You are on page 1of 14

SELECTED PAPERS FROM IEEE/CIC ICCC2013

Reputation-Based Hierarchically Cooperative


Spectrum Sensing Scheme in Cognitive Radio Networks
CHEN Huifang1, 2, XIE Lei1, 2, NI Xiong1
1

Department of Information Science and Electronic Engineering, Zhejiang University, Hangzhou 310027, China
Zhejiang Provincial Key Laboratory of Information Network Technology, Hangzhou 310027, China

Abstract: Cooperative spectrum sensing in cognitive radio is investigated to improve the detection performance of Primary User (PU).
Meanwhile, cluster-based hierarchical cooperation is introduced for reducing the overhead
as well as maintaining a certain level of sensing performance. However, in existing hierarchically cooperative spectrum sensing algorithms, the robustness problem of the system
is seldom considered. In this paper, we propose a reputation-based hierarchically cooperative spectrum sensing scheme in Cognitive
Radio Networks (CRNs). Before spectrum
sensing, clusters are grouped based on the
location correlation coefficients of Secondary
Users (SUs). In the proposed scheme, there
are two levels of cooperation, the first one is
performed within a cluster and the second one
is carried out among clusters. With the reputation mechanism and modified MAJORITY
rule in the second level cooperation, the proposed scheme can not only relieve the influence of the shadowing, but also eliminate the
impact of the PU emulation attack on a relatively large scale. Simulation results show that,
in the scenarios with deep-shadowing or multiple attacked SUs, our proposed scheme achieves a better tradeoff between the system
robustness and the energy saving compared
with those conventionally cooperative sensing
schemes.
Key words: cognitive radio networks; cooperative spectrum sensing; cluster; location correlation; reputation
January 2014
2014
China Communications January

I. INTRODUCTION
Cognitive Radio (CR) technology has been proposed as a promising technology to handle the
contradiction between the shortage of spectrum resource and the under-utilization of licensed spectrum [1]. It enables the Secondary
Users (SUs) to utilise licensed spectrum bands
assigned to the Primary Users (PUs) by opportunistic accessing [2]. The prerequisite to this
opportunistic access is to having no interference to the primary system. That is, SUs share
the licensed spectrum at the absence of PU, but
have to retreat when the PU is present. Hence,
spectrum sensing plays an important role for
SUs to opportunistically access the spectrum
allocated to the PUs in a CR network (CRN).
Spectrum sensing should be performed to
ensure that the spectrum is not occupied by PUs.
There are many spectrum sensing mechanisms,
such as energy detection, matched filter detection and cyclostationary detection, and so on.
Among them, energy detection is an engaging
method because of its simplicity and efficiency. With the problem of the multipath fading and shadowing, the detection performance
of single SU spectrum sensing may be not good enough. Thus, cooperative spectrum sensing, in which multiple SUs cooperate and share the sensing results, was considered to resolve the problem [3].
Many cooperative spectrum sensing schemes have been proposed to make the final decision based on sensing results from SUs [4-8].

Received: 2013-08-19
Revised: 2013-11-18
Editor: JIANG Xiaohong

25
12

This paper presents a


novel reputation-based
hierarchically cooperative spectrum sensing
scheme for cognitive
radio networks, which
is based on the cluster
structure of secondary
users, reputation mechanism and modified
MAJORITY combination,
to reduce the cooperation overhead and improve the system robustness, as well as to
maintain a certain level
of sensing performance.

26
13

In these schemes, the problem of multipath


fading and/or shadowing was resolved by exploiting the spatial diversity among SUs.
However, the cooperation among SUs may
also introduce communication overhead which
limits or even compromises the achievable cooperation gain. In Ref. [9], an optimal number
of SUs for cooperative sensing was derived to
maximise the utility of CRN. However, some
SUs in a real CRN environment may fall into
deep shadowing region and fail to take part in
the cooperation. In Refs. [3, 10], it had been
proven that having a small number of farther
located SUs may be more effective than having a large number of closely located SUs.
Hence, many researchers have been investigated the user selection problem in cooperative spectrum sensing, and some user selection
algorithms have been proposed [11-13]. In Ref.
[11], a user selection algorithm for CR ad-hoc
networks, which adaptively selected uncorrelated SUs based on the evaluation of the correlation experienced by SUs, was proposed to
improve the performance of cooperative sensing. In Ref. [12], an SU selection algorithm based on the channel condition between SU and
Fusion Centre (FC) was proposed to guarantee
the reliability of received sensing results, only
SUs with good report channel conditions are
selected to take part in the cooperation. In Ref.
[13], a user selection algorithm according to
the characteristics of cooperative SUs, namely
the sensing capacity and reporting delay for
local decisions, was proposed. The BPSO method was adopted to select SUs to participate
cooperative sensing.
On the other hand, clustering technology was
introduced by taking advantage of the efficient
network organisation to the system overhead
of cooperative spectrum sensing. Based on the
consideration that nodes located close to each
other will obtain similar sensing results in a high
probability, gathering SUs with geographical
correlation into a cluster to make a clusterlevel decision is a good way to decrease sensing overhead. By clustering, SUs are divided
into different clusters. Each cluster has a coordinator named as Cluster Head (CH) and

