Aerospace and Electronic Systems Magazine May 2018 - 28

Real-Time Adaptive Spectrum Sensing
Table 1.

Parameters of the Primary User Configurations
Pr(OFF)

μON-1

Pr(ON) >> Pr(OFF)

0.85

0.15

657

117

1.9

Pr(ON) ≈ Pr(OFF)

0.42

0.58

267

328

6.6

Pr(ON) << Pr(OFF)

0.14

0.86

178

1,078

4.5

We realize the SU receiver in a similar manner to the PU transmitter. The input samples are filtered with the FFT block and are
then sent to the Message Sink, which allows for their extraction
into a list that is processed inside a Python script. In this way, we
obtain the complex data, which we use for the CAF calculation
in the cyclostationary detector. For obtaining the signal variance
and the channel estimation we use the squared magnitude of the
samples. In the case of the energy detector, we only need the latter.
The processing in Python is performed by the spectrum sensing
algorithm as described previously. The radio-frequency parameters
of the receiver are the same as those of the transmitter with the exception of the gain which is 16.6/19/18 dB (for the direct-sequence
spread spectrum cyclostationary detector, OFDM cyclostationary
detector, and the energy detector, respectively) and the addition of
the FFT size equal to 64.
In order to study the scenario of an environment where the SU
devices have limited mobility within the area covered by the PU
transmitter, the receiver is randomly moved by hand in the span of
2 meters from its initial position on the table.

IMPLEMENTATION CONSIDERATIONS
In the case of a practical application of spectrum sensing in an
experimental environment, there are some considerations which
may not exist when the study is done via computer simulation. We
outline them briefly in the following.
An important thing to be regarded is the gain calibration of
the USRP-based transmitter and receiver. The purpose is to define
their gains so that the signal can be detected in the specific premise
of the experiment. This is especially important in scenarios like
ours where the transmitter and the receiver are not close to each
other, and there is no line of sight. If the gains are not sufficiently
large, the detector will not be able to differentiate the signal from
the noise because the received samples will be too weak. On the
other hand, in case the amplification is too high, we will not be able
to study signals in low SNR conditions. In this implementation we
set the gain by testing whether the PU signal is detected while the
transmitter operates without interruptions. Thus, the best balance
between the transmitter and the receiver gains is found empirically.
For the purpose of our study we examine a cyclostationary detector, which requires the cyclic frequency α and the time lag τ,
to be known. Even though, it is possible to assume their values
using our general knowledge of the modulation of the PU signal,
here we use the conclusions of some studies [47], [48] which pro28

-1
μOFF

Pr(ON)

ts, ms

pose more realistic approaches to this problem. We first consider
the analysis of the structure of different types of signals, made in
[47]. This paper proposes that the DSS signal can be detected for
α = fs and τ = 0, because if we only want to examine whether the
spectrum is occupied by such a kind of a transmission, we do not
need to identify its components (the different users whose signals
are spread along the bandwidth). As for the OFDM signal, τ = 1/
Δf (Δf is the subcarrier separation) and α = kfs(k ∈ ) because of
its multicarrier structure. Generally, k or the number of the CFs
has not clearly been defined in most of the literature [8], [9], [36],
[37], [27], in the sense that the question arises about how many and
which values of α are to be assumed. Even though some of them
can be acquired from the conclusions of [49], we utilize the empirical approach described in [48]. By plotting the cyclic correlogram
of the signal we found that the strongest peak is around 0.2 fs so this
value of α was chosen. For greater computational efficiency of the
cyclostationary method, we implement only single-cycle detection
(CAF is calculated only for one α). The correlogram is obtained by
calculating the square magnitude of the CAF summed over all the
observed samples for an initial value of α set to fs.
In general, when measurements over the air are performed with
the common USRP scenario (one being a transmitter, the other receiver) [4]-[9], the estimation of the received SNR has to be carefully considered. That is due to the need of assessing the noise
variance (and also the nominal noise power) experimentally when
the transmitter is inactive. For that reason, obtaining a very low
SNR (<< 0 dB) at the receiver, is not a trivial task. The authors
in [7] propose for the received SNR to be defined as a fraction of
the reference SNR level (called full scale SNR) which is the level
measured when the transmitter and receiver are very close to each
other. Using this method, the study in [7] is performed for SNR
even less than −15 dB. That is why we utilize it in our experiment
to define the SNR. The noise power is estimated by measuring the
channel for 5 seconds in the absence of the transmitter's signal and
finding the average of the 2,000,000 samples obtained.
Generally, the USRP receiver experiences local oscillator leakage which is dealt by using a Blackmann-Harris window function
in the FFT block as it is also suggested in [50].
Considering the need for solving the Marcum Q function of first
order [51], we suggest the algorithmic implementation proposed
in [52]. It is not only easy to implement but also allows the use
of the quad function (a part of the SciPy package which provides
many useful functions for Python [53]) in order to solve the integral in (6). From purely computational point of view, an increase

IEEE A&E SYSTEMS MAGAZINE

MAY - JUNE 2018

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