Вы находитесь на странице: 1из 10

Spectrum Sensing for Cognitve Radio

SAGAR AHIR(201211045)
Contents
1 Introduction to Cognitive Radio 2
1.1 Access Techniques for CR networks . . . . . . . . . . . . . . . 2
1.2 Approaches to Spectrum Sensing in Overlay Cognitive Radio
Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Hypothesis Testing 3
2.1 Types of hypothesis testing problems . . . . . . . . . . . . . . 4
2.2 Sequential Hypothesis Test or Sequential Probability Ratio
Test (SPRT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3 Spectrum Sensing in Multiple Antenna Based CR Systems 6
3.1 System model description of a multiple antennae enabled CR
network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2 Mathematical formulation of the spectrum sensing problem . . 8
3.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4 Conclusion 9
References 10
List of Figures
1 System Model of multiple antenna based Cognitive Radio Net-
work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1
1 Introduction to Cognitive Radio
Cognitive Radio (CR) is an emerging communication design paradigm [1] in
which the radios (devices equipped with transceivers can sense the under-
lying radio environment and adapt their transmission/reception parameters
like operating frequency, power, modulation rate, etc. The specic instance
in which the operating frequency is adjusted is known as Dynamic Spec-
trum Access(DSA). DSA has the potential to alleviate the problem of spec-
trum shortage. The problem of spectrum shortage can create problems in
allocating spectrum for the numerous wireless applications that are being
conceived. On the other hand a recent survey by Federal Communications
Commission(FCC), USA, [2] suggests that vast portions of allocated spec-
trum is vastly underutilized. A radio enabled with DSA can indeed access the
spectrum opportunistically whenever the licensed users are inactive, thereby
leading to ecient spectrum utilization. Henceforth, we refer to networks
that use DSA as CR networks. Since CR networks allow unlicensed users
to access opportunistically therefore it contains two types of users namely
primary (licensed) users and secondary (unlicensed) users. Next we discuss
about the techniques adopted by CR networks.
1.1 Access Techniques for CR networks
Access techniques for CR networks can be classied into two types :
Overlay approach (or interference-free approach): In this approach,
the secondary users access portion of the spectrum that is not used
by primary users [3]. As a result there is no virtual interference to
the primary users. So CR networks that refer to overlay approach are
referred as CR overlay CR networks.
Underlay approach (or interference-tolerant approach): In this ap-
proach, [3] the secondary user spreads the signals over a wide band
of frequencies and thereby be able to pursue its transmission. Spread-
ing over a large bandwidth would help secondary users transmit below
the noise oor of primary users. Therefore, the power of secondary
users has to be regulated so that they do not cause undue interference
to the primary users. The CR networks that employ underlay access
techniques are known as underlay CR networks.
In this work we focus on overlay access technique approach. For opportunistic
transmission of secondary users to be possible the underlying spectrum has
to be sensed and detected for the presence or absence of primary users. This
2
problem is referred to as spectrum sensing which would be the focus of our
work.
1.2 Approaches to Spectrum Sensing in Overlay Cog-
nitive Radio Networks
Approaches to spectrum sensing [4]-[6] can be categorized based on the ex-
change or no exchange of information between the SUs of a CR network
into:
1. Cooperative spectrum sensing: In cooperative spectrum Sensing the
SUs (i.e., radios) in a CR network exchange information with each other
or transmit to a centralized entity known as a fusion centre (FC). The
former approach is known as distributed cooperative spectrum sensing
and the latter approach is known as centralized cooperative spectrum
sensing. Once the information exchange among themselves or trans-
mission to a FC is complete a decision related to PUs presence or
absence is made accordingly.
2. Non Cooperative Spectrum sensing: In non-cooperative spectrum sens-
ing, PUs are detected by each SU independent of others. In other
words there is no explicit exchange of information between the SUs or
transmission of information to the FC as was the case in cooperative
spectrum sensing.
In general, cooperative spectrum sensing gives better performance than the
non-cooperative spectrum sensing. This is because in cooperative spectrum
sensing there is a diversity advantage which helps out in averaging the noise
eects that are introduced into the sensing process.
2 Hypothesis Testing
The problem of spectrum sensing can be cast under the framework of (binary)
hypothesis testing. In hypothesis testing one tries to determine as which
one of the competing hypotheses is true. This is accomplished by using
a sequence of observations made about the unknown hypothesis. In the
problem of spectrum sensing the two hypotheses correspond to the cases
where a primary user is present or absent respectively. In the following we
shall talk briey about the dierent hypothesis testing problems.
3
2.1 Types of hypothesis testing problems
Based on the information available about the model parameters of a given
problem hypothesis testing can be classied into
1. Simple hypothesis testing: In simple hypothesis testing, the probability
distribution of observations under each hypothesis is assumed to be
completely specied [7]. In other words there is no dependence of the
probability distribution on unknown parameters.
2. Composite hypothesis testing: In composite hypothesis testing the
probability distribution under each hypothesis is assumed to belong
to a family of distributions that are indexed by the unknown model
parameter. The popular generalized likelihood ratio test (GLRT) is an
instance of composite hypothesis testing [7].
For both simple and composite hypothesis testing problems one can have
a xed sample size test or sequential test. In a xed sample size test a
xed number of observations are taken using which the decision statistic is
computed. In a sequential test the decision maker defers the decision until
such time he has enough number of observations to make a decision. The
number of observations in the case of sequential tests is not xed and is indeed
random. Thus in the case of sequential hypothesis testing before even taking
a decision related to hypothesis the decision maker has to decide whether to
stop or continue taking observations. In case if he chooses to stop then he
will have to make a decision as to which one of the two hypotheses is true,
else he continues taking observations.
Both simple and composite hypothesis testing problems can use Bayesian
or non-Bayesian approaches to decide which one of the two hypotheses is true.
In the following we shall summarize the philosophy adopted in Bayesian and
non-Bayesian approach.
1. Bayesian approach: In this one assumes prior probabilities for each
hypothesis and the optimal decision rule is based on minimizing the
Bayes risk.
2. Non-Bayesian approach: In this there are no assumptions made on
the prior probabilities of each hypothesis. Two well known approaches
to hypothesis testing in this case are the Neyman-Pearson and Mini-
max. In the Neyman-Pearson approach one focuses on maximizing the
probability of detection subject to a constraint on the probability of
false alarm. In the Minimax approach the decision rule is based on
minimizing the maximum Bayes risk.
4
For the case of spectrum sensing in CR systems we intend to adopt sequential
testing, since it is well known that to achieve the same performance (in terms
of probability of error) the sequential tests take less number of observations
than the corresponding xed size tests. This is very much desired in the case
of CR systems that are subject to strict energy constraints.
2.2 Sequential Hypothesis Test or Sequential Proba-
bility Ratio Test (SPRT)
Sequential hypothesis tests are sometimes also referred to as SPRTs, since
the decision making involves computing the ratio of probabilities under each
hypothesis.
To introduce formally a SPRT we dene lower and upper thresholds A
and B such that 0 < A < 1 < B < . The SPRT can actually be dened as
follows:
Select H
0
whenever,
n