multiple SUs as cluster members. In a cluster,


cluster members transmit the local sensing data
to the CH, and CH is responsible for making a
cluster-level decision and forwarding it to the
FC. At FC, final decision will be made based
on cluster-level decisions received from different CHs. Hence, the cluster-based cooperative
spectrum sensing strategy can reduce the system overhead in terms of energy consumption,
time delay and bandwidth occupation, as well
as improve the detection performance [5, 14].
Many researchers have been studying the
cluster-based cooperative sensing issue. In Ref.
[15], a cooperative spectrum sensing scheme
based on clustering and 2-bit soften hard combination was proposed to exploit the received
information by SUs sufficiently. To reduce system overhead, a spectrum-heterogeneity-based
hierarchical spectrum sharing network was proposed [16], in which SUs near FC convey sensing result to it directly while those far away
from FC form clusters according to a distributed clustering scheme. In Ref. [17], the sensing time saving cluster-based algorithm was
proposed, in which the SNR difference of SUs
was taken into account and the sensing time
was shortened by reducing the sample size.
Meanwhile, in Ref. [18], clustering and soften
hard combination method were combined to
achieve a good tradeoff between performance
improvement and overhead saving. In Ref. [19],
a cluster-based optimal selective cooperative
spectrum sensing scheme was proposed for reducing reporting time and bandwidth while maintaining a certain level of sensing performance.
Another consideration of the cluster-based cooperative sensing scheme is to improve sensing performance when the reporting channel
suffers from a severe fading [20].
However, few studies paid attention to the
robustness of cluster-based cooperative spectrum sensing scheme in CRNs. In many clustering methods, SUs with large location correlation are grouped into the same cluster to
reduce the energy consumption for transmitting data to CH [16]. However, it is quite possible for multiple SUs within a cluster to suffer from multipath fading, shadowing or the
January
China
Communications
2014
China
Communications
January
2014

disturbance from attackers simultaneously. Hence, CH may make a wrong cluster-level decision about the status of PU and forward it.
And then, the FC may make an incorrect final
decision due to the hard combination rule used
in the second level of hierarchically cooperative sensing. With this motivation, we investigate the strategy to eliminate the harmful effect caused by SUs abnormal behaviour on
the final decision at FC in hierarchically cooperative spectrum sensing in this paper.
Based on the assumption that multiple SUs
within a cluster may become abnormal because
of fading, shadowing or PU Emulation Attack
(PUEA) [21], we propose a reputation-based
hierarchically cooperative spectrum sensing
scheme to improve the robustness of the CR
system. In our proposed hierarchically cooperative scheme, there are two levels of cooperation. In the first level of cooperation within
each cluster, the CH collects all of the sensing
data from SUs in the same cluster, combines
the Equal
EGC rule
makes a clusterthem with the
Gainand
Combination
(EGC)
level
decision
about
the
status
of
the
PU.
In
rule and makes a clusterlevel decision about
secondoflevel
of cooperation
at alevel
higher
lathe status
the PU.
In the second
of coyer
consisting
of
CHs
and
FC,
FC
collects
all
operation at a higher layer consisting of CHs
of theFC,
cluster-level
one-bit
decisions
from CHs,
and
FC collects
all of
the cluster-level
combines
them
with
a
modified
MAJORITY
one-bit decisions from CHs, combines them
rule and
makes aMAJORITY
final decision
the sta-a
with
a modified
ruleabout
and makes
tus
of
the
PU.
final decision about the status of the PU.
Moreover, in our proposed scheme, the reputation mechanism is integrated into the data
fusion process with MOJARITY rule at FC to
improve the robustness of the CRN. Hence, a
slide-window-based reputation degree calculation method is presented. That is, the reliability of different clusters is evaluated using their
reputation degrees which reflect the history of
their behaviour in the cooperation and can be
used to calculate the importance factors of
cluster-level decisions in the fusion process.
Then, the FC differentiates each clusters contribution to the final decision to improve the
robustness. In addition, since clusters are
formed based on the location correlation coefficients of SUs before spectrum sensing process, we also present a clustering method based
location correlation coefficient between SUs.
January 2014
2014
China Communications January

Fig.1 CRN model for hierarchically cooperative spectrum sensing

Finally, we evaluate the proposed scheme in


different scenarios. Simulation results show that
our proposed scheme can improve the robustness and the detection accuracy compared with
some exiting cooperative sensing schemes.
The remainder of the paper is organised as
follows. Section II introduces system model
and the problem we aim to resolve. Section III
presents the reputation-based hierarchically
cooperative spectrum sensing scheme in CRNs.
A location-aware clustering method and a reputation degree calculation method are also presented. Simulation results and analysis are given in Section IV. Finally, we conclude the
paper in Section V.

II. SYSTEM MODEL AND


PROBLEM FORMULATION
We consider a CRN consisting of 1 PU, 1 FC
located in the centre of a CRN, and K cooperative SUs located far away from the PU. It is
assumed that the PU signal received by SUs is
corrupted by the Additive White Gaussian Noise (AWGN) with zero-mean and variance
2 . In the spectrum sensing process, energy
detection mechanism, which has been widely
adopted because of its low complexity and
needing no prior knowledge about the PU, is
used as the fundamental brick.
Figure 1 shows the CRN model for hierarchically cooperative spectrum sensing. All SUs
are grouped to M different clusters by a clustering

27
14

where ni j and hi s j denote the noise and


received PU signal in the j-th sample at the
i-th SU, respectively, s j is the transmitted
PU signal which is distorted by the channel
gain hi from the PU to the i-th SU. It is assumed that hi keeps constant during LS . H 0
and H1 are the hypothesis of the absence of
PU and the presence of PU, respectively.
Each SU accumulates N samples within
LS as its local sensing data. That is, local
sensing data at the i-th SU, X i , is

