Energy-Detection Based Spectrum Sensing for Cognitive Radio

Zhai Xuping, Pan Jianguo

Key Laboratory of Special Fiber Optics and Optical Access Networks, Ministry of Education,
School of Communication and information engineering, Shanghai University, 200072, China
Email: [email protected]

Keywords: Cognitive radio, Spectrum sensing, Energy detection, shadowing fading. Three spectrum sensing or signal detection
Cooperative detection techniques are used in traditional systems: matched filter,
energy detector and cyclostationary feature detector. The
Abstract matched-filter-based detection is the optional way for any
signal detection[4]. A significant drawback of the matched-
In this paper we investigated the energy detection technique filter-based detection is that a cognitive radio would need a
for spectrum sensing in cognitive radio systems. The dedicated receiver for every primary user class.
detection performance and the impact of the noise uncertainty Cyclostationary feature detection method is recognized as a
on the detection probability are analyzed theoretically. To more effective way for the detection of very weak signals in a
deal with the hidden terminal problem and the local spectrum background of noise[2]. The computational complexity of
sensing in wireless signal detections, a distributed M- cyclostationary feature detection needs further researches.
cooperative sensing scheme is proposed. Through analysis Energy detection is a sub-optional technique which performs
and simulation, we have shown the benefits of the proposed non-coherent detection. Although there are several drawbacks,
scheme in increasing the agility of cognitive radio systems. such as the threshold setting, detection of spread spectrum
With small tradeoffs between the detection probability and signal etc., One most advantage of energy detection is its
the false alarm probability, the scheme improves the spectrum simplicity in implementation and the short sensing time.
sensing ability greatly in low SNR situations. Some studies [5] reveal that it is essential to perform two
levels of detection: coarse sensing and fine sensing. The
1 Introduction energy detection can operate as the first level coarse sensing
due to its low computational complexity and short sensing
The radio frequency spectrum is becoming scarce. time. With varying conditions on the channel between a
Conventional fixed spectrum allocation however results in primary user and a cognitive radio due to severe fading,
low utilization of the allocated spectrum. Actual shadowing and building penetration losses, spectrum sensing
measurements show that most of the allocated spectrum is can become extremely difficult. To improve the performance
vastly underutilized with utilization ranging from 15% to 85~ of these detection techniques in such scenarios, cooperative
in the bands below 3 GHz at any specific location and sensing scheme is proposed [2,6,7]. In [7], a centralized
time[I,2]. At frequencies above 3GHz the actual utilization is cooperative sensing scheme is proposed where the access
dramatically lower. Recently cognitive radio [3] is proposed point collects sensing results form all users, monitors the
as a novel technology to improve the spectrum utilization. A channel conditions and performs channel allocations.
"Cognitive Radio" is a radio that is able to sense the spectral
environment over a wide frequency band and exploit this In this paper, we focus on the energy detection based
information to opportunistically provide wireless links that spectrum sensing for cognitive radio. The performance of
best meet the user communications requirements. By allowing energy detection is analyzed under noise uncertainty. To
secondary networks to borrow unused radio spectrum from improve the detection probability, a distributed M-
primary licensed networks or to share the spectrum with the cooperative sensing scheme is proposed and its performance
primary networks, cognitive radio is expected to greatly is analyzed. Simulation and analysis shows the great benefits
improve the spectrum utilization. of the proposed scheme. The rest of this paper is organized as
follows. In section 2, the principle and the process of energy
One of the fundamental issues for cognitive radio is to sense detection is reviewed briefly. The effect of noise uncertainty
the spectrum environment accurately and determine whether on energy detection and the detection limit is analyzed in
the primary user is active or not over a specific band reliably. section 3. Then in section 4, a distributed M-cooperative
Therefore, in order to guarantee non-interference with the sensing scheme is proposed. The performance of the scheme
primary user, cognitive radio must detect very weak signals. is analyzed and simulated. Finally the conclusions are drawn
This sensing operation is considered to be more difficult in a in section 5.
degraded wireless channel due to random fading, especially
2 Principle of Energy Detection
This work is supported by foundation for development from the Energy detection is a sub-optional signal detection technique
education committee of Shanghai (No. 05AZ46), and by the
which has been extensively used in radiometry. The detection
foundation of education committee of Shanghai for outstanding
young teachers.
process can be performed in both time domain and frequency
domain. To measure the signal power in a particular
frequency band in time domain, a bandpass filter is applied to Where 1
Q(x}= ~fexp -y
"12ft
00

