Aerospace and Electronic Systems Magazine May 2018 - 30

Real-Time Adaptive Spectrum Sensing
[0.95, 1.15] for λˆ (Figure 6). The number of samples
N is 7,500 and 550,000 for the "Fewer Samples" and
"More Samples" cases, respectively. It is of interest
to us to analyze the values of Pd for the respective
SNR levels (-40, -30, and −20 dB) for λˆ = 1 (which
corresponds to the actual threshold obtained by the
bounded threshold selection). For SNR of −40 dB,
this value is 0.5 for the "Fewer Samples" case and
0.53 for the "More Samples" case as shown in Figure
6, respectively. This explains why the curves in Figure 5 start from these points.
As for the difference in the performance between
the Fixed and the Mobility cases, we can observe
that it is insignificant and therefore, we can make the
same conclusion that we did for the cyclostationary
detector.
We also assess the effectiveness of the PU states
estimation scheme. Figure 7 shows the cumulative
distribution functions of the estimated probability
Figure 5.
 ( ON ) of the PU
that the PU is in active state Pr
Probability of Detection versus SNR for the Energy Detector in Fixed and Mobility cases
in two scenarios for the number of samples.
transmitter and the two types of detectors. It is obvious that the detectors follow a similar to the original
pattern
of
the distribution of the active states (marked in blue
one utilizes the same framework (and magnitude of the number of
in Figure 7 but differing significantly. The detectors perceive the
samples) as the cyclostationary detector, while the second one is
band as occupied for most of the time even though the pattern
realized by increasing the sampling rate and the sensing times of
of transmission of the PU shows that in reality, it is otherwise
the receiver. Consequently, this requires changing the parameters
(according to the settings of the PU configurations, defined in
of the PU as well. Under the "Fewer Samples" scenario, the detecExperimental Setup and Implementation Characteristics). This
tor analyzes at maximum of several thousand samples. That is why
is where the influence of the processing time becomes evident.
it is only able to reach the required probability of detection (0.9)
In this experiment, it is not included into the analysis of the sensat levels higher than −10 dB. As it was noted in [28], the detector
ing time and its impact on the operation of the framework is not
needs more than 1.5 million samples to achieve the recommended
detection rate in −20 dB SNR under the
standard-specific requirement for probability of false alarm (0.1). Despite of this,
we reached this goal with just several
hundred thousand samples in the "More
Samples" scenario. The significant disadvantage of the second scenario is that
the time to process so many samples increases by orders of magnitude. From the
plots, it can be seen that for both of the
scenarios, the curves start from detection
rate of 0.5 or a little above it. The reason
can be found by examining (11) with different values for the number of samples,
the SNR and the ED decision threshold. Measurement trials have led us to
believe that the probability of detection
cannot fall beneath 0.5 in such SNR levels and the amount of observed samples.
To elaborate further, we perform
similar analysis to the one in [28].
We set σ w2 = 1, normalized threshold
λˆ = λ N and examine the range of pos- Figure 6.
sible values of the probability of detecProbability of Detection versus the normalized threshold for fixed SNR levels in two scenarios for the
tion for a fixed SNR in the interval of number of samples.
30

IEEE A&E SYSTEMS MAGAZINE

MAY - JUNE 2018

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine May 2018