25-08-2017, 09:32 PM
Blind Spectrum Sensing using Bayesian Sequential
Testing with Dynamic Update
Blind Spectrum Sensing using.pdf (Size: 283.51 KB / Downloads: 24)
Abstract—
Cognitive radios require accurate spectrum sensing
decisions to minimize interference both to themselves and to
primary and/or other secondary spectrum users. In dynamic
spectrum environments, where interference may appear or disappear
on any channel at any time instant, robust spectrum
sensing is challenging particularly if only blind methods are
available. Blind sensing methods for single spectrum sample
vector operation are most sensitive at detecting changes in the
interference environment, whereas sequential testing methods use
more data to increase the reliability of detection decisions but
are insensitive to spectrum dynamics. This paper reviews the
Bayesian sequential testing approach and analyses the effect
of parameter estimation on detection performance. A reduced
complexity, two dimensional hidden Markov modeling method
is proposed to improve the sensitivity of sequential testing to
spectrum dynamics. The efficacy of this method is established
by comparison with pure sequential testing and single spectrum
sample vector detection.
I. INTRODUCTION
INSPIRED by the work of Mitola [1], spectrum sensing for
cognitive radio applications is a topic which has captured
the attention of many researchers. Recent review studies [2–4]
collectively cite some 300 articles related to spectrum sensing,
many of which address blind spectrum sensing methods such
as energy detection as also discussed in [5–7] and elsewhere.
Recently, it has been recognized by a number of researchers
[8–10] that most prior studies do not consider the case of
dynamic blind spectrum sensing where interference can appear
and/or disappear during a sensing interval.
HYPOTHESIS TESTING AND THE GLRT
The secondary cognitive radio system begins wideband
spectrum sensing based on the first DFT sample vector, denoted
R0, using the hidden Markov modeling (HMM) method
described in [7, 13] to determine which hypothesis
SEQUENTIAL TESTING WITH ESTIMATED PARAMETERS
In order to perform a meaningful analysis of the effect of
parameter estimation on the sequential testing probabilities
developed in section IV, it is first necessary to describe
likely detection scenarios. Recall that each DFT contains N
frequency bins. For the initial sampling of the nth frequency
bin, when L is small, the best estimates for σ2Zand σ2C
will
be drawn from a subset of the available N ×L samples. Thus,
define that estimates of σ2Z
will be obtained using NZ samples
and estimates of σ2C
will be obtained using NC samples.
DYNAMIC UPDATE FOR TIME-VARYING INTERFERENCE
A serious limitation of the Bayesian sequential testing
method described in sections IV to VI is the assumption
that all elements in the sequence r are identically distributed.
Sequences which violate this assumption are likely to produce
poor detection performance. In a dynamic environment, perhaps
where primary systems transmit periodically or where
other secondary users transmit opportunistically, interference
can be expected to appear and disappear at any time. Important
requirements for cognitive radio operation are to prevent
collisions by detecting interference as soon as it appears,
and to maximize spectral efficiency by accessing available
spectrum as soon as interference is not present.