IEEE ICC 2015 - Workshop on Cognitive Radios and Networks for Spectrum Coexistence of Satellite and Terrestrial Systems

(CogRaN-Sat)

A Trust-value based Cooperative Spectrum Sensing

Algorithm for Mobile Secondary Users
Xinyu Wang, Min Jia, Qing Guo, and Xuemai Gu
Communication Research Center
Harbin Institute of Technology
Harbin, China
Email: wangxinyu920124@sina.com, jiamin@hit.edu.cn, qguo@hit.edu.cn, guxuemai@hit.edu.cn

Abstract—Cognitive radio is able to effectively increase In a cognitive radio network (CRN), final decisions made
spectral utilization. However, cooperative spectrum sensing gives by the fusion center (FC) may be affected by the fake results
malicious users chances to interfere with its decision processes. If uploaded by MUs. In light of this, conventional trust-value
a mobile secondary user’s energy detection results become based cooperative spectrum sensing algorithms for static CRNs
unreliable, conventional trust-value based cooperative spectrum (CTBSS) are proposed in [2-4], where an SU’s trust value (TrV)
sensing algorithms, which are used to resist malicious attacks, depends on how well its locally detected results match most
cannot distinguish whether it’s caused by a reliable user moving detected results. SUs with larger TrVs are assigned with larger
into a deep-fading area or it attacking maliciously. This is the weighting coefficients (WeCs). However, those conventional
main reason why the detection performances of conventional
trust-value based cooperative spectrum sensing algorithms for
algorithms are terribly bad when secondary users are mobile.
This paper proposes a trust-value based cooperative spectrum
static CRNs tend to overly punish reliable users (RUs) in the
sensing algorithm aiming at mobile secondary users. We divide cases of deep fading or large path losses and sometimes they
the whole region into cells according to different areas’ actual even remove RUs mistakenly which will cause the detection
channel conditions so that the detected results of users in any one performances greatly decrease. And every SU has the
of cells are very close to each other but those in different cells are opportunity to move into areas in deep fading. That’s the main
quite different. Our proposed approach removes malicious users reason why those conventional trust-value based cooperative
independently in each cell based upon their trust values. And sensing algorithms for static CRNs can not be applied to
larger weighting coefficients are given to cells with better channel mobile scenarios directly. Studies on spectrum sensing aiming
conditions. Then this paper analyzes the effects of the average at mobile scenarios, namely mobile SUs, are rather limited.
velocity of secondary users on the detection performance. The algorithm is targeting mobile SUs proposed in [5] which
Simulation results show that when secondary users are mobile, must detect many times to achieve the requested accuracy, thus
the detection performance of our algorithm is much better than it will cause significant interference to the primary user (PU).
that of conventional trust-value based cooperative spectrum A trusted collaborative spectrum sensing for mobile CRNs
sensing algorithms proposed for static secondary users and is based on location reliability and malicious intention (LRMI) is
better than that of a trusted collaborative spectrum sensing for proposed in [6] aiming at malicious attacks when SUs are
mobile cognitive radio networks based upon location reliability mobile. However, the computed user trust evaluations based
and malicious intention which also takes location channel upon Dempster-Shafer theory are not accurate enough to
differences into considerations.
correctly remove MUs according to the simulation results.
Keywords—cognitive radio; cooperative spectrum sensing; To overcome these problems, this paper firstly proposes a
mobility; malicious users; trust value trust-value based cooperative spectrum sensing algorithm for
mobile secondary users (TBSS-MU). Then, we analyze the
I. INTRODUCTION effects of the average velocity of SUs on its detection
performance. Finally, this paper compares the proposed trust-
Cognitive radio (CR) [1] has been widely used to increase
value based cooperative spectrum sensing algorithm, TBSS-
the utilization of spectrum. The basic idea is that secondary
MU, with CTBSS to illustrate the important effects of
users (SUs) access frequency bands of authorized or
considering location channel differences on the detection
unauthorized users by finding suitable opportunities to increase
performance when SUs are mobile. Moreover, we compare the
spectrum utilization, providing the authorized or unauthorized
proposed TBSS-MU algorithm to the LRMI algorithm which
users are not affected. Thus, effective spectrum sensing is
both take location channel differences into considerations and
essential to CR. Spectrum sensing consists of single-node
are both proposed aiming at mobile scenarios, to highlight our
sensing and multi-node cooperative sensing. But the detection
proposed algorithm’s superior detection performance in mobile
performance of the former is largely affected by the so-called
scenarios. The numerically simulated results are presented to
hidden terminal problem, channel fading effects and multi-path
verify that the primary user detection performance (PUDP) and
effects. Thus, studies on the latter are more significant than the
the malicious user detection performance (MUDP) of the
former. Data fusion is one of the most attractive fields in
proposed TBSS-MU algorithm are both significantly better
cooperative sensing, which however is prone to attacks by
than those of CTBSS and LRMI.
malicious users (MUs).