Fig.2 Frame structure for hierarchically cooperative spectrum sensing

theirspatial
spatial
correlamethod
accordingtototheir
method according
correlation.
SU
each
cluster
selected
A SUAn
each
cluster
is selected
as CH
action.
AinSU
inineach
cluster
isis selected
as CH
according
certain
rules.
During
the
cording to to
some
certain
rules.
During
the coac-cording
to some
some
certain
rules.
operation, each
SU
cooperation,
each
SU
senses
and
transmits the
co-operation,
each
SUsenses
sensesand
andtransmits
sensing data to its CH, and CH is responsible
for gathering sensing data from SUs within a
cluster, combining and making a cluster-level
decision which is sent to the FC. Finally, the
FC performs a data fusion process of received
cluster-level decisions
decisions from
from CHs
CHs and makes
cluster-level
decision.
the final decision.
In this work, we did not consider the noise
and interference in two kinds of control channel, the control channel between SUs and CH
and the control channel between CHs and FC.
Figure 2 shows the TDMA frame structure.
The total frame is fixed to T consisting of
spectrum sensing time, s , and data transmission time, t T s . s consists of Local
Sensing (LS) time, LS , SUs reporting and
cluster-level decision time, intra , and CHs
reporting and final decision time, inter . When
the FC infers that PU is absent in spectrum
sensing process, FC allocates data transmission time, t , to SUs according to a certain
resource allocation rule. In this paper, we only
focus on the spectrum sensing process during
s . For each SU, LS is fixed.
During each local sensing interval, LS , SU
samples N times to perform energy detection
of PU signal. The j-th sample at the i-th SU,
xi j , is
n j
H0
xi j i
hi s j ni j H1
i 1, 2, , K , j 1, 2, , N

28
15

(1)

Xi
xi j , i 1, 2, , K
2

(2)

j 1

During SUs reporting and cluster-level decision time, intra , each SU transmits its sensing data to its CH through the control channel
which is assumed ideal.
CH fuses all received sensing data from
CH fuses all received sensing data from
SUs in its cluster to obtain the combination
SUs in its cluster to obtain the combination
result. With Equal Gain Combination (EGC)
result. With EGC rule, the combination result
rule, the combination result of the m-th cluster,
m-th cluster, Xc,m, is
of the, is
X c,m
X c ,m

1
Km

Km

X
i 1

i,m

, m 1, 2, , M

(3)

where X i .m and K m are the sensing data of the


i-th SU in cluster m and the number of SUs in
cluster m, respectively, and

K
m 1

K.

CH in cluster m makes the cluster-level decision between H 0 and H 1 by comparing X c , m


with the threshold of cluster m, m ,
H1

X c .m m , m 1, 2, , M
H0

(4)

where H 0 and H 1 denote the final decision


made by the FC that the presence of the primary signal and the absence of the primary
signal, respectively. Referring to the Neyman-Pearson formulation, m can be derived
according to the given false alarm probability.
Then, a one-bit cluster-level decision, Ym ,
is obtained as
0, X c , m m
, m 1, 2, , M
Ym
1, X c , m m

(5)

During CHs reporting and final decision time,


January
China
Communications
2014
China
Communications
January
2014

inter , Ym is transmitted to the FC by the CH in


cluster m via the control channel between CH
and FC which is also assumed ideal.
The FC combines these one-bit cluster-level decisions to perform the final decision
according to the following logic rule
M
H 0
U Ym
m 1
,H 1

(6)

where U is the sum of all cluster-level decisions. Eq. (6) demonstrates that the FC infers
the primary signal being transmitted, H 1 ,
when there are at least out of M clusters inferring H 1 ; otherwise, it infers the primary
signal not being transmitted, H 0 . The OR
rule corresponds to the case of 1 , the
AND rule corresponds to the case of M ,
and the MAJORITY rule corresponds to
M 2.
However, conventional hierarchically cooperative spectrum sensing scheme discussed
above fails to take the robustness into consideration. In a real network, shadowing and/or
PUEA will happen in a relatively large area. It
is possible that multiple SUs belonging to a
cluster are located at the shadowing area, as
shown in Figure 1. Moreover, if PUEA is
launched, most SUs within a cluster are disturbed due to the location correlation. In such
two situations, the corresponding CH(s) maybe make a wrong cluster-level decision about
whether the PU is present or not. Hence, false
cluster-level decision transmitted to FC will
introduce serious impact on the final decision.
The AND rule and OR rule become invalid in
the scenarios with shadowing and PUEA, respectively, while the MAJORITY rule remains
a low detection performance.
Therefore, the main purpose of this work is
to present a robust hierarchically cooperative
spectrum sensing scheme to deal with the impact
of shadowing or PUEA on the final decision.

III. REPUTATION-BASED
HIERARCHICALLY COOPERATIVE
SPECTRUM SENSING SCHEME
To resolve the problem mentioned in Section
January 2014
2014
China Communications January

II, we propose a reputation-based hierarchically cooperative spectrum sensing scheme in


CRNs. Our proposed scheme consists of two
levels of cooperation. The first level is the cooperation implemented in each cluster, which is
a cooperative sensing with SUs sensing data
combination. The second level is a cooperative sensing with cluster-level decisions combination performed at FC. Moreover, we also
present a location-aware clustering method, a
modified MOJORITY rule and a reputation
degree calculation method in this section.

3.1 Location-aware clustering


Before spectrum sensing, all SUs are divided
into M clusters by a proper clustering method
and a CH is selected for each cluster. Selecting
the number of clusters, M, is out of the scope
of this paper, and we assume that M is known
before clustering.
Here, we assume that SUs are clustered according to their correlation coefficients in a way
that SUs with highly correlated received signals will be grouped into the same cluster.
Spatially close SUs are likely to be affected by
the same environmental conditions, and/or by
the same PUEA attack, and they can suffer
from common biases in spectrum sensing.
Furthermore, PU is located at a place far away
from all SUs. Choosing the most correlated
SUs for each cluster is equivalent to clustering
SUs based on their location correlation between each other. With this assumption, we
present a location-aware clustering method to
divided SUs into clusters.
As we know, the core issue of a locationaware clustering method is to define a spatial
correlation coefficient between two SUs,
which captures the correlation characteristics
of SUs. Hence, we set up a location correlation matrix, , based on the distance and the
environmental impact factor as
1,1

i ,1


K ,1

1,i

i ,i


K ,i

1, K


i , K


K , K

29
16

where i , j denotes the spatial correlation coefficient between SUs i and j,


i , j exp di , j

[22], i 1, 2,, K , j 1,2,, K , and i ,i 1 .


