting channel conditions on optimization results.
In other words, the half voting rule may not always be optimal,
but dependent on specific detecting channel conditions.
In this paper, to thoroughly minimize the probability of
total miss-detection and false-alarm errors, we investigate the
optimization of parameters for both local spectrum sensing and
cooperative spectrum sensing. Specifically, according to the
law of total probability in probability theory, we firstly propose
a novel target function for optimization of a variety of sensing
parameters. In local spectrum sensing with energy detection,
based on our established target function, the joint optimization
of time-bandwidth product and sensing threshold is developed
through the method of extremum seeking. After the analysis of
local spectrum sensing, we further derive the closed-form
expression of the optimal fusion rule for cooperative spectrum
sensing with a given number of total collaborative SUs and
certain spectrum occupancy statistics of PU. Finally, we also
evaluate the optimal area of joint parameters selection for local
spectrum sensing and the effect of detecting channel conditions
on the optimization for cooperative spectrum sensing.
The rest of this paper is organized as follows. In Section II,
we describe the system model and establish the target function.
The time-bandwidth product and sensing threshold for local
spectrum sensing are analyzed and optimized in Section III.
The optimization of fusion rule for cooperative spectrum
sensing is developed in Section IV. In Section V, the
simulation results of optimization and performance evaluations
are provided. Finally, we draw the conclusions in Section VI.
II.

SYSTEM MODEL AND PROBLEM FORMULATION

In a centralized CR spectrum sensing manner, there are


mainly three elements: PU, SU, and fusion center. PU has the
exclusive priority to the licensed spectrum bands and should
not be harmfully interfered by the operations of any other SUs.
SU senses spectrum bands and performs local observations of
PU signal according to a certain local spectrum sensing
algorithm. Since the energy detection is widely adopted with

This research was supported in part by China National 973 Project under
Grant No.2009CB320400 and National Science Foundation of China under
Grant No.60772110.

978-1-4244-3693-4/09/$25.00 2009 IEEE

low computational complexity and no requirement of prior


knowledge, we conduct our work with energy detection as the
local spectrum sensing algorithm. At the fusion center, to
exploit the multi-user cooperation gain, local spectrum sensing
results obtained by spatially distributed SUs are combined and
a global decision is then made according to a certain fusion rule.
The goal of CR spectrum sensing is to detect between two
hypotheses: H1 or H0, which means PU signal present or absent
B

g s ( t ) + n ( t ) ,
r (t ) =
n ( t ) ,

H1 ,

(1)

H0 ,

Under these two hypotheses, the traditional sensing


performance metric for CR spectrum sensing optimization is
usually pursued by either minimizing the probability of missdetection Pm for a given probability of false-alarm Pf or
minimizing Pf for a given Pm. In this paper, however,
considering both of these two kinds of detection errors should
associatedly be decreased as largely as possible for better
comprehensive sensing performance, we hence focus on the
minimization of the probability of total detection errors. First of
all, a target function based on the law of total probability is
established for following optimization of various sensing
parameters, and meanwhile it denotes the comprehensive
sensing performance metric of CR spectrum sensing in this
paper. In particular, since the miss-detection and false-alarm
errors take place corresponding to H1 and H0 cases respectively,
our target function is thus formulated in terms of a linear
combination of such two kinds of detection errors probabilities
with the spectrum occupancy statistics of PU as weights
B

Pf = Pr (Y > | H 0 ) =

2 , ,

( u, 2 )
(u )

(4)
(5)

where is the sensing threshold, Qu(.,.) is generalized Marcum


Q-function, (.,.) and (.) are incomplete and complete Gamma
functions respectively. It is also worth noting that in the case of
a certain fading detecting channel condition Qu(.,.) in Pm of (4)
should be further averaged over the particular statistics of SNR
of corresponding fading detecting channel while Pf of (5)
remains the same since Pf is independent of the SNR of
detecting channel.
B

where r(t) is the signal received by a SU, s(t) represents the


transmitted PU signal, n(t) is the zero-mean additive white
Gaussian noise (AWGN), and g indicates the amplitude gain of
detecting channel between PU and SU.