978-1-4673-6305-1/15/$31.00 ©2015 IEEE 1635

 IEEE ICC 2015 - Workshop on Cognitive Radios and Networks for Spectrum Coexistence of Satellite and Terrestrial Systems
(CogRaN-Sat)

forwarded to the FC along with the numbers of the cells they

currently belong to. Assume that these energy values and cell
PU numbers are transmitted correctly. The FC decides whether the
PU is active or not. Assume that the noise signal power
FC received at each cell is the same, and the received PU signal
power is related with its channel conditions.
Our channel model considers both shadow effects and path
Mobile Users loss. The PU signal power received by a secondary user ui with
the distance of d i ,k from the PU [7] is
Fig. 1. The CRN model with a static PU and some mobile SUs.

P j (dBm)  Pt (dBm)  K (dB)

i ,k
II. SYSTEM MODEL   
 10 j log10[d i ,k d 0 ]  j dB(dB)
A. CRN Model
In real-world scenarios, there always exist some areas where K (dB)  20 log10 (4d 0 )  with  being the
with better channel conditions and others with worse in a wavelength of the PU signal [8]. Pt is the transmit power of
large-sized area of interest. Each mobile SU has the the PU. d 0 represents a reference distance for the antenna far
opportunity to move into areas in deep fading. Conventional field.  j is referred to as the path loss exponent of cell c j .  j dB
algorithms fail to consider individually SUs with different indicates the shadow fading of c j . k denotes the kth detection.
channel conditions so they regard RUs in deep fading as MUs
mistakenly making their detection performances decrease.
Thus conventional trust-value based algorithms can no B. Detection Model
longer apply to mobile CRNs. Our algorithm solves this Each SU adopts energy detection since it’s easy to realize
problem by dividing the whole region into cells to update SUs’ and needs no a prior information. H1( k ) and H0(k ) represent the
TrVs and remove MUs independently in each cell based upon hypotheses on the presence or absence of the PU signal, during
those TrVs. The whole large-sized area is divided into equal- the kth detection, respectively. Assume that the bandwidth of
j
sized cells. Then as to SUs moving into the same cell, they the PU signal is W and the sample interval is T.  i,k is referred
are close geographically and thus close path loss. to as the instantaneous signal-to-noise ratio experienced at ui
Meanwhile, their surroundings are similar and thus shadow in c j during the kth detection and f  ij,k ( x ) stands for the
fading and multi-path effects are alike. So compared with j
probability distribution function of  i,k . Qm is a marqum-
SUs in different cells, SUs in the same cell have more
similar channel conditions and closer detected results. Then function and  is a gamma function. P1 and P0 represent the
even if an SU move into a cell with bad channel conditions, probabilities of the presence or absence of the PU signal,
j
only if that SU is reliable, its detected results would still be respectively. When Yi,k  2 , where 2 is the preset threshold of
j
consistent with other detected results of SUs in that specific the FC and Yi ,k is the detected energy value of ui in c j during
cell. In fact, MUs are in the minority of SUs. Since RUs have the kth detection, the FC considers the PU is active; otherwise,
similar detected results in any cell, only a few SUs’ results, i.e. idle. Thus, the detection probability Pd , the false alarm
MUs’ fake results, are different from most results in that cell. probability Pf and the error probability Pe [9] are as follows
So our algorithm can identify MUs independently in each cell
by decreasing TrVs of SUs with detected results different from
most detected results of the cell they currently belonging to to
increase the detection performance. Then when SUs are in cells
 Pd  P(Yi ,jk  2 | H1( k ) )  Q
x
m( 2 i j,k , 2 ) f j ( x)dx 
i ,k

in deep fading, their detected results are indeed small but still
close to each other. Thus our algorithm can avoid the
phenomenon of conventional algorithms that remove RUs in
deep fading mistakenly since only those randomly transmitted
 
