You are on page 1of 4

Efcient Energy Detector for Spectrum Sensing in

Complex Gaussian Noise


Guowei Zhang, Ju Liu, Lei Chen, Weidong Guo
School of Information Science and Engineering, Shandong University,
Jinan, Shandong, 250100, China
Email: zhangguowei@sdu.edu.cn
AbstractThe energy detector is a very useful detector for
spectrum sensing in cognitive radio systems and has been widely
employed for sensing optimization. However, there is certain mis-
match between the theoretical and actual detection performance
of the conventional energy detector in complex Gaussian noise,
which will undermine the optimization accuracy. To solve this
problem, an efcient energy detector in complex Gaussian noise
is proposed and its theoretical detection performance expressions
are derived based on the approximated distributions of its test
statistic under different hypothesis. Numerical results show that
compared with the conventional energy detector the new energy
detector offers superior detection performance and its theoretical
detection performance is almost as same as the actual one,
which conrms the new energy detectors efciency and is of
great importance for the design of optimal spectrum sensing in
cognitive radio systems.
I. INTRODUCTION
Traditional static management of radio resource has caused
a serious under-utilization of the frequency spectrum, accord-
ing to the measurements conducted by the Federal Communi-
cations Commission(FCC) [1]. On the other hand, the prolif-
eration of wireless equipments and services requires a heavy
demand of radio resource. Cognitive radio technology [2] has
been proposed as an attractive technique to cope with this
problem by improving spectrum utilization for future wireless
systems. In cognitive radio systems, spectrum sensing is a key
enabling functionality and needs to reliably detect the presence
and absence of the primary user so that the secondary user can
opportunistically access the under-utilized frequency band of
the primary user with little or no performance degradation to
the primary user.
Various spectrum sensing algorithms have been proposed,
such as the energy detector [3], [4], matched lter detector [5],
and cyclostationary feature detector [6]. The energy detector
is one of the most commonly employed spectrum sensing
schemes, since it does not require any prior knowledge about
the structure of the primary users signal or the channel.
Cooperative sensing has been proposed to enhance the de-
tection performance via the cooperation of several secondary
users, and cooperative sensing optimization has been further
researched in under complex Gaussian noise by utilizing the
theoretical performance indexes of the conventional energy
detector. However, to our best knowledge, there has been
no investigation about the match between the theoretical
and actual performance indexes of the conventional energy
detector under complex Gaussian noise, which is critical to the
accuracy of the spectrum sensing optimization. We nd that
there is a serious mismatch between the theoretical and actual
performance indexes of the conventional energy detector under
complex Gaussian noise. This will undermine the accuracy of
the spectrum sensing optimization.
To solve this problem, this paper proposes a new and
efcient energy detector for the spectrum sensing of cognitive
radio under complex Gaussian noise. Also the theoretical
expressions of the performance indexes of the new energy
detector are derived. The efcient energy detector is based on
more accurate distribution approximations of the test statistic
with a simple modication to the conventional energy detector.
Simulations conrm the superiority of the proposed energy
detector to the conventional one, not only the match of
the performance indexes, but also a more efcient detection
performance.
II. SYSTEM MODEL
We consider the problem of detecting the presence of one
primary user at a given channel based on the signal observed
by the secondary user. Specially, assuming a general additive
white Gaussian noise (AWGN) channel between the primary
and secondary users, the sample signal () received by a
secondary user, dened as () (

) with 1/

being the
sampling rated. Then the spectrum sensing problem becomes
the following hypothesis test problem:
() =
{

1
: () +()

0
: ()
= 1, . . . , (1)
where
0
represents the hypothesis that the primary user
is idle and
1
represents the hypothesis that the primary
user is active. = 1, 2, . . . , index the -th signal sam-
ples with being the assigned sensing time. () is the
received signal by the secondary user with () denotes the
transmitted symbol by the primary user, and () is the
additive noise at the secondary user and is further modeled
as an independent and identically distributed (i.i.d.) circularly
symmetric complex Gaussian (CSCG) random variable with
zero mean and variance
2
, that is () (0,
2
) and

(),

() (0,
2
/2) where

() and

() represent
the real and imaginary components of the noise, respectively.
We also assume that () and () are independent of each
other.
978-1-4244-7555-1/10/$26.00 2010 IEEE
As to the detection performance of spectrum sensing algo-
rithm, there are two measurements: the detection probability

dened as the probability that the secondary successfully


declares the presence of primary signal when in fact the
primary user is active and the false alarm probability

which
is the probability that the secondary user falsely announce the
presence of primary signal when the secondary user is actually
idle.
III. NEW AND EFFICIENT ENERGY DETECTOR
A. Conventional Energy Detector
The test statistic of the conventional energy detector under
complex Gaussian noise is dened as [7][9]
=

