Beruflich Dokumente
Kultur Dokumente
I. I NTRODUCTION
HE last decade has witnessed the increasing popularity
of wireless services. Based on fixed spectrum allocation
methodology, in many countries, most of the available radio
spectrum has been assigned for various services. On the
other hand, careful studies of the spectrum usage pattern
have revealed that the allocated spectrum experiences low
utilization. In fact, recent measurements by Federal Communications Commission (FCC) have shown that 70% of the
Manuscript received October 25, 2006; revised August 18, 2007; accepted
January 2, 2008. The associate editor coordinating the review of this paper
and approving it for publication was W. Yu. This paper was presented in part
at the IEEE International Conference on Communications (ICC), Glasgow,
UK, June 2007.
The authors are with the Institute for Infocomm Research, A STAR,
21 Heng Mui Keng Terrace, Singapore 119613 (e-mail: {ycliang, yhzeng,
athoang}@i2r.a-star.edu.sg.
Digital Object Identifier 10.1109/TWC.2008.060869.
c 2008 IEEE
1536-1276/08$25.00
Fig. 1. A deployment scenario for IEEE 802.22 WRAN: the primary users
are TV receivers and wireless microphones.
1327
(1)
(2)
1328
N
1
|y(n)|2 .
N n=1
(3)
f
,
(5)
s
u2
where Q() is the complementary distribution function of the
standard Gaussian, i.e.,
2
1
t
Q(x) =
exp
dt.
(6)
2
2 x
Under hypothesis H1 , denote p1 (x) as the PDF of the
test static T (y). For a chosen threshold , the probability of
detection is given by
p1 (x)dx.
(7)
Pd (, ) = P r(T (y) > |H1 ) =
if s(n) and u(n) are both circularly symmetric and complexvalued, and
12 =
1
E|s(n)|4 + E|u(n)|4 (s2 u2 )2 + 2s2 u2 , (9)
N
fs
Pd (, ) = Q
1
.
(10)
u2
2 + 1
For a target probability of detection, Pd , the detection threshold can be determined by
fs
= Q1 (Pd ).
1
(11)
u2
2 + 1
From (5), on the other hand, this threshold is related to the
probability of false alarm as follows:
1
Q (Pf ) =
1
fs .
(12)
u2
Therefore, for a target probability of detection Pd , the probability of false alarm is related to the target detection probability
as follows:
2 + 1Q1 (Pd ) + fs .
(13)
Pf = Q
On the other hand, for a target probability of false alarm, Pf ,
the probability of detection is given by
1
Pd = Q
(14)
(Q1 (Pf ) fs ) .
2 + 1
Finally, for a given pair of target probabilities (Pd , Pf ), the
number of required samples to achieve these targets can be
determined from (11) and (12) by cancelling out the threshold
variable . The minimum number of samples is given by
2
1
(15)
Nmin = 2 Q1 (Pf ) Q1 (Pd ) 2 + 1 .
1329
respectively. If we define
R0 (, ) =
T
C0 (1 Pf (, ))P (H0 ),
T
(16)
R1 (, ) =
T
C1 (1 Pd (, ))P (H1 ),
T
(17)
and
Fig. 2. Frame structure for cognitive radio networks with periodic spectrum
sensing ( : sensing slot duration; T : data transmission slot duration.)
(18)
R( ) = R0 (, ) + R1 (, )
(19)
Pd (, ) Pd ,
(20)
s.t.
s.t.
) = R0 (, )
R(
(21)
Pd (, ) Pd .
(22)
1330
1
where = 2 + 1Q (Pd ). Thus, from (23), we can see
that the achievable throughput of the secondary network is a
function of the sensing time .
Theorem 1: Under the assumptions (AS1) - (AS3), if the
primary signal is complex-valued PSK and the noise process
is CSCG, for any target probability of detection, there exists
an optimal sensing time which yields the maximum achievable
throughput for the secondary networks.
Proof: It can be verified from (23) that
( )
fs (1 T )
R
( + fs )2
=
exp
C0 P (H0 )
2
2 2
1
1
(24)
+ Q( + fs ).
T
T
Obviously,
1
1
<
lim R ( )
( ) < 0,
lim R
(27)
(28)
( ) 1 C1 P (H1 )(1 Pd )
lim R
T
+.
(29)
time for each mini-slot, and again T the duration for each
frame. We fix the the total sensing time in each frame to
= M 1 , and the number of samples for each mini-slot is
N1 = N/M (without loss of generality, we assume that N1
is an integer.)