k=1
L(Y
k
) A
Select H
1
whenever,
n

k=1
L(Y
k
) B
Take another sample if,
A
n

k=1
L(Y
k
) B
It is indeed convenient to express the above test by expressing the decision
statistic in the logarithmic form. To that end, we dene Z
k
= ln (L(y
k
)),

n
=

n
k=1
Z
k
, a = ln(A) and b = ln(B) . The SPRT test in terms of
n
can be written as follows:
Select H
0
whenever,
n
a
Select H
1
whenever,
n
b
Take another sample if, a <
n
< b
5
The SPRT dened above can indeed be shown to be optimal for the case of
Bayesian binary sequential hypothesis testing problem [7].
If we dene N = min {n :
n
a or
n
b} to be the stopping time of
the SPRT, then it can be shown that P[N < ] = 1 which implies that
the SPRT test terminates in nite time [7]. Next, we need to determine the
values for the lower threshold A (hence a) and upper threshold B (hence
b) in terms of the probability of false alarm P
F
and probability of miss P
M
.
Alternatively, one can determine P
F
and P
M
in terms of A and B respectively.
This can be accomplished approximately via Walds identity using the zero
overshoot approximation. In zero overshoot approximation we approximate

n
by b or a whenever
n
goes past either of them. Indeed, if the increments
of Z
k
are small in comparison with thresholds, a and b then the random walk
could not have been too far away from the threshold it is going to cross.
Thus using Walds identity in conjunction with the zero overshoot ap-
proximation one can express P
F
and P
M
in terms of A and B as
P
F
=
1 A
B A
(1)
P
M
=
A(B 1)
B A
. (2)
Once the desired values of P
F
and P
M
are chosen one can compute A and B
using (1) and (2) respectively.
3 Spectrum Sensing in Multiple Antenna Based
CR Systems
In this work we specically focus on the problem of spectrum sensing in mul-
tiple antennae enabled CR networks. A detailed description of the problem
considered along with a road map for future research is given in this section.
3.1 System model description of a multiple antennae
enabled CR network
Consider a cognitive radio network in which the radios of various SUs sense
a channel [8], [9], which is primarily allocated to the PU as depicted in Fig.
3.1. Let N denote the number of radios in the network such that each ra-
dio is equipped with M number of antennas. Every antenna element of a
given radio makes a measurement about the energy on a particular channel
of interest. Thus our spectrum sensing approach is non-coherent in nature.
6
Figure 1: System Model of multiple antenna based Cognitive Radio Network
Further, we assume that each radio uses a selection combiner that outputs
the maximum out of the M measured energy values at a given radio. Other
types of combining schemes like maximal ratio combining (MRC) also can
be used, although it may entail additional overhead in terms of channel esti-
mation. Note that for selection combining of various antenna measurements
one does not require channel estimation. Using the output of the selection
combiner a given radio computes the likelihood ratio (LR)/ log likelihood
ratio (LLR) that can be used as a decision statistic to decide on the pres-
ence or absence of a PU on a given channel. The computed LR/LLR is then
sent over an error free channel to the FC. The FC accumulates the LR/LLR
values from dierent radios across time instants. The accumulated LR/LLR
value (denoted as ACC LR/ACC LLR) at the FC is then used to decide
on the presence or absence of a primary user according to an SPRT as given
below:
Accept H
0
if ACC LR (ACC LLR) < LOW THR
Accept H
1
if ACC LR (ACC LLR) > HIGH THR
Continue taking measurements if LOW THR < ACC LR (ACC LLR) < HIGH THR,
where H
0
and H
1
correspond to the hypothesis of PU absent and PU present
respectively. The symbols LOW THR and HIGH THR correspond to the
low and high threshold values used and can be set in terms of the probability
of false alarm and probability of miss as per the equations (1) and (2).
7
3.2 Mathematical formulation of the spectrum sensing
problem