Pf  P Yi,jk  2 | H 0( k )  (TW , 2 / 2) (TW ) 
fake results of MUs have large differences from most detected
results in the same cell. This is the reason why we divide the Pe  P0  Pf  P1  (1  Pd ) 
whole area into cells. As the signal attenuation is generally
stable in a small area, it is not necessary to have tiny size cells
and it is reasonable we consider a CRN as illustrated in Fig. 1, Cooperative sensing can reduce the limitation of single-
which consists of a static PU, an FC and N mobile SUs node sensing mentioned above. We choose soft-decision
including M MUs. Assume that the entire region is divided into cooperative sensing due to its better detection performance [10].
L equal-sized square cells, with each cell assigned a unique cell
number through positioning technology. Here we divide the C. Malicious Attack Model
whole region as shown in Fig. 1 as an example, but actually our We adopt the strong malicious attack model same as [6] to
algorithm can apply when it is divided into cells in any size or contrast the detection performance of our algorithm with that
with any number. Detected energy values by the SUs are of LRMI reasonably: if the PU is idle, the MUs send the

1636
 IEEE ICC 2015 - Workshop on Cognitive Radios and Networks for Spectrum Coexistence of Satellite and Terrestrial Systems
(CogRaN-Sat)

information of (Yi ,jk  ) to increase Pf ; otherwise, send   k (ui )   k (ui )

(Yi ,jk  ) to decrease Pd . And the MUs send cell numbers 4  tan(  max( (u ))  3 ) ,  k (ui )  max( 2 )
 k (ui )   k i (6)
randomly. Malicious attack deviation  is chosen randomly. 1  (u )
,  k (ui )  max( k i )
 2
III. ALGORITHM DESCRIPTIONS
Then update ri according to ri  ri   k (ui ) and adjust it
Before our algorithm begins, first we initialize the within the restraint of the preset upper limit rmax and lower
following parameters: ri  0 , 0 (c j )  1 , e0 (c j )  0 and limit rmin of ri . Consider users with TrVs smaller than 1 to be
E0 (c j )  0 where k ( c j ) stands for the weighting coefficient malicious and remove their detected results in the FC.
of c j computed from the kth detection; ek (c j ) stands for the
As for computing the cells’ WeCs, first update the above-
number of accumulated RUs in c j computed after the kth mentioned ek (c j ) and E k (c j ) according to (7) and (8) based on
detection; Ek (c j ) is referred to as the accumulated average the detected results sent during the k th detection. Then
value of the RUs’ detected results in c j after the kth detection. calculate the weighting coefficient of c j ,  k (c j ) , according to
Then each SU independently performs energy detection and (9) using updated E k (c j ) computed from (8). So when the PU
sends its detected result to the FC along with the cell number of
is active, our algorithm will assign the cells with larger energy
the cell it currently belonging to. Now all preparatory work is
values, i.e., better channel conditions, with larger WeCs to
completed. Our algorithm consists of two main parts, namely
increase the objective function and thus Pd . When the PU is
calculating SUs’ TrVs and then removing MUs, and computing
idle, since the received noise signal power of each cell is
cells’ WeCs. The TrV updating method must meet the
approximately equal, the computed WeCs are approximately
following criteria: if Yi ,jk is close to most detected energy values equal and thus Pf would hardly change. Then we increase Pd
of the SUs in the same cell as ui , we consider Yi ,jk to be reliable on the condition that Pf is constant, i.e., improving the PUDP.
and increase ui ’s TrV ( ri ). Otherwise, we decrease ri . Define The MUDP is referred to as the ability of distinguishing
the distance-function k (ui ) to quantify the difference between between MUs and RUs. Better PUDP means higher Pd when
Yi ,jk and the above-mentioned most detected energy values: Pf is constant. The MUDP is related to TBSS-MU’s first part,
while the PUDP is collectively decided by the first and second
parts. Only properly distinguishing between MUs and RUs, can
 
 k (ui )  Yi ,jk  avg Yi ,jk
N'
i 1
 
std Yi ,jk
N'
i 1
(5) the proper weighting coefficients be calculated. Then the more
accurate weighted objective function will be computed to
improve the PUDP. Finally, the FC calculates the final
where avg and std mean computing the average value and the weighted objective function according to (10) to decide
standard deviation, respectively, and N ' is the number of SUs whether the PU is active or not.
in the same cell that ui belongs to during this detection. In fact,
if we just regard the numerator of (5) as the distance-function,
we can still ensure the SUs with detected results more different ek (c j )  ek -1 (c j )  e j (7)
from most results from the particular cell they currently belong
to have larger distance-functions compared to other SUs in this
very cell. However, since the cells have different locations and 

