Documente Academic
Documente Profesional
Documente Cultură
I. I NTRODUCTION
Due to the scarcity of the spectrum, the Federal Communications Commission (FCC) has approved a guideline based on
the cognitive radio concept to enhance the utilization of the
spectrum [13]. In general, secondary users (SUs) in a cognitive radio network are allowed to access spectrums occupied
by primary users (PUs) in a primary network given that 1)
those spectrums are vacant, or 2) interferences created by SUs
to PUs occupied those spectrums are under an acceptable level.
The first scenario has already been considered for the TV white
space usage, while the later has attracted considerable interest
in the literatures. In both cases, spectrum sensing is required in
order to check the presence of PUs. A miss detection can cause
considerable interference to PUs, while a false alarm reduces
the efficiency of spectrum usage. Therefore, highly reliable
spectrum sensing is a crucial task to ensure high throughput
and efficiency of spectrum usage for cognitive radio networks.
In the literatures, there are a few spectrum sensing methods
including matched filtering, energy detection, cyclostationarity
detection, and eigenvalue-based detection; see, e.g., [4, 5].
Each method has its own advantages and disadvantages. If all
information about the transmitted signal of the PU is known
a priori, matched filtering is the optimum detector. Since
it requires perfect knowledge of the PU signaling features
such as bandwidth, modulation type, pulse shaping, frame
format, etc., it may not be implementable in practice. For
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE DySPAN 2010 proceedings
x t
e
x t+
dt (1)
Rx ( ) = lim
t t t
2
2
2
x (f ) =
Sx (f )
,
Sx (f + 2 )Sx (f 2 )
(3)
(4)
Xi (t, f ) =
N
s 1
(5)
n=0
lim Sx (t, f )t .
f 0 t
H0 : ym (n) = wm (n)
H1 : ym (n) = hm s(n) + wm (n),
(6)
(7)
(8)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE DySPAN 2010 proceedings
(9)
(10)
(11)
H y(n).
z(n) = c
(18)
H
h
.
h
then we have
H0 : y(n) = w(n),
H1 : y(n) = hs(n) + w(n).
(12)
Since the detection probability is related to the signal-tonoise ratio (SNR) of the received signal, increasing the SNR
translates into better detection. If the SU has the channel state
information (CSI), h, under H1 , the maximum ratio combining
(MRC) can be used to enhance the reception of PUs signal.
Choosing the combining coefficients as
c=
h
,
h
(13)
(14)
Rs ( )
R
w ( ),
B. BMRC-SCF Detector
Prior to presenting details of the BMRC-SCF, we provide
the following important lemma to emphasize the motivation
of using the cyclic SCF.
Lemma 4.1: The cyclic SCF has built-in ability to handle
noise uncertainty.
2
2
= w
,
Proof: Let the estimated noise power be
w
where is called the noise uncertainty. We can envision this
as noise w(t)
(f
)|
=
|
(f
)|.
This
states
that
the
noise
uncertainty
by |
w
w
(16)
H1
max |
z (f )| H0 1
(19)
H1
avg |
z (f )| H0 2
(20)
f F
Algorithm 2 :
f F
(15)
=
This combining method is called the blind
where c
MRC (BMRC).
Next, we present the BMRC with cyclic SCF (BMRC-SCF)
detector, followed by the BMRC with modified cyclic SDF
(BMRC-MSDF) detector.
max |
z (f )|,
f F
(21)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE DySPAN 2010 proceedings
L
m 1
gm (l)um (n l),
(23)
l=0
M
(24)
Notice that
cm |2 = 1. For Case 1, w(n) follows
m=1 |
the same distribution as any um (n), denoted as u(n) for
simplicity, thus we have
D
z
u,
(25)
gm (l)um (n l),
(26)
m=1 l=0
(f ) =
Sw
V2 = |G|V1 .
(30)
(V )
cm wm (n).
m=1
M L
m 1
Gm f
S (f ),
Gm f +
2
2 um
m=1
M
w (f ) = G
um (f ),
(27)
gm (l), and
which is a
estimation.
is obtained
(28)
where we define
M
m=1 Gm f + 2 Gm f 2
G=
M
. (29)
M
2
2
m=1 Gm f + 2
m =1 Gm f 2
In the Appendix, it is proven that 0 < |G| 1. For white
noises, |G| = 1, while for colored noises, 0 < |G| < 1.