Let us consider the following hypotheses for the ith minislot:
H1 : yi (n) = hi si (n) + ui (n),
(30)
H0 : yi (n) = ui (n),
(31)
The channel coefficients hi s are zero-mean, unitvariance complex Gaussian random variables;
The noises are independent of each other for each of the
M mini-slots;
The signal power and noise power are constant over the
M mini-slots, i.e., E[|si (n)|2 ] = s2 , and E[|ui (n)|2 ] =
u2 for all i;
The primary user is either active or inactive for all the
M mini-slots.
Data Fusion: Process the measurements from all minislots jointly and then make the final decision based on
the calculated statistics;
Decision Fusion: Process each mini-slots data separately,
and make individual decisions. The final decision is then
made by fusing the individual decisions.
N1
1
|yi (n)|2 ,
N1 n=1
(32)
Using data fusion, the test statistic used for final decision is
then represented as
T (y) =
M
gi Ti (y),
(33)
i=1
with mean 1 = u2
12
M
2
i=1 gi (|hi |
+ 1) and variance
M
1 2
g [|hi |4 E|s(n)|4 + E|u(n)|2
N1 i=1 i
(|hi |2 s2 u2 )2 ],
M
1 2
g [|hi |4 E|s(n)|4 + E|u(n)|2
N1 i=1 i
(|hi |2 s2
u2 )2
2|hi |2 s2 u2 ],
u4
,
N1
M
u4
2
2
2
gi |hi | .
1 + 2
1 =
N1
i=1
02 =
(37)
(38)
Pd = Q (f2 (g1 , , gM ; Pf )) ,
(39)
where
f1 (g1 , , gM ; Pd ) = 1 Q
(Pd ) +
N1
M
gi |hi | , (40)
f2 (g1 , , gM ; Pf )
M
1
1
2
=
Q (Pf ) N1
(41)
gi |hi | ,
1
i=1
M
where 1 = 1 + 2 i=1 gi2 |hi |2 . Since Q(x) is a decreasing function of x, from the primary users interest, for a target
probability of false alarm Pf , we want to design the optimal
gi by maximizing the probability of detection, which gives:
min
M
i=1
gi2 =1
f2 (g1 , , gM ; Pf ).
(42)
max
g1 , ,gM :
M
i=1
gi2 =1
f1 (g1 , , gM ; Pd ).
(43)
N1
M
gi |hi |2 , (44)
i=1
f2 (g1 , , gM ; Pf ) Q1 (Pf )
N1
M
i=1
gi |hi |2 . (45)
M
i=1
M
gi2 =1
gi |hi |2 .
(47)
i=1
Pf = Q 2 Q1 (Pd ) +
Pd = Q
i=1
g1 , ,gM :
g1 , ,gM :
(35)
(36)
(46)
|hi |2
gi =
.
M
4
|h
|
i
i=1
(34)
1331
1
2
M
N1
|hi |2
M i=1
Q1 (Pf )
M
N1
|hi |2
M i=1
(48)
(49)
M
2
where 2 = 1 + 2
i=1 |hi | .
M
Finally, let us consider a special case where hi = 1 for all
i, and gi is chosen as gi = 1M . From (38), for a target Pd ,
the probability of false alarm is given by:
(50)
2 + 1Q1 (Pd ) + M N1 .
Pf = Q
Since M N1 = N , the use of M mini-slots for the case of
static channel does not provide any performance gain when
data fusing is applied.
From (38), for a fading channel environment, in order to
achieve same probability of detection for each frame, the
detection threshold will change from one frame to another,
thus the probability of false alarm will vary. Let us denote, for
a given sensing duration , Pf ( ) as the average probability
of false alarm over all frames. We then define the normalized
achievable throughput as
B( ) =
T
(1 Pf ( )).
T
(51)
B. Decision Fusion
(i)
(i)
0
N1
(i)
2
,
(52)
|hi | 1
Pd = Q
u2
2|hi |2 + 1
1332
and the probability of false alarm for the ith slot is given by
0
(i)
Pf = Q
1
N1 .
(53)
u2
Once the decision is made for each time slot, there are different
rules available for making final decision on the presence of
the primary user.
1) Optimal Decision Fusion Rule: Let Ii be the binary
decision from the ith time slot, where Ii {0, 1} for i =
1, , M . The optimal decision fusion rule is the ChairVarshney fusion rule [16], which is a threshold test of the
following statistic:
M
(i)
(i)
Pd
1 Pd
0 =
Ii log (i) + (1 Ii ) log
(i)
Pf
1 Pf
i=1
P (H1 )
.