ej
thus different channel conditions, the detected results of SUs in Ek -1 (c j )  ek -1 (c j )  Yj
i 1 i , k (8)
different cells differ considerably, even by several order of E k (c j ) 
ek (c j )
magnitudes. For example, the difference between the SUs’
actual detected results in the cells with good channel conditions
is much larger than that in deep fading since the former results
have higher order of magnitudes and thus larger difference  
value. Then just the difference between locally detected results  k ( c j )  L   E k (c j )
  E k (c j ) 
 (9)
and the average value of each cell cannot reflect reasonably  j 
and accurately whether their detected results are reliable or not
among the entire region of interest. The standard deviation in
(5) is to make the numerator and the denominator have the   

N 
same order of magnitude, which makes the distance-functions  H1(k)   k (c j )  Yi ,jk   2
 i 1 
comparable in the whole region of interest. U  (10)
 

N 
 H 0(k)   k (c j )  Yi ,jk   2
Assume that k (ui ) is the trust value deviation of ui after   i 1 
the kth detection. If  k (ui ) is small, consider Yi ,jk to be reliable
and the computed k (ui ) should be positive to increase the where e j denotes the number of RUs belonging to c j during the
trust value, ri . If not, consider it to be unreliable and the 
computed k (ui ) should be negative to decrease ri . There are kth detection; Yi ,jk is the detected results of RUs in c j during
many functions meeting this requirement and we choose the 
the kth detection; Yi ,jk denotes all detected results of RUs and
following one:
N  is the number of all the RUs during the kth detection; U

1637
 IEEE ICC 2015 - Workshop on Cognitive Radios and Networks for Spectrum Coexistence of Satellite and Terrestrial Systems
(CogRaN-Sat)

denotes the final decision. Note that the premise of our 1

algorithm is that the ratio of M/N (  ) is small. Considering that
the WeCs of cells should not increase indefinitely to avoid
individual cells from playing absolute leading roles, we set the 0.8
sum of those coefficients as L.

Detection probability
Now analyze the influence of SUs’ average velocity on the 0.6
detection performance of TBSS-MU. During the finite time N=50 M=10 TBSS-MU
duration  t , the number of SUs ever existing in each cell ( N e ) N=50 M=15 TBSS-MU
consists of two parts, i.e., the initial value N L in the static 0.4 N=30 M=6 TBSS-MU
state; SUs moving into this particular cell caused by mobility. N=50 M=10 CTBSS
We consider the latter as follows. Assume ui ’s average N=50 M=15 CTBSS
N=30 M=6 CTBSS
velocity is Vui ( 0) so that it travels the distance of S ui in  t , 0.2
N=50 M=10 LRMI
where Sui   t Vui . The average number of cells where ui N=50 M=15 LRMI

i
 i

ever travels is Su K u   s , where  s is the cell size and K ui 0
0 0.2 0.4
N=30 M=6 LRMI
0.6 0.8 1
is the parameter related to ui ’s mobility model. Each SU False alarm probability
performs one detection every Ts sec so ui performs t Ts
detections totally, which means the FC receives t Ts results Fig. 2. Comparison of ROC curves of three algorithms.

 
from ui during  t . Thus, it sends  t Ts  Sui K ui   s  reliable (RUs) or fake results (MUs) to recover harmful effects
detected results to the FC in c j . Now let us analyze from the caused by c j . The discussion above demonstrates the positive
perspective of c j . All N SUs upload  t Ts   N results during effects of mobility of SUs on the MUDP. Thus, it helps
 t , so users in each cell send  t Ts   N L results on average. calculate more accurate WeCs and a properer objective
The expected number of detected results from each cell is a function to improve the PUDP.
stable value when the mobility models are given. In conclusion,
the second part of N e during  t can be expressed as IV. SIMULATION RESULTS
 t   Sui  N  t Vui Assume that the whole area is a 1000m 1000m square area,
  N L   t   . So N e 
 Ts   Ts K ui  s  L Kui  s whose center is 1000 m away from the PU. The noise signal
power is 5 dBm and the PU signal power is 200 mW. The SUs
N   t  Vui  take 1 ms for each detection [11] and they detect every second.
  1 . In short, as Vui becomes larger, N e gets Vui  40m/s . and TW=5. Choose  j and  j dB from 2 to 5 and 2
L  K ui   s 