=1
()
2

2
=

=1

()
2
+

()
2

2
. (2)
And several kinds of distribution have been proposed to
approximate the distribution of the test statistic under
0
and
1
, respectively. Specically, under
0
the probability
density function (PDF) of can be approximated by the Chi-
square distribution with degrees of freedom or the Gaussian
distribution with mean and 2 variance. That is,

(
0
) =

(
0
) :
1
2
/2
(

2
)

2
1
exp
(

2
)

(
0
) :
1

4
exp
{

()
2
4
}
(3)
where () is the gamma function. On the other hand, under

1
the PDF of can be approximated by the non-central Chi-
square distribution with degrees of freedom and parameter
or the Gaussian distribution with (+) mean and 2(+2)
variance:

(
1
) =

(
1
) :
1
2
(

2
)exp(
2+
2
)
1
(

2)

(
1
) :
1

4(+2)
exp
{

()
2
4(+2)
}
(4)
where =

=1
()
2

2
is times the average SNR at
the output of the energy detector and

() is the th-order
modied Bessel function of the rst kind.
The decision rule of the energy detector is given by

0
(5)
where is the corresponding decision threshold. According to
the (3) and (4), the theoretical detection performance of the
conventional energy detector for a specic decision threshold
will be

= {
0
} =
{

:
(

2
,

2
)
(

2
)

: (

2
)
(6)
and

= {
1
} =
{

: 1 (, , )

: (

2(+2)
)
(7)
where () is the complementary cumulative distribution
function (CDF), i.e., () =

exp(
2
/2)/

2; and
() is the noncentral chi-square CDF, i.e., (, , ) =

=0
[
(
1
2
)

!
exp

2
](
2
+2
).
B. The Efcient Energy Detector
For the conventional energy detector, we nd that the theo-
retical detection probability is far from the actual one, which
implies that there is lack of accurate theoretical detection
performance expressions of the conventional energy detector
for the existing sensing optimizations under complex Gaussian
noise. To solve this problem and improve the detection perfor-
mance of the conventional energy detector, an efcient energy
detector is proposed in this paper with certain improvement
of the conventional energy detector. The test statistic of the
improved energy detector is

=1
()
2

2
/2
=

=1

()
2

2
/2
+

=1

()
2

2
/2
. (8)
As a result of the modication, under
0
,

is just the sum


of the squares of 2 Gaussian random variables with zero
mean and unit variance, which follows a central chi-square
distribution with 2 degrees of freedom and

0
) =
1
2

()

(1)
exp
(

2
)
. (9)
Also according to the central limit theorem when the number
of samples is large enough (e.g., 10 in practice), the test
statistic

is asymptotically normally distributed with mean


2 and variance 4. That is

0
) =
1

8
exp
{

2)
2
8
}
. (10)
Similarly, under
1
the test statistic

would follow a Gaus-


sian distribution with mean 2( +) and variance 4( +2)

1
) =
1

8( + 2)
exp
{

2 2)
2
8( + 2)
}
.
(11)
In [10], Gamma distribution has been proposed to approximate
the distribution of the test statistic under
1
in the situation of
real signals. Here we extend this approximation to the situation
of complex signals. That is,

1
) =
1

()

(1)
exp
(

)
(12)
where = 2(1 +/) and = .
The decision rule of the improved energy detector is similar
to (5)

(13)
where

is the corresponding decision threshold which is


changed accordingly. Thus based on (9) (12) the theoretical
detection performance of the improved energy detector will be

= {

0
} =

:
(,

2
)
()

: (

4
)
(14)
and

= {

1
} =
{

: (

22

4(+2)
)

: 1 (

, , )
(15)
where (, , ) =

0
1

()

(1)
exp(

) denotes the
CDF of a Gamma distribution with scale parameter and
shape parameter .
Note that although here we focus on the situation of AWGN
channel, the results (14) and (15) can be easily extended
to the fading channels. Specically, the detection probability
will be averaged over the particular statistics of the SNR ()
of the corresponding fading channel, while the false alarm
probability remains the same as a result of its independence
of the SNR of the channel.
IV. NUMERICAL RESULTS
In this section, we provide simulations to illustrate the
performance of the proposed energy detector, and compare
it with that of the conventional energy detector. Given the
target false alarm probability, the decision threshold can be
calculated according to the approximated distributions under