Pm = Pr (Y < | H1 ) = 1 Qu

Then, based on our established target function as proposed


in Section II, the probability of total detection errors in local
spectrum sensing with adopting energy detection can be
obtained by substituting (4) and (5) into (2)

Pe , l = P ( H1 ) 1 Qu

2 ,

)) + P ( H

( u, 2 )
(u )

(6)

Pe  P ( H1 ) Pm + P ( H 0 ) Pf ,

III. OPTIMIZATION FOR LOCAL SPECTRUM SENSING


As mentioned above, energy detection is usually executed
as the local spectrum sensing algorithm in CR technology. At a
SU, after squaring and integrating the received signal, the
decision statistic of energy detection, Y, has the distribution [5]
H1 ,
H0 ,

(2)

where P(H1) and P(H0) represent the probabilities of PU


presence and absence respectively, and in the meantime they
also satisfy the constraint equation of P(H1)+ P(H0)=1.

22u ( 2 ) ,
Y = 2
2u ,

According to (6), for the given SNR of detecting channel as


well as the known spectrum occupancy statistics of PU which
are both assumed to be accessible by certain measurements, the
probability of total detection errors Pe,l in local spectrum
sensing thus become a binary function with two variables of
parameters, i.e., time-bandwidth product u and sensing
threshold . Therefore, we can develop the optimization of such
two sensing parameters jointly for local spectrum sensing
through the method of extremum seeking, searching the
extreme minimum value of Pe,l to be exact as below

(3)

where u is the time-bandwidth product, is the SNR of


2
detecting channel, and 22u ( 2 ) and 2u
represent non-central
and central chi-square distributions each with 2u degrees of
freedom and a non-centrality parameter 2 for the former one.
When the detecting channel is assumed to be the AWGN
channel, the probabilities of miss-detection and false-alarm of
local spectrum sensing at a SU with energy detection are as [6]

Pe, l
u u = u
Pe , l
= 

= 0,
= 0,

2 Pe , l
u2

u = u

2 Pe , l
2

= 

> 0,
(7)
> 0,

where (u ,  ) are the jointly optimal sensing parameters gotten


over the two dimensional space of both the time-bandwidth
product and the sensing threshold for local spectrum sensing.
IV.

OPTIMIZATION FOR COOPERATIVE SPECTRUM SENSING

During CR spectrum sensing, due to the practical problem


of hidden terminal and the serious impact of fading detecting
channel, purely depending upon the implementation and
optimization of local spectrum sensing may not be reliable
enough to ensure achieving the sensing performance
requirement. In reality, by taking advantage of the multi-user
spatial diversity gain, several spatially distributed SUs are
usually incorporated to execute cooperative spectrum sensing.
Through cooperative spectrum sensing, the sensing
performance can therefore be improved efficiently. In the
process of cooperative spectrum sensing, every SU firstly
performs local spectrum sensing and reports the observation
result to a fusion center independently, and then the fusion
center makes a global decision.

At the fusion center, according to the difference of how to


make the global decision, there are logically three kinds of
fusion rules, i.e., OR-rule, AND-rule, and K-out-of-N-rule. In
the OR-rule, the global decision declares a PU exists as long as
one SU detects the PU. In the AND-rule, PU is viewed to be
appeared only when all SUs report PU presence. In the K-outof-N rule, the global decision announces an appearance of PU
when at least K SUs discover a PU exists. For the sake of
simplicity in our analysis, we assume that every SU has the
same sensing ability. Hence, the probabilities of miss-detection
and false-alarm of the three kinds of fusion rules during
cooperative spectrum sensing are respectively given as
Pm = Pm, i N ,
OR rule :
N
Pf = 1 (1 Pf , i ) ,

(8)

P = 1 (1 P ) N ,
m, i
AND rule : m
N
Pf = Pf , i ,

(9)

( )
( )

N
j

N
N j
Pm = 1 j (1 Pm , i ) Pm, i ,
K out of
j=K
N
N rule :
N j
Pf = N Pf , i j (1 Pf , i ) ,
j

j=K
B

(10)