This implies that the false alarm probability based on colored
noise is less than that based on white noise for any given
zH z
.
N
(35)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE DySPAN 2010 proceedings
f F
(36)
Algorithm 4 :
max
f F
avg
M
m=1
M
f F m=1
w
m (n) =
L
m 1
gm (l)um (n l).
(39)
l=0
Employing samples of w
m (n) to calculate the cyclic SCF and
determining its magnitude, we obtain the following:
w
m (f )| = |
um (f )|.
(40)
The result in (40) states that the effect of the noise correlation
disappears. Hence, exploiting the cyclic SCF together with the
linear combiner is effective against correlated noise.
B. LC-MSDF Detector
Algorithm 3 :
H1
|
ym (f )| H0 4
(37)
H1
|
ym (f )| H0 5
(38)
Since no channel knowledge is required here, channel estimation issues are irrelevant. This could be advantageous
in practice, where a complicated channel estimation might
be required. Note that unlike thresholds for pre-combining
detectors considered previously, thresholds for post-combining
detectors are dependent on the number of antenna at the SU.
Similar to the BMRC-SCF detector, the LC-SCF detector
is also able to handle colored noise as proven in the lemma
below.
Lemma 5.1: The LC-SCF detector has built-in ability to
effectively handle both uncorrelated and correlated noise environments.
Proof: For the LC-SCF detector, we need to check its
effectiveness against correlated noise at one antenna only due
The LC-MSDF detector is proposed for complexity reduction as in the case of the BMRC-MSDF detector. The detection
of the PU is carried out as shown below:
avg
M
f F m=1
1
|Tym (f )| H
H0 6 ,
(41)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE DySPAN 2010 proceedings
TABLE I
T HE FALSE ALARM PROBABILITY FOR ALL DETECTORS IN CORRELATED
NOISE SCENARIO .
Perfect ( = 0)
Cyclic ( = 0)
Conv ( = 0)
Perfect ( = 0.5)
Cyclic ( = 0.5)
Conv ( = 0.5)
0.9
0.8
0.7
0.6
0.067
0.021
0.092
0.5
0.4
Pf a
0.3
LC-SCF (Algo. 3)
LC-SCF (Algo. 4)
LC-MSDF
0.093
0.01
0.064
0.2
0.1
0
20
15
10
0.9
SNR (dB)
Fig. 1.
0.8
0.7
0.8
0.7
0.6
d
BMRCSCF (Algo. 1)
BMRCSCF (Algo. 2)
BMRCMSDF
LCSCF (Algo. 3)
LCSCF (Algo. 4)
LCMSDF
0.9
BMRCSCF (Algo. 1)
BMRCSCF (Algo. 2)
BMRCMSDF
LCSCF (Algo. 3)
LCSCF (Algo. 4)
LCMSDF
0.5
0.4
0.3
0.6
0.2
0.5
0.1
0.4
0
20
15
10
SNR (dB)
0.3
Fig. 3.
0.2
0.1
0
20
15
10
SNR (dB)
Fig. 2.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE DySPAN 2010 proceedings
2
2
2
m
m
m
1
0.9
m =m
0.8
2
m m
|am bm |2
=
.
2
m
0.7
0.6
|bm am |2
m
m =m
0.5
R EFERENCES
BMRCMSDF (M = 6)
LCMSDF (M = 6)
BMRCMSDF (M = 4)
LCMSDF (M = 4)
BMRCMSDF (M = 2)
LCMSDF (M = 2)
0.4
0.3
0.2
0.1
20
15
10
SNR (dB)
Fig. 4.
|bm am |2
VIII. A PPENDIX
Defining am = Gm f + 2 and bm = Gm f
magnitude squared of G can be expressed as
, the
|am |2 |bm |2 +
m m =m am bm am bm
2
2
2
2
m |am | |bm | +
m
m =m |am | |bm |
|G|2
(a)
1,
m m =m
(b)
m
m =m
|am |2 |bm |2 ,