d i , j is the distance between SUs i and j.
is the environmental impact factor. From Ref.
[22], 0.12 in the city environment, while
0.006 in the suburbs.
Based on the location correlation matrix, ,
we propose a location-aware clustering method. Since the number of clusters, M, is known, we only consider grouping K SUs to M
clusters based on location correlation matrix,
finding corresponding CHs in M clusters, and
calculating the number of SUs in each cluster,
K m ( m 1, 2, , M ). Let

K SU

1, 2, ,
i i
K denote the set of SUs, CH denote the
set of CHs and CH represent the initial
set of CHs, Cm ( m 1, 2, , M ) denote the
set of SUs in the m-th cluster and Cm
( m 1, 2, , M ) denote the initial set of SUs
in the m-th cluster, Cm denote the number
of SUs in Cm , K denote the set of SUs still
not clustered and K = K for initialization. D
is the radius of the coverage area of a CRN.
The main ideas of location-aware clustering
method are: First, according to the number of
clusters, M, FC sets the expected positions of
the CHs of M clusters [20], and selects M
temporal CHs which are located at the nearest
M corresponding expected positions. Second,
the FC groups other (K-M) SUs into M clusters
according to the maximum location correlation coefficient. Third, the CH which is the SU
having the largest sum of location correlation
coefficient with other SUs in the same cluster
is selected for each cluster. And the number of
SUs in each cluster is also computed.
Our proposed location-aware clustering method is described as Algorithm 1 as follows.
Algorithm 1 Location-aware clustering method
Step 1: the FC collects the location information of all
SUs and its own with the assistant of Global Positioning System.
Step 2: the FC calculates the location correlation
matrix, .
Step 3: the FC selects M CHs from SUs temporally.
1: for (m=1 to M, m++) do
2: Set the expected position of the CH in the m-th

30
17

cluster,

3:

2m D
2m
D
CH m : cos
, sin

M 2
M
2
Finds an SU that satisfied
i* arg min d CH
SU iK

4:

,i

i m , CH
CH SUi ,
*

C
Cm SUi , K
K SU i
m
*

5: end for
Step 4: the FC groups other (K-M) SUs into M clusters.
6: for (n=1 to K-M, n++) do
7:
The FC takes an SU from K , SU j K .
8:
Finds a CH that satisfied
i* arg max i , j
i

SUiCH

C
C i SU j , K
K SU j .
i

9: end for
Step 5: the FC selects the CH in each cluster.
10: The FC initialises the set of CHs, CH
=
.
11: for (m=1 to M, m++) do
12:
Finds an SU in each cluster that satisfied
i* arg max i , j
*

SUiCm

SU jCm

K m Cm , CH
CH SUi .
*

13: end for


Step 6: the FC broadcasts the clustering results to the SUs.

3.2 The first level of


cooperative spectrum sensing
3.2.1 Local sensing at SUs

During a local sensing interval, LS , each SU


samples the target spectrum band to implement energy detection.
Each SUs sensing data is obtained by Eqs.
(1-2). According to Eqs. (1-2), X i i2 at the
i-th SU follows a central chi-square distribution with a freedom degree of N under H 0 ,
while it follows a non-central chi-square with
a freedom degree of N and a non-centrality
parameter of i under H1 . That is,
2
H0
X i N
~
, i 1, 2, , K
(7)
2
2
i N i H1
where i N i , i denotes the average SNR
N

at the i-th SU and i hi 2 s 2 j N i2 .


j 1

According to the central limit theorem [23],


when N is large enough, X i follows asymptotical Gaussian distribution as

N i2 , 2 N i4 , H 0

Xi ~
(8)
2
4
( N i ) i , 2 N 2i i , H 1
January
China
Communications
2014
China
Communications
January
2014

inter , Ym is transmitted to the FC by the CH

3.2.2 Fusion process at CHs

During SUs reporting and cluster-level decision time, intra , each SU transmits its sensing
data to its CH.
The combination result of the m-th CH,
X c , m , is obtained with Eq. (3). According to
Eqs. (3) and (8), it is easy to conclude that
X c , m also follows the normal distribution as

X cc, m

N Km 2 2 N Km 4
i , H0

i , K 2
m i 1

K m i 1

Km

1
~
( N i ) i2 ,

K m i 1
, H
2 Km
1
4
2 N 2i i

K m i 1

(9)

m 1, 2, , M

According to Eqs. (3) and (4), the false alarm probability of EGC-based cooperative spectrum sensing scheme, Pf , m , in the m-th cluster is
Km

K m m N i2

i 1
Pf , m Q
Km

2 N i4

i 1

, m 1, 2, , M (10)

In CRNs, it is assumed that the spectrum


resource licensed to PU can be used by SUs
only when the PU is absent. Hence, NeymanPearson formulation is taken into consideration here. That is, the detection problem is formulated as to maximise the detection probability with a given false alarm probability.
With the target false alarm probability, Pf , m ,
the decision threshold in the m-th cluster, m ,
is determined as
1 1
Q Pf , m
m
K m

2 N N
(11)
i 1
i 1

Km

4
i

Km

2
i

m 1, 2,, M

According to Eqs. (4-5, 9-11), CH in the m-th


cluster makes a cluster-based decision between
H 0 and H 1 , which is denoted by a one-bit
variable, Ym .

in cluster m via the control channel between


CH and FC.
We assign different weight coefficients to
different clusters according to their reputation
degree to differentiate their contribution to the
final decision made at FC, where the clusters
reputation degree is a parameter denoting the
reliability of the cluster. The clusters with high
reputation degrees get large weight coefficients and play more important roles in the final
fusion process. By contrast, the clusters with low
reputation degrees are assigned small weight
coefficients to decrease their interference.
The FC receives all of the one-bit clusterlevel decisions, Ym ; m 1, 2, , M , and combines them with a fusion function, , as
Z w1 , w2 , , wM , Y1 , Y2 , , YM , where Z is
the fusion result at FC, wm is the weight coefficient of the m-th cluster. For the m-th cluster, wm
is computed using its reputation degree, rm , as
rm

, m 1, 2, , M
wm
(12)
M
rm
m 1

FC makes the final decision between H 0 and


H 1 according to the decision rule as follows
H

(13)

where is the final decision threshold.