x
(

2
2} y.
the target signal and the power of the signal samples is the
measured. To measure the signal power in frequency domain,
the time domain signal is transformed to frequency domain In case that the signal power is unknown, we can first set the
using FFT and the combined signal power over all frequency false alarm probability Pia be a specific constant. By
bins in the target frequency band is then measured. Fig.l equation (5), the threshold y can be determined. Then, for the
shows the detection process. fixed number of samples N, the detection probability ~ can
y(t) {\
be evaluated by substituting the threshold in (4). Using
equation (4) and (5), we can determine the required minimum
(a) Implementation in time domain
observation window N min which is a function of signal to
. ratIon
nOIse . S h
'NR, were u
SNR = _x .
(b) Implementation in frequency domain Uw
Fig.l the process of energy detection
(6)
Without loss of generality, a complex baseband equivalent of Ideally, if the observation window is not limited, an energy
the energy detector is considered. The detection is the test of detector can meet any desired ~ and Pia simultaneously. In
the following hypotheses:
fact, there is a minimum SNR below which signal cannot be
H 0: Y[n] = W [n] signal absent
detected.
H t: Y[n]=X[n]+W[n] signal present (1)
n = 1, ... ,N; where N is the observation window 3. Effect of Noise Uncertainty
Where X[n] is the sample of the target signal with power u; .
Noise in most communication systems is an aggregation of
The noise sample W[ n] is assumed to be additive white
various independent sources including not only thermal noise,
Gauss (AWGN) with zero mean and variance u;.
The but also interferences due to nearby unintended emissions. So
decision statistic for energy detector is: the assumption that the noise is a Gaussian random process is
N 2 always appropriate. Further, the variance of the noise could
T = L(Y[n]) (2) vary over time and it cannot be estimated exactly.
n=1

In the absence of coherent detection, the signal samples X[ n]

The above section shows that the detection threshold r is
can be modeled as a Gaussian random process with variance
proportional to u;. In most practical situations, u; (and r
u; . So Y[ n] is also a Gaussian random process. It can be consequently) would need to be estimated at the receiver.
A A

show that in the low SNR regime, the number of required Denote the estimates by u~ and r. Assume the error in
samples N is large. By central limit theorem, the test statistic
can be approximated as a Gaussian distribution.
estimating u; is :
A

n
Ho: T,.... Normal ( Nu;,2Nu;) (1-&I)U; ~u~ ~(1+&2)U~, O~&1 <1; &2 ~O
(3)
HI: T - Normal (N( 0"; + a; ),2N( 0"; +0"; L et U A1+ -
=- &2 ~ 1 ,to
1-81
meet th e d '
etectIon .
requuement 0f

The performance of detection is measured by two parameters

(Pla,Pd ) , the relationship between the SNR and the
~ and Pia . Each pair is associated with a particular
observation window can be determined as follows.
threshold"( that tests the decision statistic:
T > y decide signal present (U -l)JN +J2UQ-l(~/)-J2Q-l(~)
SNR = (7)
T < y decide signal absent J2Q-l(~)+JN
Then ~ and Pia can be evaluated as:
For large N, equation (7) can be simplified into
(4) J2l)Q-l (P ) - J2Q-l (P )
SNR = U -1 + af d
IN (8)

(5)
=U-l+o(lv )~(SNR)min
Where (SNR)min ~ (U -1) .
Equation (8) shows that for a specific U, there exists a In the proposed scheme, the exchanged sensing results are
minimum SNR below which the detection requirement of whether the signal is active or not in the interested frequency
(Pfa'~) cannot be met simultaneously. That is to say, under band.
Now the performance of the proposed cooperative sensing
the circumstance of noise uncertainty, the energy detection scheme can be evaluated as follows. We assume that every
cannot be applicable in the low SNR situation. Fig.2 shows
cognitive radio node is shadowed with equal probability ~h'
the limitation. To be used in cognitive radio in any situations,
we have to resort to some other schemes to compensate for its When a node is shadowed, the detection probability is
limitation. ~h,d and the false detection probability is ~h,f. When a node