larger, the times of TrV updating of each SU in one cell dB to 20 dB randomly, respectively.  ~ (10dBm,5dBm).
P0  P1  0.5. 1  10. L  9 .
becomes less and more SUs proceed TrV updating in one cell.
Then under the malicious attack model mentioned above, Fig. 2 compares the PUDP of TBSS-MU, LRMI and
the MUs transmit random cell numbers. So MUs’ malicious CTBSS algorithms in terms of receiver operating
attacks can be considered as uniformly distributed in all the characteristics (ROC) to prove TBSS-MU is effective (N=50,
cells. However, the cells have different locations and thus their M=10; N=50, M=15; N=30, M=6). By invoking computed trust
channel conditions are quite different so the detected results of values from the workspace of MATLAB and observing Fig. 2,
SUs have big or small differences. Then the same malicious find out that the performances of the MUDP and the PUDP of
attacks in different cells bring about varying degrees of attack TBSS-MU are much better than those of LRMI and CTBSS in
effects. Among all the cells, the ones with detected results mobile scenarios under malicious attacks. TBSS-MU’s better
more different from transmitted results of MUs are attacked detection performance than LRMI is because our algorithm can
more obviously. So in those cells, the computed average values remove the MUs more accurately according to the computed
and standard deviations of detected results disagree with actual TrVs. And CTBSS becomes non-effective in such cases
situation and they cannot reflect real conditions, which may because it does not take location channel differences into
reduce RUs’ TrVs and increase MUs’ TrVs mistakenly. considerations and it neglects the case that RUs can be in deep-
fading areas. Focus on the curves of TBSS-MU to find out
Assume that one of those cells is c j . Since SUs’ TrVs tend to
when N is constant (50), the larger M is, the worse the PUDP
be updated wrongly in c j , as Vui gets larger, the times of TrV will be. And when  is constant (20%), the larger N is, the
mistakenly updating of each SU in c j becomes less and more better the PUDP will be. In a CRN where N=50, the larger M is,
SUs proceed TrV updating in c j . Thus the negative effects of the more wrongly detected results will be sent to the FC during
c j are dispersed onto more SUs and each SU is less impacted. every detection. Then the difference between the calculated
averages or standard deviations and true values are too large,
We call this the dispersion effects. Moreover, as for the MUs, resulting in inaccurate k (ui ) and thus ri , so the FC cannot
larger Vui means the cells altering faster which suffer violent correctly increase RUs’ TrVs or decrease MUs’. Thus, the FC
attacks. This makes the dispersion effects more obvious. So as cannot remove MUs thoroughly and even remove individual
Vui increases, the mistaken updating increment of each user
RUs mistakenly. So the larger M is, the poorer the MUDP will
becomes smaller in c j during  t . Then soon after ui moves out be, and thus the computed WeCs and objective function are
of c j , ri will increase or decrease to its true value by providing no longer accurate making the PUDP poor. If  is constant, the

1638
 IEEE ICC 2015 - Workshop on Cognitive Radios and Networks for Spectrum Coexistence of Satellite and Terrestrial Systems
(CogRaN-Sat)

1
Detection probability 1

Detection probability
0.5 0m/s
0.5
M=5
0
0 20 40 60 80 100 0 20m/s
False alarm probability

Average velocity 0 0.2 0.4 0.6 0.8 1

0.2 The percentage of malicious users

False alarm probability

M=10 1
40m/s
0.1
0.5
M=15
0 60m/s
0 20 40 60 80 100
Average velocity 0
0 0.2 0.4 0.6 0.8 1
Error probability