(54)
+ log
P (H0 )
If 0 0, then the primary user is present; otherwise, there
is no primary user.
2) Logic-OR (LO) Rule: LO rule is a simple decision
rule described as follows: if one of the decisions says that
there is a primary user, then the final decision declares
M that
there is a primary user. Mathematically, define = i=1 Ii ,
if 1, then the primary user is present; otherwise, there is
no primary user.
Assuming that all decisions are independent, the probability
of detection and probability of false alarm of the final decision
are, respectively,
Pd = 1
Pf = 1
M
(i)
(1 Pd ),
i=1
M
(i)
(1 Pf ).
Pf =
i=1
M
(i)
Pd ,
(i)
Pf .
Pd =
j=0
M
M +j
M M
2 j
Pd,02
, (59)
(1 Pd,0 )
M
2
+j
M M
2
M
M +j
M M
2 j
Pf =
Pf,02
, (60)
(1 Pf,0 )
M
2
+j
j=0
c!
.
respectively, where kc = k!(ck)!
A. Data Fusion
Suppose the channel coefficients from the primary user to
each receiver are known. Using maximal ratio combining,
M
i=1
(56)
(57)
i=1
4) Majority Rule: Another decision rule is based on majority of the individual decisions. If half of the decisions or
more say that there is a primary user, then the final decision
declares
Mthat there is a primary user. Mathematically, define
= i=1 Ii , if M
2
, where x
denotes the smallest
integer not less then x, then the primary user is declared to
be present; otherwise, there is no primary user. Assuming
(1)
that all decisions are independent, and supposing that Pd =
(M )
(1)
(M )
Pd
= Pd,0 and Pf = Pf
= Pf,0 , the probability
hi
yi (n).
M
2
i=1 |hi |
y(n) = u(n),
(63)
(64)
(65)
1
2
Pf = Q 3 Q (Pd ) + N
|hi | ,
(66)
(58)
(61)
(62)
y(n) =
M M
2
(55)
i=1
Pd =
i=1
M
where 3 =
2 i=1 |hi |2 + 1. On the other hand, for
a target probability of false alarm, Pf , the probability of
detection is given by
M
1
1
2
Pd = Q
(Q (Pf ) N
|hi | ) .
(67)
3
i=1
When the channel coefficients are unknown, similar to (32)
and (33), a simple way for data fusion is to choose gi = 1M .
In this case, we obtain:
M
N
1
2
|hi | ,
(68)
Pf = Q 4 Q (Pd ) +
M i=1
1333
10
5.4
5.3
10
10
5.2
5.1
4.9
4.8
3
10
0.8
1.2
1.4
1.6
1.8
2.2
2.4
Number of samples
2.6
4
x 10
4.7
1.5
2.5
3.5
4.5
(69)
B. Decision Fusion
Suppose that Pd,i and Pf,i are the probability of detection
and probability of false alarm from the ith secondary user,
respectively. Similar to the multiple mini-slot case, the probability of detection and probability of false alarm of the final
decision can be derived accordingly, from which the sensingthroughput tradeoff can be carried out.
0.97
0.96
M
N
1
1
2
Pd = Q
|hi |
Q (Pf )
,
4
M i=1
M
2
where 4 = 1 + 2
i=1 |hi | .
M
0.95
0.94
0.93
0.92
0.91
0.9
0.89
Simulated
Theory
0.88
1.5
2.5
3.5
4.5
0.5
0.45
0.4
Pf of the network
0.35
0.3
0.25
0.2
0.15
0.95
0.9
Normalized throughput
1334
0.85
0.8
0.75
M=1
M=2
M=3
M=4
M=5
0.7
0.1
0.65
0.05
0
1.5
2.5
3
3.5
Sensing time (ms)
4.5
Fig. 6.
The average probability of false alarm Pf ( ) using multi-slot
spectrum sensing.
4
5
Sensing time (ms)
0.95
0.9
Normalized throughput
0.85
0.8
0.75
M=1
M=2
M=3
M=4
M=5
0.7
0.65
4
5
Sensing time (ms)
1335
0.975
2.4
0.95
2.2
LogicAND
LogicOR
LogicMajority
0.9
0.85
1.8
1.6
1.4
1.2
1
0.8
0.8
channel known
channel unknown
0.75
5
6
7
Number of minislots
0.6
10
0.4
2
10
12
14
16
18
20
Number of users
Fig. 9.