Here, a modified MAJORITY rule is presented and used in the second level of cooperative spectrum sensing as
Z w1 , w2 , , wM , Y1 , Y2 , , YM
M

wmYm

M
M l 1
if Ym M

m 1
wm 1 Ym
m

1.1 if Ym M
m 1

(14)

Then, the global threshold is set as 1.


Hence, the global detection probability, Qd ,
and the global false alarm probability, Qf , are
Qd P Z H 1 , Qf P Z H 0 (15)

3.3 The second level of


cooperative spectrum sensing

3.4 Clusters reputation degree


calculation

During CHs reporting and final decision time,

Considering that different clusters have different

January 2014
2014
China Communications January

31
18

of H 0 and H 1 , respectively. Pf ,m and Pd,m

LT

are the reliability of the m-th clusters past decisions in case of H 0 and H 1 , respectively.
To estimate the reliability of each cluster
based on past detection performance, we present a slide-window-based reputation degree
calculation method using L past cluster-level
and final decisions. Slide window structure is
shown in Figure 3.
Owing to the cooperation gain of cooperative spectrum sensing scheme, the final decision is more accurate than cluster-level decisions. Hence, the final decision is used as a
reference and the cluster-level decision in a
certain window slot to be correct if it is agreement with the final decision made at this
sensing interval.
For the m-th cluster, xm, y is the number of

...
for the mth clusterlevel decisions

Ym(n-L+1)

...

Ym(n-1)

Ym(n)

for final decisions

O(n-L+1)

...

O(n-1)

O(n)

Fig.3 Slide window structure

times in the window size of length L for which


the cluster-level decision is H x when the
final decision is H y , and x, y 0,1 .
Hence,
m
0,0

l n L 1
n

m
0,1

l n L 1

1,0

1,1m

Fig.4 The result of location-aware clustering method

reliability based on their past detection performance, it is more efficient to take into account
the reliability of cluster-level decisions made
by different clusters. In our proposed reputationbased hierarchically cooperative spectrum sensing scheme in Sections 3.2 and 3.3, the combining weight coefficients assigned for each
cluster-level decision reflects the reliability of
this clusters past decisions. It is evident that
the cluster with higher detection probability
and lower false alarm probability has a higher
reliability, and therefore should contribute more
to the final decision made by FC. Hence, the
reliability of each cluster can be defined as
rm p0 1 Pf , m p1 Pd, m

(16)

where p0 and p1 are the prior probability

32
19

l n L 1
n

l n L 1

1 Ym l 1 O l
1 Ym l O l

Ym l 1 O l

(17)

Ym l O l

where Ym l and O l are the one-bit decisions made by the m-th CH and FC, respectively, at the l-th interval of the observation
window L.
We do not assume any knowledge of the
probability density of the CHs decisions, p0
and p1 . In our proposed reputation degree calculation method, we approximate p0 , p1 , cluster-level detection probability and false alarm
probability by relative frequencies. That is,
1 m
m
p 0 0,0
1,0
, p1 L1 0,1m 1,1m
L

Pd, m

m
1,1m
0,0
,
P
f ,m
m
m
m
1,1m
1,0
0,1
0,0

(18)
(19)

Eq. (16) is also rewritten as

rm p 0 1 Pf , m p 1 Pd, m

(20)

January
China
Communications
2014
China
Communications
January
2014

With the approximate value in Eqs. (18) and


(19), the reliability of the m-th CH is calculated with Eq. (20) and used to compute the
weight coefficient of the m-th cluster, wm , using Eq. (12).

IV. SIMULATION RESULTS AND


ANALYSIS
First, we give the simulation result about our
proposed location-aware clustering method. The
simulation parameters are set as follows. The
coverage area of the CRN is 1 400 m 1 400 m.
K=40 and M=6. FC is located at the centre of
the coverage area, and its position is (0, 0). SUs
are uniformly distributed in the coverage area.
For the city environmental condition, 0.12 .
Figure 4 shows the result of our proposed
location-aware clustering method. From Figure 4,
we observe that 40 SUs are grouped into 6
clusters, and the number of 6 clusters is 7, 7, 7,
6, 6, and 7, respectively. Hence, when SUs in
the coverage area are uniformly distributed, the
number of SUs in each cluster is nearly the same.
Second, the performance of our proposed
reputation-based hierarchically cooperative spectrum sensing scheme is evaluated. The traditional EGC-based cooperative spectrum sensing scheme aimed at improving the detection
probability, the conventional cluster-based hierarchically cooperative spectrum sensing scheme aimed at decreasing the communication
overhead, and our proposed reputation-based
hierarchically cooperative sensing scheme taking both robustness and energy efficiency into
consideration are denoted by EGC-based cooperative sensing scheme, conventional hierarchical scheme and reputation based hierarchical scheme, respectively. According to the
different hard fusion strategy used at FC, we
denote conventional hierarchical scheme as
conventional hierarchical scheme: MAJORITY,
conventional hierarchical scheme: AND and
conventional hierarchical scheme: OR, respectively. Simulations are performed in a
CRN with K=20 SUs grouped into M=4 clusters according to their location correlation. We
January 2014
2014
China Communications January

Fig.5 The performance comparison of four different schemes, where all SUs in 1
cluster are suffering from deep shadowing

assume that each cluster contains 5 SUs with


similar average SNRs. In each sensing interval,
an SU samples N=20 times to obtain local sensing data. For each SU, i2 1 , and i=1,
2, , K . The reputation degree of each cluster
is calculated using cluster-level decisions within L=20 sensing intervals. Moreover, each result is obtained through 100 000 simulations.
Figure 5 shows the performance comparison
of four different spectrum sensing schemes, where