is not shadowed, the detection probability is Pnsh,d and the

false detection probability is P"sh,f' According to OR rule, if
one or more than one nodes among M nodes have detected
the presence of the signal, the target node would determine
that the signal is present. So the detection probability of the
distributed M-cooperative sensing scheme is:
M M-n [ ( ( )M-n ]
~=LC~~.(I-~h) . l-l-~h,d )n l-Pnsh ,d (9)
n=O
The false alarm probability of the distributed M-cooperative
sensing scheme is:

lfa == LM C~ . ~~ .(1- ~h )M-n .[1-( 1- ~h,f )n . ( 1- Pnsh,f )M-n ] (10)

n=O

Fig.2 effect of noise uncertainty

Fig.3 - fig.6 shows the performance of the proposed
distributed M-cooperative sensing scheme. Both the analysis
4 Distributed M-Cooperative Sensing Scheme results and the simulation results are included. The analysis
No matter what a high sensitivity the detector has, the results are obtained by equation (9) and (10). The simulation
detection of wireless signal is essentially a local spectrum results are obtained by Monte Carlo experiments.
sensing[2]. The hidden terminal problem is difficult to
prevent due to the propagation loss of wireless signals, the Fig.3 shows the detection probability ~ against the detection
shadowing, and severe multipath fading. Even through the probability of non-shadowing node P"sh,d with the number of
probability of this scenario is low, cognitive radio should not cooperative sensing nodes M=l (no cooperation), M=3, and
cause interference to the primary user. In this section, a M=5. It shows that the cooperation in sensing can greatly
cooperative sensing scheme is proposed in which not only the improve the sensing sensitivity. Even the number of
effectiveness of the cooperation is considered, but also the cooperative sensing node is small, such as M==3, i.e., the
overhead of cooperating and protocol is taken into account. target node communicates with other two nodes, the detection
Suppose that any cognitive radio in the interested network has probability can also be greatly improved (non-cooperation:
more than one neighbouring node. Here if one node can ~ == 0.23. 3-cooperation: Pd == 0.57). In addition, it can be
communicate with another node directly, these two nodes are
called as neighbouring nodes. Some nodes are shadowed, or seen that in low SNR regions (the Pnsh,d is small), the
inside buildings with high penetration loss. The proposed cooperation is more effective. This is desired in cognitive
distributive M-cooperative sensing scheme performs the radio to prevent the hidden terminal problem.
following steps.
Step 1: Every cognitive radio node senses the specific ~:
spectrum frequency band independently using the energy ~ ~~
0.9
/
detection techniques. /' /
Step 2: Every node chooses randomly M nodes among 0.8
/ ./
/ V
its neighbouring nodes according to some rules (such as the
strongest M signals, the first M signals from different 0.7
/
/~.
directions, etc.) If the number of neighbouring nodes for a ~
0.6
/
IV
node is less than M, all neighbouring nodes are chosen. ~/
Step 3: Every cognitive node broadcasts its sensing results 0.5
V '''G'''Theory M= 1
via some way (such as a shared channel, etc.). 0.4
/ -+-Simulation M= I
",c"'Theory M=3
Step 4: Cognitive node received the broadcast message /~ -e-Simulation M=
"'6'''Theory M=5
determines if this message is useful and update its sensing
V
0.3
'-""""'Simulation M=

results according to the OR rule if necessary.

OJ 0.4 0.5 0.6 0.7 0.8 0.9
Pnsh.d
Fig.4 reveals the effect of shadowing on the detection detection performance is analyzed theoretically. To deal with
probability. It can be seen that with the increase of the the hidden terminal problem and the local spectrum sensing, a
probability of shadowing, the detection probability becomes distributed M-cooperative sensing scheme is proposed.
decreased with a flat slope. It illustrates that the proposed Through analysis and simulation, we have shown the benefits
scheme can improve the detection performance of node of the proposed scheme in increasing the agility of cognitive
undergoing shadow fading effectively. radio systems.