Consequently, an optimization problem in cooperative


spectrum sensing naturally arises, i.e., which is the optimal
fusion rule with the given total number of collaborative SUs.
Specifically, by comparing (8) and (9) with (10), it is worth
emphasizing that both OR-rule and AND-rule are two special
cases of K-out-of-N-rule, respectively corresponding to K=1
and K=N cases. Therefore, the optimization of fusion rule for
cooperative spectrum sensing can be transformed into the
problem of deriving the optimal parameter K for the purpose of
maximally decreasing the probability of total detection errors.
After substituting (10) into (2), we can then obtain the target
function for optimization of the parameter K as

( )
( )

j
Pe , c = P ( H1 ) 1 N (1 Pm , i ) Pm, i N j
j
j=K

N
N j
N
j
P
1 Pf , i ) .
+ P ( H0 )
j f ,i (
j=K

(11)

Since K is a discrete variable, the optimization process for


parameter K can thus be developed by performing the discrete
difference operation on (11)
Pe , c ( K ) = Pe , c ( K ) Pe , c ( K 1)
K 1
= P ( H1 ) N (1 Pm, i ) Pm, i N K +1
K 1
N K +1
P ( H 0 ) N Pf , i K 1 (1 Pf , i )
.
K 1

( )
( )

l = min
K

1 Pf , i
P ( H1 )

ln

N ln P

P
H
(
)
m, i
0

+ 1 , N ,
1 Pf , i
1 Pm, i

ln Pm, i + ln Pf , i

(12)

l for
Then, the optimal parameter value of fusion rule K
cooperative spectrum sensing can be obtained by letting (12) be

(13)

l being an
where means the floor function for keeping K
integer. From (13), we can note that the optimal fusion rule is
dependent on the spectrum occupancy statistics of PU and the
detecting channel condition since it directly affects Pm,i.
B

where Pm,i and Pf,i are the probabilities of miss-detection and


false-alarm at the ith distributed SU (SUi) derived from (4) and
(5), and N is the total number of SUs participating in
cooperative spectrum sensing.
B

equal to zero and meanwhile should always be kept no greater


than the amount of collaborative SUs

V.

SIMULATIONS AND RESULTS

In this section, the optimization of parameters for both local


and cooperative spectrum sensing is evaluated numerically.
According to our established target function, the following
simulations mainly focus on analyzing the comprehensive
sensing performance and seeking the optimal selections of
various parameters, in order to minimize the probability of total
detection errors occurring during CR spectrum sensing.
Fig.1 illustrates the comprehensive sensing performance
metric quantified by the probability of total detection errors
over the two dimensional space of time-bandwidth product u
and sensing threshold in local spectrum sensing. From the
figure, it can be clearly observed that the total detection errors
occur less frequently when the options of u and are selected
around the diagonal area of the two dimensional space of u and
compared with other areas. Intuitively speaking, in CR
spectrum sensing applications, the selections of u and from
such optimal area can produce better comprehensive sensing
performance. Moreover, the difference of achieved
performance and the slender shift of optimal selection area
between Fig.1 (a) and (b) reveals that the optimization and
selection of parameter values of u and can be slightly varied
according to the particular spectrum occupancy statistics of PU
represented by P(H0) and P(H1).
Then, we analyze the optimal solution for the parameter K
of fusion rule. Here, we present a paradigm that totally N=8
SUs participate in the cooperative spectrum sensing. Fig.2 and
Fig.3 depict the probability of total detection errors with all
possible values of K conducted in AWGN and Rayleigh fading
environment cases respectively. From these two groups of
figures, it is obvious that the optimal K through the examined
range of sensing threshold is located at either the half voting as
K=4 or the minority voting as K=2 for either AWGN or
Rayleigh fading detecting channels. Noticeably, such meaningful difference demonstrates that the selection of optimal fusion
rule should depend on the exact probability distribution and
condition of detecting channels. In addition, by comparing (a)
with (b) of both Fig.2 and Fig.3, it reveals that the parallel shift
of sensing threshold corresponding to the optimal K is also due
to the variance of spectrum occupancy statistics of PU.
VI.