33
20

Fig.6 The comparison of the detection probability with different number of shadowing SUs

Figure 5 (a) and 5 (b) demonstrate the robustness and the energy efficiency of four sensing
schemes with multiple deep shadowing SUs,
respectively. In simulations, the local SNRs of
20 SUs grouped into 4 clusters are {(-3.5,
-3.8, -3.3, -3.5, -3.4), (-4.8, -4.5, -4.5, -4.7,
-4.4), (-5.5, -5.4, -5.7, -5.6, -5.4), (-9.5,
-9.6, -9.4, -9.5, -9.6)} in dB.
Figure 5 (a) shows the detection probability
of four different schemes, where 1 cluster of
SUs is suffering from deep shadowing. From
Figure 5 (a), we observe that EGC-based cooperative scheme obviously has the best performance because the soft data fusion strategy
takes full advantage of all the local sensing
data. The cluster-based network structure of
the other three hierarchically cooperative sensing schemes makes them lose some amount of
information which is useful to the decisionmaking at FC. The conventional hierarchically
cooperative scheme with AND rule has the
worst performance in the CRN containing shadowing SUs because it cannot tolerate any
H 0 decision from clusters. The conventional
hierarchically cooperative scheme with MAJORITY rule obtains the second worst performance since it only puts emphasis on the ene-

34
21

rgy saving and fails to deal with the impact of


the shadowing SUs. However, our reputationbased hierarchically cooperative scheme takes
robustness issue into account on the basis of
high energy efficiency.
Figure 5 (b) shows the communication overhead of four different schemes, where 1 cluster of SUs is suffering from deep shadowing.
From Figure 5 (b), we observe that the communication overhead of our proposed reputation-based hierarchically cooperative scheme
and the conventional hierarchically cooperative
schemes is almost the same, and is much smaller than that of EGC-based cooperative scheme.
According to Figure 5 (a) and 5 (b), we conclude that our proposed reputation-based hierarchically cooperative scheme outperforms the
conventional hierarchically cooperative schemes in deep shadowing circumstance, as well
as keeps a low communication overhead. That
is, the proposed cooperative scheme achieves
a good tradeoff between the system robustness
and energy saving.
Figure 6 depicts the detection performance
comparison with different number of deep-shadowing SUs in order to demonstrate our proposed hierarchically cooperative scheme can
outperform conventional hierarchically cooperative scheme with MAJORITY rule. For no
shadowing SUs case, SNRs at SUs are {(-3.5,
-3.8, -3.3, -3.5, -3.4), (-4.8, -4.5, -4.5, -4.7,
-4.4), (-5.5, -5.4, -5.7, -5.6, -5.4), (-4.5, -4.6,
-4.4, -4.5, -4.6)} in dB. For 2 shadowing SUs
case, SNRs at SUs are {(-3.5, -3.8, -3.3, -3.5,
-3.4), (-4.8, -4.5, -4.5, -4.7, -4.4), (-5.5, -5.4,
-5.7, -5.6, -5.4), (-9.5, -9.6, -4.4, -4.5, -4.6)}.
For 4 shadowing SUs case, SNRs at SUs are
{(-3.5, -3.8, -3.3, -3.5, -3.4), (-4.8, -4.5, -4.5,
-4.7, -4.4), (-5.5, -5.4, -5.7, -5.6, -5.4), (-9.5,
-9.6, -9.4, -9.5, -4.6)} in dB.
From Figure 6, we observe that, when no
shadowing SU exists, the detection performance of our proposed reputation-based hierarchically cooperative scheme is very close to
that of conventional hierarchically cooperative
scheme with MAJORITY rule. However, when
there are 2 shadowing SUs in the network, the
January
China
Communications
2014
China
Communications
January
2014

detection performance of our proposed reputation-based hierarchically cooperative scheme


is better than that of conventional hierarchically cooperative scheme with MAJORITY rule.
And when the number of shadowing SUs is 4,
the detection performance of our proposed reputation-based hierarchically cooperative scheme has little degradation compared to the case
with 2 shadowing SUs, while the detection
performance of conventional hierarchically cooperative scheme with MAJORITY rule degrades rapidly.
Referring to IEEE 802.22 [24], a cooperative spectrum sensing model makes sense only
if its detection probability can achieve 0.9
with the false alarm probability of 0.1. From
Figure 6, we observe that, when the number of
shadowing SUs reaches 2 or 4, our reputation-based hierarchically cooperative scheme
still satisfies the requirement of IEEE 802.22,
but the conventional hierarchically cooperative scheme with MAJORITY rule does not.
Hence, with the introduction of reputation mechanism to modify the MAJORITY rule, our reputation-based hierarchically cooperative scheme can deal with the interference of shadowing
SUs and improve the robustness effectively.
Figure 7 shows the detection performance
of four different schemes with the interference
of PUEA. In this simulation, we suppose that
there are 4 SUs within a cluster suffering from
the PUEA and inclined to send false local sensing data to their CH. SNRs of 20 SUs are
{(-9.5, -9.6, -9.4, -9.5, -9.6), (-9.3, -9.2, -9.4,
-9.3, -9.1), (-9.9, -9.7, -9.8, -9.6, -9.8), (-4.5,
-4.6, -4.4, -4.5, -9.3)} in dB, and 4 SUs in
the 4th cluster are attacked by PUEA.
From Figure 7, we observe that conventional hierarchically cooperative scheme with
OR rule has the worst performance. The achieved false alarm probability of our proposed
reputation-based hierarchically cooperative scheme and conventional hierarchically cooperative scheme with MAJORITY rule is smaller
than that of EGC-based cooperative scheme.
This is because the cluster-based network structure of hierarchically cooperative spectrum
January 2014
2014
China Communications January