0
. Then the theoretical detection probability can be obtained
by using the approximated distributions and the corresponding
decision threshold. Note that the decision threshold can be
calculated according to the Chi-square CDF (

and

) or
the Gaussian CDF (

and

). 10000 realizations have been


conducted to calculate both the actual false alarm probability
and the actual detection probability. And QPSK signals are
generated as the primary users signals.
Match of the false alarm probability: In Fig.1, we draw
the simulated false alarm probability against the target false
alarm probability for different detectors. It is seen that for the
conventional energy detector (ED:

or ED:

), its actual

doesnt match well with the target

. In other words, the


distribution approximation of the conventional energy detector
(3) is not quite efcient. This is a good motivation for our
research to improve the spectrum sensing reliability of energy
detector. Shown in Fig.1, the proposed energy detector using

(New-ED:

) has the best performance with its actual

almost as same as the target

. And the actual

of the
proposed energy detector using

(New-ED:

) is more
close to the target one than the conventional energy detectors.
These results conrm the accuracy of the proposed distribution
approximations of the test statistic of our efcient energy
detector under
0
, as well as the efciency of the proposed
theoretical false alarm probability expressions (14).
Match of the detection probability: Next, simulations are
conducted to investigate the match between the theoretical
detection probability and the actual detection probability of
the energy detectors. Here both

) and

) are
0 0.2 0.4 0.6 0.8
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
A
c
t
u
a
l

V
a
l
u
e
s

o
f

P
f
.
(a) Target Values of P
f
with N=10
0 0.2 0.4 0.6 0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
A
c
t
u
a
l

V
a
l
u
e
s

o
f

P
f
.
(b) Target Values of P
f
with N=20
ED:
C
ED:
G
NewED:
C

NewED:
G

Fig. 1. False alarm probability comparison of different detectors in complex


Gaussian noise with different .
employed as the decision threshold to calculate the actual
detection probability of these detectors while only

)
is used to calculate the theoretical detection probability based
on the previous ndings.
0.2 0.4 0.6 0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
D
e
t
e
c
t
i
o
n

P
r
o
b
o
.
(a) Target P
f
, SNR=10dB
0.2 0.4 0.6 0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
D
e
t
e
c
t
i
o
n

P
r
o
b
o
.
(b) Target P
f
, SNR=6dB
Simu:
C
Simu:
G
Theo: P
d
Gau
Theo: P
d
Chi
Fig. 2. Comparison of the theoretical

to the simulated

of the
conventional energy detector with = 20.
Fig. 2 plots the theoretical and actual

of the conventional
energy detector versus the target

. We can observe that


although the conventional energy detector using

(Simu:

) can achieve a higher overall actual

than that using

(Simu:

), none of the approximated theoretical functions


(Theo:

or Theo:

) can match well with the actual


values and the difference between these values are obviously
not negligible. These inaccurate theoretical detection perfor-
mance expressions (6) and (7) of the conventional energy
detector will undermine the efciency of the spectrum sensing
optimization.
In Fig. 3, we investigate the accuracy of the theoretical
detection probability expressions of the proposed energy de-
tector. As we can see, the proposed energy detector using

(Simu:

) can achieve a higher overall actual

than that
0.2 0.4 0.6 0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
D
e
t
e
c
t
i
o
n

P
r
o
b
o
.
(a) Target P
f
, SNR=10dB
0.2 0.4 0.6 0.8
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
D
e
t
e
c
t
i
o
n

P
r
o
b
o
.
(b) Target P
f
, SNR=6dB
Simu:
C

Simu:
G

Theo: P
d
Gau
Theo: P
d
Gam
Fig. 3. Comparison of the theoretical

to the simulated

of the proposed
energy detector with = 20.
using

(Simu:

). Also note that using

the proposed
energy detector can make sure its actual

almost as same
as the target one, as displayed in Fig. 1. As to the accuracy
of the different theoretical expressions of

, it is easy to
see that the Gamma approximation (Theo:

) can match
very well with the actual detection performance. While the
values of the Gaussian approximation (Theo:

) deviate a
little from the actual detection performance. It means that the
theoretical expressions

in (14) and

in (15) should
be used for the design of spectrum sensing optimization.
Detection probability comparison: Fig.4 plots the simulated

of the energy detectors under different scenarios where

and

are employed for the conventional energy detector


(ED) and the proposed energy detector (New-ED), respec-
tively. In Fig.4-(a), it is observed that our efcient energy
detector can achieve a better detection performance than the
conventional energy detector for a realistic region of the target

. For example, when the target

= 0.05 and = 50,


the proposed energy detector has an actual

around 0.2
which is almost three times of that of the conventional energy
detector. Fig.4-(b) shows the simulated

of different energy
detectors versus SNR with the same target

of 0.1. One
sees that obvious performance improvement can be achieved
by the proposed energy detector over the conventional energy
detector with about 1dB average gain. So the new energy
detector is of great superiority to the original energy detector
in terms of both the actual detection performance and the
accuracy and efciency of theoretical detection performance
expressions which are critical to the sensing optimization.
V. CONCLUSIONS
A new and efcient energy detector is proposed for spec-
trum sensing in cognitive radio under complex Gaussian noise,
as well as the theoretical expressions of its detection perfor-
mance. Series of simulations verify the efciency of the pro-
posed energy detector and its superiority to the conventional
one. That is, its theoretical detection performance can match
10
2
10
1
10
2
10
1
10
0
S
i
m
u
l
a
t
e
d

D
e
t
e
c
t
i
o
n

P
r
o
b
o
.
(a) Target P
f
SNR=10dB
10 8 6 4 2 0 2
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S
i
m
u
l
a
t
e
d

D
e
t
e
c
t
i
o
n

P
r
o
b
o
.
(b) SNR (dB) Target P
f
=0.1
ED: N=20
NewED: N=20
ED: N=50
NewED: N=50
Fig. 4. Actual

comparison of the energy detectors under different


scenarios.
well with the actual detection performance which conrms
the accuracy of the approximated distribution functions. And
the proposed energy detector can achieve a higher detection
probability than the conventional one when the target

is
within certain realistic region. Some interesting extensions of
this work may include studying the sensing optimization based
on the new energy detector.
ACKNOWLEDGMENT
This work was supported by National Natural Science Foun-
dation of China (60872024), the Cultivation Fund of the Key
Scientic and Technical Innovation Project (708059), Natural
Science Foundation of Shandong Province (No.Y2007G04),
open research fund of National Mobile Communications Re-
search Laboratory (W200802), and the State Key Lab. of
Integrated Services Networks (ISN9-03).
REFERENCES
[1] F. C. Commission, Spectrum policy task force, Rep. ET Docket, Tech.
Rep., Nov. 2002, no.02-135.
[2] S. Haykin, Cognitive radio: brain-empowered wireless communica-
tions, IEEE J. Select. Areas Commun., vol. 23, no. 2, pp. 201220,
Feb. 2005.
[3] V. Kostylev, Energy detection of a signal with random amplitude, in
Proc. ICC, New York, May 2002, pp. 16061610.
[4] F. Digham, M.-S. Alouini, and M. Simon, On the energy detection of
unkown signals over fading channels, IEEE Trans. Commun., vol. 55,
no. 1, pp. 2124, Jan. 2007.
[5] H.-S. Chen, W. Gao, and D. G. Daut, Signature based spectrum sensing
algorithms for ieee 802.22 WRAN, in IEEE Intern. Conf. Comm., Jun.
2007, pp. 64876492.
[6] S. Haykin, D. J. Thomson, and J. H. Reed, Spectrum sensing for
cognitive radio, Proc. IEEE, vol. 97, no. 5, pp. 849877, May 2009.
[7] Z. Quan, S. Cui, and A. H. Sayed, Optimal linear cooperation for
spectrum sensing in cognitive radio networks, IEEE J. Select. Signal
Processing, vol. 2, no. 1, pp. 2840, Feb. 2008.
[8] J. Shen, T. Jiang, S. Liu, and Z. Zhang, Maximum channel throughput
via cooperative spectrum sensing in cognitive radio networks, IEEE
Trans. Wireless Commun., vol. 8, no. 10, pp. 51665175, Oct. 2009.
[9] W. Zhang, R. Malik, and K. Letaief, Optimization of cooperative
spectrum sensing with energy detection in cognitive radio networks,
IEEE Trans. Wireless Commun., vol. 8, no. 12, pp. 57615766, Dec.
2009.
[10] Y. Chen, Improved energy detector for random signals in gaussian
noise, IEEE Trans. Wireless Commun., vol. 9, no. 2, pp. 558563, Feb.
2010.