1
M=20 The percentage of malicious users
1

Error probability
80m/s
0.5
0.5
0 100m/s
0 20 40 60 80 100
Average velocity 0
0 0.2 0.4 0.6 0.8 1
The percentage of malicious users
Fig. 3. The PUDP of TBSS-MU versus the average velocity of SUs.
Fig. 4. The PUDP of TBSS-MU versus the percentage of MUs.
larger N is, the bigger space diversity gains will be obtained
normally, so the detection performance will be better. As for
our algorithm, N getting smaller means N ' is smaller, and thus ACKNOWLEDGMENT
calculated averages and standard deviations can hardly reflect This work was supported by the National Natural Science
real conditions since the amount of statistical samples is not big Foundation of China under Grant No. 61201143, Innovation
enough causing the detection performance of TBSS-MU poor. Foundations for Shandong Province of Aerospace (Grant No.
In Fig. 3, we study the PUDP of TBSS-MU with different 2014JJ002) and the Fundamental Research Funds for the
average velocity Vui . Assume N=50 and M =5, 10, 15 and 20, Central Universities (Grant No. HIT. IBRSEM. 201309).
respectively. Fig. 4 studies the PUDP of TBSS-MU with
different percentages of MUs when Vui is 0m/s, 20m/s, 40m/s, REFERENCES
60m/s, 80m/s and 100m/s, respectively. By invoking computed [1] Mitola J., Maguire G. Q. Jr., “Cognitive Radio: making software radios
more personal”, IEEE Pers. Commun., vol. 6, no. 4, pp. 12-18, Aug.
trust values from the workspace of MATLAB and observing 1999.
Fig. 3, we find out that regardless of M, as Vui increases, the
[2] R. Chen, Park J.-M., and Bian K., “Robust distributed spectrum sensing
MUDP and the PUDP of TBSS-MU become better, namely in cognitive radio networks”, in Proc. IEEE Int. Conf. on Computer
larger Pd and smaller Pf and Pe . Thus, the mobility of SUs Commun. (INFOCOM), Phoenix, AZ, pp. 31-35, Apr. 2008.
plays a positive role on the detection performance of TBSS- [3] C. Sun, W. Zhang, and Khaled L., “Cooperative spectrum sensing for
MU, which is consistent with the derivation result above. Then cognitive radios under bandwidth constraints”, in Proc. IEEE Wireless
compare four situations with different M to find out if Vui is Commun. and Networking Conf. (WCNC), Kowloon, pp. 1-5, March
2007.
constant, larger M means worse PUDP which is consistent with
[4] Kaligineedi P., Khabbazian M., and Bhargava V. K., “Malicious user
simulation results of Fig. 2. By invoking computed trust values detection in a cognitive radio cooperative sensing system”, IEEE Trans.
from the workspace of MATLAB and observing Fig. 4, we Wireless Commun., vol. 9, no. 8, pp. 2488-2497, Aug. 2010.
find out that regardless of Vui , as  increases, the MUDP and [5] Arshad K., Moessner K., “Mobility driven energy detection based
the PUDP of TBSS-MU become gradually worse, namely Pd spectrum sensing framework of a cognitive radio”, in Proc. IEEE UK-
getting smaller and Pf and Pe getting larger and that when  is India-IDRC Int. Workshop on Cognitive Wireless Systems (UKIWCWS),
New Delhi, pp. 1-5, Dec. 2010.
constant, larger Vui means better MUDP and PUDP, which is [6] Jana S., K. Zeng, W. Cheng, and Mohapatra P., “Trusted collaborative
consistent with the observations in Figs. 2 and 3. spectrum sensing for mobile cognitive radio networks”, IEEE Trans. On
Information Forensics and Security, vol. 8, no. 9, pp. 1497-1507, Sept.
2013.
V. CONCLUSIONS
[7] Andrea G., “Wireless communications”, Stanford, USA.: Stanford Univ.
In this paper, we have proposed a trust-value based Press, pp. 35-44, 2003.
cooperative spectrum sensing algorithm for mobile SUs when [8] Goldsmith A. J., Varaiya P. P., “Capacity of fading channels with
there exist MUs in the CRN. The algorithm firstly divides the channel side information”, IEEE Trans. on Information Theory, vol. 43,
entire region into cells according to different areas’ actual no. 6, pp. 1986-1992, Nov. 1997.
channel conditions in order to take location channel differences [9] Digham F. F., Alouini M.-S., and Simon M. K., “On the energy
detection of unknown signals over fading channels”, IEEE Trans. on
into considerations, and then removes MUs independently in Commun., vol. 55, no. 1, pp. 21-24, Jan. 2007.
each cell based on their TrVs, and finally provides larger WeCs [10] A. Ghasemi, E. S. Sousa, “Opportunistic spectrum access in fading
for cells with better channel conditions. The numerically channels through collaborative sensing”, Journal of Commun., vol. 2, no.
simulated results have shown that the MUDP and PUDP of our 2, pp. 71-82, Mar. 2007.
algorithm are significantly better than those of CTBSS and [11] WRAN WG on Broadband Wireless Access Standards, IEEE 802.22
LRMI in mobile scenarios. [Online] Available: www.ieee802.org/22.

1639