Fig.7 The performance comparison of four different schemes, where four SUs in 1
cluster are suffering from PUEA

Fig.8 The performance comparison under different average SNRs

sensing helps to eliminate the bad influence of


false sensing data caused by PUEA. Moreover,
the performance of our proposed reputationbased hierarchically cooperative scheme is
much better than that of conventional hierarchically cooperative scheme with MAJORITY
rule, which demonstrates that the reputation
mechanism introduced in our scheme is an
effective way to remove the effect of PUEA.
Figure 8 shows the missed detection probability of three different schemes under diffe-

35
22

rent average SNRs to analyse the detection performance in various environments. The target
global false alarm probability is set as
0.15; m 1, 2, , M .

f ,m

From Figure 8, we observe that EGC-based


cooperative scheme has the lowest missed detection probability. The performance of our
proposed reputation-based hierarchically cooperative scheme is better than that of conventional hierarchically cooperative scheme with
MAJORITY rule. To satisfy Qd 0.9 defined
in IEEE 802.22, average SNR thresholds for
EGC-based cooperative scheme, our proposed
reputation-based hierarchically cooperative scheme and conventional hierarchically cooperative scheme with MAJORITY rule are about
-7.5 dB, -5 dB and -4.5 dB, respectively.

V. CONCLUSION
Considering both the robustness and energy
saving, we proposed a reputation-based hierarchically cooperative spectrum sensing scheme for CRNs in this paper. In our proposed
hierarchically cooperative scheme, there were
two levels of cooperation. In the first level of
cooperation taking place within a cluster, the
CH collects all of the sensing data from SUs
in the same cluster, combines them with the
EGC rule and makes a cluster-level decision
about the status of the PU. In the second level
of cooperation happening at a higher layer
consisting of CHs and FC, FC collects all the
cluster-level one-bit decisions from CHs, combines them with a modified MAJORITY rule
and makes a final decision about the status of
the PU. In addition, the reputation mechanism
was introduced in the second level of cooperation. That is, in modified MAJORITY rule, FC
assigned weight coefficients to the clusterlevel decisions according to their reputation
degrees of clusters.
Simulation results show that, if there are
deep-shadowing SUs located near each other,
our proposed reputation-based hierarchically
cooperative scheme can improve the detection
performance significantly, as well as keep a

36
23

low communication overhead compared with


conventional hierarchically cooperative schemes.
Meanwhile, if there are SUs suffering from
PUEA, the performance of our proposed reputation-based hierarchically cooperative scheme is better than that of EGC-based cooperative scheme and conventional hierarchically
cooperative schemes.
Therefore, the introduction of reputation mechanism and hierarchically cooperative strategy
helps to improve the system robustness against
deep shadowing and interference from PUEA.
In the future work, we will explore other performance metrics besides the detection probability, the false alarm probability, and the communication overhead. Furthermore, other promising ways toward a robust cluster-based solution, such as cooperative spectrum sensing based on self-organization paradigms surveyed
in Ref. [25], will be investigated.

ACKNOWLEDGEMENT
This work was partially supported by the National Natural Science Foundation of China
under Grant No. 61071127; and the Science
and Technology Department of Zhejiang Province under Grants No. 2012C01036-1, No.
2011R10035.

References
[1]

HAYKIN S. Cognitive Radio: Brain-Empowered


Wireless Communications[J]. IEEE Journal on
Selected Areas in Communications, 2005, 23(2):
201-220.

[2]

MITOLA J. Cognitive Radio: An Integrated


Agent Architecture for Software Defined Radio[D]. Stockholm, Sweden: KTH Royal Institute of Technology, 2000.

[3]

GHASEMI A. SOUSA E S. Collaborative Spectrum Sensing for Opportunistic Access in Fading Environments[C]// Proceedings of the First
IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks:
November 8-11, 2005. Baltimore, MD, USA.
IEEE Press, 2005: 131-136.

[4]

ZHAO Qing, SADLER B M. A Survey of Dynamic Spectrum Access[J]. IEEE Signal Processing Magazine, 2007, 24(3): 79-89.

[5]

YUCEK T, ARSLAN H. A Survey of Spectrum


January
China
Communications
2014
China
Communications
January
2014

Sensing Algorithms for Cognitive Radio Appli-

on Wireless Communications, Networking and

cations[J]. IEEE Communications Surveys and

Information Security: June 25-27, 2010. Bei-

Tutorials, 2009, 11(1): 116-130.


[6]

[7]

HAYKIN S, THOMSON D, REED J. Spectrum

jing, China. IEEE Press, 2010: 183-187.


[16]

Sensing in Cognitive Radio[J]. Proceedings of

lication of Clustering Structure in the Hierar-

the IEEE, 2009, 97(5): 849-877.

chical Spectrum Sharing Network Based on

CHEN Huifang, ZHOU Ming, XIE Lei, et al. Fa-

Cognitive Radio[C]// Proceedings of 3rd Int-

ult-Tolerant Cooperative Spectrum Sensing Sch-

ernational Conference on Cognitive Radio Ori-

eme for Cognitive Radio Networks[J]. Wireless

ented Wireless Networks and Communications:

Personal Communications, 2013, 71(4): 2379-

May 15-17, 2008. Singapore. IEEE Press, 2008:

2397.
[8]

[9]

HAO Jianjun, LI Jin. Determination of Thresh-

1-5.
[17]

sing Time Saving Cluster-Based Cooperative

sor Networks[J]. China Communications, 2011,

Spectrum Sensing Scheme[C]// Proceedings

8(1): 14-19.

of 12th International Conference on Commu-

YANG Jinsheng, SHAO Xinming. Optimal Num-

nication Technology: November 11-14, 2010.

ative Spectrum Sensing[C]// Proceedings of

Nanjing, China. IEEE Press, 2010: 1244-1247.