Let the two hypotheses corresponding to PU be
H
0
: PU is absent
H
1
: PU is present
The measurement model [9] corresponding to each of the two hypotheses
recorded by the m
th
antenna (i = 1, 2, , M) of the n
th
cognitive radio
(n = 1, 2...N) is given by
H
0
: y
i
(t) = v
i
(t) if PU is absent
H
1
: y
i
(t) = h
i
(t)s(t) +v
i
(t) if PU is present
(3)
where i is the antenna index of a given radio, s(t) denotes the PU signal
at time instant t with energy E
s
, v
i
(t) CN(0,
2
n
) is circularly symmet-
rical complex Gaussian random variable, and all h
i
(t) CN(0,
2
h
) are in-
dependent and identically distributed complex Gaussian random variables.
This assumption implies that the underlying fading process is Rayleigh dis-
tributed. The energy W
i
associated with a given measurement of the i
th
antenna of a given radio is given by
W
i
= |y
i
|
2
, (4)
where |.| denotes the magnitude operator of a complex number.
3.3 Future Work
In the following we shall lay out a road map for our future work:
At rst we would like to optimize the number of antennas M re-
quired at each node to achieve the prescribed values of probability
of false alarm and probability of miss. To this end, we can con-
sider minimizing the maximum of the expected number of observa-
tions under both the hypothesis. If we denote N as the random num-
ber of observations required to decide between H
0
and H
1
, then we
can determine optimal number of antennas M

such that it minimizes


max (E[N|H
0
], E[N|H
1
]).
Since the optimal value of the number of antennas is based on the ap-
proximate expressions used for the upper and lower decision threshold
values, therefore we would like to corroborate via simulations if the
value of the optimal number of antennas determined above indeed hold
for the exact case as well.
8
Repeating the above mentioned things for other types of antenna com-
bining schemes like MRC will also be considered.
We shall also explore the possibility of using composite sequential hy-
pothesis testing based framework for the spectrum sensing problem
considered here.
4 Conclusion
In this report we presented a brief summary of the work pursued in summer
2013 towards the M.Tech thesis. Broadly, we reviewed CR networks and the
problem of spectrum sensing as applied to it. We then reviewed statistical
hypothesis testing under whose framework the problem of spectrum sensing is
considered. Finally, we presented the problem of spectrum sensing in multiple
antennae enabled CR networks with a few pointers to future research.
9
References
[1] E. Biglieri, A. J. Goldsmith, L. J. Greenstein, N. B. Mandayam and H. V.
Poor, Principles of Cognitive Radio , New York: Cambridge, December
2012.
[2] FCC, Spectrum Policy Task Force, ET Docket No. 02-135, Nov. 2002.
[3] T. Issariyakul, L. S. Pillutla, and V. Krishnamurthy,Tuning Radio Re-
source in an Overlay Cognitive Radio Network for TCP: Greed Isnt
Good, IEEE Communications Magazine, July 2009.
[4] I. F. Akyildiz, B. F. Lo and R. Balakrishnan,Cooperative spectrum sens-
ing in cognitive radio networks: A survey, Physical Communication, vol.
4, pp. 40-62, March 2011.
[5] K. B. Letaief and W. Zhang, Cooperative communications for cognitive
radio, Proc. IEEE, vol. 97, no. 5, pp. 878893, May 2009.
[6] T. Yucek and H. Arslan, A survey of spectrum sensing algorithms for
cognitive radio application, IEEE Communications Surveys & Tutorials,
vol. 11, no. 1, pp. 116-130, March 2009.
[7] B.C.Levy, Principles of Signal Detection and Parameter Estimation,
New York: Springer, 2008.
[8] Q. Zou, S. Zheng, and A. H. Sayed, Cooperative spectrum sensing via
sequential detection for cognitive radio networks, IEEE Transactions on
Signal Processing, vol. 58, no. 12, December 2010.
[9] A. Singh, M. R. Bhatnagar and R. K. Mallik, Cooperative spectrum
sensing in multiple antenna based cognitive radio networkd using and
improved energy detector, IEEE Communication Letters, Vol. 16, no.
12, January 2012.
10

Вам также может понравиться