In the proposed distributed M-cooperative sensing scheme in

this paper, the exchanged sensing information is whether the
C."II.:.".
signal is active or not in the interested frequency band. In
--"'''-...,'''--t....-.......,
c. ~~ ; : ;. i , "''\'=--':~o.:..:''':'-~' ;. ., addition, no MAC protocol is considered. In future, we plan
to investigate the soft information exchange and how much
C. 7~ .... .... ; ...... .. ;...... .. ..;. .. .. .. ..:...... .. i .. G .. Theory M= 1 can be gained. Also we wish to develop cooperative protocols.
-&-Simulation M= 1
.. e Theory M= 3
C. ~ .. .. .... : ........ .. :.......... ..~ .......... ,:...... .. i ~Simulation M=3
.. 6 .. Theory M=5

C.5
~'---- ~ .......... Simulation M=5 0.035' .. G Theory M=I
-e- Simulation M= I
... e.. Theory M=3

..
~~~ 0.03 ~ Simulation M~Ii ; : ;.:/ i , -t
C.4~ i : : i ; ~ ; .,
.. 6.. Theory M=5
0.025 .... .......... Simulation M~5.

C. 1 C. 15 C. 2 C.25 C. 3 C. 35 C. 4 C. 4c C. 5 r:f 0.021- ; ; ; ;..~ ;''' ; ; .........-:-

Ps/1
Fig.4 Pd VS Psh(Psh,d=0.1,Pnsh,d=0.6) 0.0151- .. .. ~ :.. .. .. "JI'!' .. .. .. : ~ ~'!"'"!' : : .

O.OII- ".. ; .. -/o

Fig.5 studies the dependence of the detection probability on _ i.:...:~: .. ; .. .. .. .. .. _ -1

the detection probability of shadowing nodes. This study 0.OO5".. ,.:AF .. : ;.. .. .. .. : ~~:::::: ..':f ; -1

reveals that, with the increase of the detection probability of

shadowing nodes, the detection performance is improved little.
This verifies that the contribution of shadowing nodes in the
cooperative sensing is small due to the low SNR. Fig. 6 Pfa VS Pnsh,t<:Psh=O.3,Psh,FO.00l)

References
0.9 ~~r"
I~~~ [1] FCC Spectrum Policy Task Force, "FCC Report of the
0.81- .... ...... : ...... ....:...... .. ..,,~ .. .. ...... : ...... .. I .. G .. Theory M= I
Spectrum Efficiency Working Group," Nov 15, 2002,
--e- Simulation M= I
... e .. Theory M= 3 http://www.fcc.gov/sptf/files/SEWGFinaIReprotl.pdf.
~ O. 71- .. ,, .. ; .. : .. .. .. ..,,: ; .. 1 ~Simulation M= 3
'''6'''Theory M=5
[2] Danijela Cabric, Shridhar Mubaraq Michra, Robert W.
"""""*'- Simulation M= 5 Brodersen. "Implementation Issues in Spectrum Sensing
0.61- ; ;.. ~ : .. ; .. .. ;.. .,
~
for Cognitive Radios", In Proc. of Asilomar Conf. on
Signals, Systems, and Computers, 2004. pp.772-776.
[3] FCC, ET Docket no. 03-322. Notice of Proposed Rule
Making and Order, December 2003.
Ot.I 0.15 0.2 0.2~ 0.3 0.35 0.4 0.4~ 0.5 [4] John Proakis, Digtal Communications, 3rd edition, Mc
Psh.d Graw Hill.
Fig.5 Pd VS Psh,d(Psh=0.3,Pnsh,d=0.6) [5] Neihart, N.M. Roy, S. Allstot, DJ. , "A Parallel,
Multi-Resolution Sensing Technique for Multiple
The false alarm probability is analyzed in fig.6 when the OR Antenna Cognitive Radios," In Circuits and Systems,
rule is adopted. It can be seen that the false alarm probability 2007. ISCAS 2007. IEEE International Symposium on.
is also boosted. In cognitive radio, a high false alarm pp.2530-2533.
probability means a low utilization of spectrum. Although the [6] G. Ganesan and Y. G. Li, "Cooperative spectrum
proposed scheme may lower the utilization of the spectrum, it sensing in cognitive radio networks," in Proc. IEEE
prevents the interference between the cognitive radio and the Symp. New Frontiers in Dynamic Spectrum Access
primary user to a large extent. We think that it is worthwhile Networks (DySPAN'05), Baltimore, USA, Nov. 8-11,
to make this trade off. 2005, pp. 137-143.
[7] R. W. Brodersen, A. Wolisz, D. Cabric, S. M. Mishra,
5. Conclusion and Future Work and D. Willkomm, "Corvus: a cognitive radio approach
for usage of virtual unlicensed spectrum," July 2004,
In this paper we investigated the energy detection technique whitepaper.
for spectrum sensing in cognitive radio systems. The