CONCLUSIONS

In this paper, in order to associatedly decrease the


probability of total detection errors, we propose a novel target

function for optimization of a series of sensing parameters,


which characterizes the comprehensive sensing performance
metric. Then, pursuing the minimization of such established
target function, we develop the joint optimization of timebandwidth product and sensing threshold for local spectrum
sensing. Furthermore, we derive the closed-form expression of
the optimal fusion rule for cooperative spectrum sensing. By
performing simulations and discussing results with variant
spectrum occupancy statistics of PU, we find the diagonal area
over the two dimensional space of time-bandwidth product and
sensing threshold is the optimal area of joint parameters
selection for local spectrum sensing, and we also numerically
evaluate the effect of different detecting channel conditions on
the optimization of fusion rule which implies the half voting is
optimal for the AWGN detecting channel case whereas the
minority voting is optimal for the Rayleigh fading case in
cooperative spectrum sensing.

REFERENCES
[1]

Ian F. Akyildiz, Won-yeol Lee, et al., Next generation/dynamic


spectrum access/cognitive radio wireless networks: a survey, Computer
Networks, vol. 50, no. 13, pp. 2127-2159, Sep. 2006.
Yue Wang, Chunyan Feng, et al., A robust and energy efficient
cooperative spectrum sensing scheme in cognitive radio networks, 11 th
International Conference on Advanced Communication Technology,
Gangwon-Do, South Korea, Feb. 2009, pp. 640-645.
Peh E. and Ying-Chang Liang, Optimization for Cooperative Sensing
in Cognitive Radio Networks," IEEE Wireless Communications and
Networking Conference 2007, Hong Kong, China, Mar. 2007, pp. 27-32.
W. Zhang, R. Mallik, et al., Cooperative Spectrum Sensing
Optimization in Cognitive Radio Networks, IEEE International
Conference on Communications 2008, Beijing, China, May 2008, pp.
3411-3415.
H. Urkowitz, Energy detection of unknown deterministic signals," Proc.
IEEE, vol. 55, pp. 523-531, Apr. 1967.
F. F. Digham, M. S. Alouini, et al., On the energy detection of
unknown signals over fading channels, IEEE Transactions on
Communications, vol. 55, no. 1, pp. 21-24, Jan. 2007.

[2]

[3]

[4]

[5]
[6]

(b) P(H0)=0.5, P(H1)=0.5


Probability of Total Detection Errors, P e,l

Probability of Total Detection Errors, P e,l

(a) P(H0)=0.8, P(H1)=0.2

10

-1

10

10
8
6

25
4
2

10

30

20

15

10

-1

10

10
8
6

25
4
2

Time-Bandwidth Product, u

Time-Bandwidth Product, u

Sensing Threshold,

30

20

15

10

Sensing Threshold,

Figure 1. Probability of total detection errors vs. time-bandwidth product and sensing threshold of local spectrum sensing (SNR=10dB)

10

10

(a) P(H0)=0.8, P(H1)=0.2

10

Probability of Total Detection Errors, P e,c

Probability of Total Detection Errors, P e,c

10

-1

-2

10

K=1
K=2
K=3
K=4
K=5
K=6
K=7
K=8
15

20

25

30

35

40

45

10

10

(b) P(H0)=0.5, P(H1)=0.5

-1

-2

10

50

K=1
K=2
K=3
K=4
K=5
K=6
K=7
K=8
15

20

25

30

35

40

45

50

Sensing Threshold,

Sensing Threshold,

Figure 2. Probability of total detection errors vs. sensing threshold when 8 SUs cooperate with different fusion rules in AWGN environment (SNR=10dB)
(a) P(H0)=0.8, P(H1)=0.2

10

Probability of Total Detection Errors, P e,c

Probability of Total Detection Errors, P e,c

10

-1

10

-2

10

10

K=1
K=2
K=3
K=4
K=5
K=6
K=7
K=8
15

20

25

30

35

Sensing Threshold,

40

45

50

10

10

(b) P(H0)=0.5, P(H1)=0.5

-1

-2

10

K=1
K=2
K=3
K=4
K=5
K=6
K=7
K=8
15

20

25

30

35

40

45

50

Sensing Threshold,

Figure 3. Probability of total detection errors vs. sensing threshold when 8 SUs cooperate with different fusion rules in Rayleigh fading environment (SNR=10dB)

Das könnte Ihnen auch gefallen