Fig. 10. Optimal sensing time for distributed spectrum sensing with various
decision rules: SN Rp = 15dB, T = 100ms.
1
E[(|s(n)|2 + |u(n)|2 + s(n)u (n)
N
+ s (n)u(n) s2 u2 )2 ].
(72)
2
N
1
= E
(s(n) + u(n))2 (s2 + u2 ) (74)
N n=1
=
1
E[(s2 (n) + u2 (n) + 2s(n)u(n)
N
s2 u2 )2 ].
(75)
Eq. (9) can then be derived by using the zero-mean and mutual
independence property of s(n) and u(n).
A PPENDIX B: C ONCAVITY OF R( )
Proposition 5: Under the assumptions (AS1) - (AS3), if the
primary signal is complex-valued PSK and the noise process
is CSCG, then Pf ( ) is decreasing and convex for the range
of in which Pf ( ) 0.5.
dPf ( )
d
fs 1/2
exp(( + fs )2 /2).(77)
=
2 2
=
1336
R EFERENCES
0.995
0.99
0.985
0.98
0.975
LogicAND
LogicOR
LogicMajority
0.97
2
10
12
14
16
18
20
Number of users
16
14
12
10
1.5
2.5
3.5
Number of users
Fig. 12. Optimal sensing time for distributed spectrum sensing using logicAND decision fusion rule.
(78)
then
( ) = C0 P (H0 ) Pf ( ) 1 Pf ( ) 1 . (79)
R
T
T
T
When Pf ( ) is decreasing and convex in , Pf ( ) is negative and increasing in . Further, Pf ( ) is decreasing in .
( ) is decreasing in
Therefore, from (79), it follows that R
) is concave in .
, 0 T , which further implies R(
From Propositions 5 and 6, it follows that R( ) is concave
for the range of in which Pf ( ) 0.5. This further implies
that there is a unique maximum point of R( ) within this
range.
Ying-Chang Liang (SM00) received the PhD degree in Electrical Engineering in 1993. He is now
a Senior Scientist in the Institute for Infocomm Research (I2R), Singapore, where he has been leading
research activities in the area of cognitive radio
and cooperative communications and standardization activities in IEEE 802.22 wireless regional
networks (WRAN), for which his team has made
fundamental contributions in physical layer, MAC
layer, and spectrum sensing solutions. He also holds
adjunct associate professorship positions in Nanyang
Technological University (NTU) and the National University of Singapore
(NUS), both in Singapore, and an adjunct professorship position at the
University of Electronic Science and Technology, China (UESTC). He has
been teaching graduate courses in NUS since 2004. From Dec. 2002 to Dec.
2003, Dr Liang was a visiting scholar with the Department of Electrical
Engineering, Stanford University. His research interests include cognitive radio, dynamic spectrum access, reconfigurable signal processing for broadband
communications, space-time wireless communications, wireless networking,
information theory and statistical signal processing.
Dr Liang served as an Associate Editor for the IEEE T RANSACTIONS
ON W IRELESS C OMMUNICATIONS from 2002 to 2005, Lead Guest-Editor of
the IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS , S PECIAL
I SSUE ON C OGNITIVE R ADIO : T HEORY AND A PPLICATIONS, and GuestEditor of the Computer Networks Journal (Elsevier) Special Issue on Cognitive Wireless Networks. He received the Best Paper Awards from IEEE
VTC-Fall1999 and IEEE PIMRC2005, and the 2007 Institute of Engineers
Singapore (IES) Prestigious Engineering Achievement Award. Dr Liang has
served for various IEEE conferences as a technical program committee (TPC)
member. He was Publication Chair of the 2001 IEEE Workshop on Statistical
Signal Processing, TPC Co-Chair of the 2006 IEEE International Conference
on Communication Systems (ICCS2006), Panel Co-Chair of the 2008 IEEE
Vehicular Technology Conference Spring (VTC2008-Spring), TPC Co-Chair
of the 3rd International Conference on Cognitive Radio Oriented Wireless
Networks and Communications (CrownCom2008), Deputy Chair of the 2008
IEEE Symposium on New Frontiers in Dynamic Spectrum Access Networks
(DySPAN2008), and Co-Chair, Thematic Program on Random matrix theory
and its applications in statistics and wireless communications, Institute for
Mathematical Sciences, National University of Singapore, 2006. Dr Liang is
a Senior Member of IEEE. He holds six granted patents and more than 15
filed patents.
1337