[18]

tion for Cooperative Spectrum Sensing in Cog-

nce and Wireless Technology Conference: July

nitive Radio Networks[J]. IEEE Transactions on

26-30, 2011. Harbin, China. 2011: 902-905.

Wireless Communications, 2008, 7(11): 4502-

MISHRA S M, SAHAI A, BROADERSEN R W.

4507.
[19]

Cooperative Spectrum Sensing Scheme in Cog-

Conference on Communications: June 11-15,

nitive Radio[J]. EURASIP Journal on Wireless

2006. Istanbul, Turkey. IEEE Press, 2006: 1658-

Communications and Networking, 2013: 176.


[20]

ter-Based Cooperative Spectrum Sensing over

Correlation-Aware User Selection for Coop-

Correlated Log-Normal Channel with Noise

erative Spectrum Sensing in Cognitive Radio

Uncertainty in Cognitive Radio Networks[J].

Areas in Communications, 2012, 30(2): 297-306.

IET Communications, 2012, 6: 2725-2733.


[21]

worthy Spectrum Sensing in Cognitive Radio

et al. Robust Cooperative Spectrum Sensing

Networks[C]// Proceedings of 1st IEEE Work-

Schemes for Fading Channels in Cognitive

shop on Networking Technologies for Software

Radio Networks[J]. Science China Information

Defined Radio Networks: September 25, 2006.

XIA Wenfang, YUAN Wei, CHEN Wenqing, et

Reston, VA, USA. IEEE Press, 2006: 110-119.


[22]

al. Optimization of Cooperative Spectrum SenProceedings of 2010 International Conference

GUDMUNDSON M. Correlation Model for Shadow Fading in Mobile Radio Systems[J]. Elec-

sing in Ad-Hoc Cognitive Radio Networks[C]//

tronics Letters, 1991, 27(23): 2145-2146.


[23]

GNEDENKO B V, KOLMOGOROV A N. Limit

on Global Telecommunications: December 6-10,

Distributions for Sums of Independent Ran-

2010. Miami, FL, USA. IEEE Press, 2010: 1-5.

dom Variables[J]. American Journal Mathematics, 1954, 105: 28-35.

NI Xiong, CHEN Huifang, XIE Lei, et al. Reputation-Based Hierarchically Cooperative Spe-

[15]

CHEN Ruiliang, PARK J-M. Ensuring Trust-

YUE Wenjing, ZHENG Baoyu, MENG Qingmin,

Sciences, 2011, 54(2): 348-359.

[14]

REISI N, AHMADIAN M, JAMALI V, et al. Clus-

CACCIAPUOTI A S, AKYILDIZ I F, PAURA L.

Ad Hoc Networks[J]. IEEE Journal on Selected

[13]

NHAN N-T, KOO I. A Cluster-Based Selective

[C]// Proceedings of 2006 IEEE International

1663.

[12]

MA Jun, LI Ye. Soft Combination and Detec-

2011 Cross Strait Quad-Regional Radio Scie-

Cooperative Sensing among Cognitive Radios

[11]

WEI Dong, FENG Chunyan, GUO Caili. A Sen-

old for Energy Detection in Cognitive Radio Sen-

ber of Secondary Users in Weighted Cooper-

[10]

GONG Lei, CHEN Jie, TANG Wanbin, et al. App-

[24]

CORDEIRO C, CHALLAPALI K S, BIRRU D, et al.

ctrum Sensing Scheme in Cognitive Radio Net-

IEEE 802.22: The First Worldwide Wireless Sta-

works[C]// Proceedings of 2nd IEEE/CIC Inter-

ndard Based on Cognitive Radios[C]// Proc-

national Conference on Communications in

eedings of 1st IEEE International Symposium

China: August 12-14, 2013. Xian, China. IEEE

on New Frontiers in Dynamic Spectrum Acc-

Press, 2013: 1-6.

ess Networks: November 8-11, 2005. Balti-

DUAN Jiaqi, LI Yong. A Novel Cooperative

more, MD, USA. IEEE Press, 2005: 328-337.

Spectrum Sensing Scheme Based on Cluster-

[25]

ZHANG Zhongshan, LONG Keping, WANG Jian-

ing and Softened Hard Combination[C]// Proc-

ping. Self-Organization Paradigms and Optim-

eedings of 2010 IEEE International Conference

ization Approaches for Cognitive Radio Tech-

January 2014
2014
China Communications January

37
24

nologies: A Survey[J]. IEEE Wireless Commu-

Information Science and Electronic Engineering,

nications, 2013, 20(2): 36-42.

Zhejiang University, China. He received his B.S. deg-

Biographies

ree in electronic engineering, M.S. and Ph.D. degrees


in communications and information systems from

CHEN Huifang, Associate Professor with the Dep-

Zhejiang University in 1994, 1997 and 2002, respec-

artment of Information Science and Electronic Engi-

tively. His current research interests include informa-

neering, Zhejiang University, China. She received her

tion theory and coding, coded modulation system,

Ph.D. degree at the Department of Information Sci-

and security issue in wireless networks. Email: xiel

ence and Electronic Engineering, Zhejiang University

@zju.edu.cn

in 2000. From October 2005 to September 2007, she


was a post-doctoral research fellow with the Dep-

NI Xiong, graduate student in the Department of

artment of Computer Science, Shizuoka University,

Information Science and Electronic Engineering, Zhe-

Japan. Her current research interests include infor-

jiang University, China. He received his B.S. degree in

mation theoretic security and its applications, and

information and communication engineering from

wireless networks. Email: chenhf@zju.edu.cn

Zhejiang University in 2010. His current research


interest is spectrum sensing in cognitive radio net-

XIE Lei, Associate Professor with the Department of

38
25

works. Email: 21031117@zju.edu.cn

January
China
Communications
2014
China
Communications
January
2014

You might also like