Aerospace and Electronic Systems Magazine May 2018 - 22

Real-Time Adaptive Spectrum Sensing
vations of the closed form expressions for the probability of missdetection for different fading channels. In comparison to [34], the
detector is made more resilient by the introduction of a bounded
decision threshold. The bounds are chosen to optimize the ratio of
Pmd and Pfa. The noise uncertainty is also accounted for by integrating the Pmd and Pfa over its interval.

CYCLOSTATIONARY DETECTION-BASED SPECTRUM SENSING
Cyclostationary detectors have received much attention in the scientific community. Most of them ([36], [37], [38]) define the test
statistics as the sum of the Spectral Correlation Functions (also
called Cyclic Spectral Densities) for one (single-cycle detector)
or more (multicycle detector) cyclic frequencies (CFs) of the PU
signal. It is also often the case that the derivation for Pfa and Pd
for AWGN channel is used. The solution in [36] proposes a cooperative spectrum sensing scenario, in which each SU estimates the
Spectral Correlation Function (SCF) of one of the cyclic frequencies and the central SU combines the results to decide whether the
band was occupied or not. Such an approach decreases the computational complexity of the operation. The authors in [37] compared the same kind of cyclostationary detector against an ED and
a matched filter, showing that the cyclostationary detector outperforms the other two. A hybrid sensing method is introduced in [38].
The ED is applied initially, and then, if the decision is not clear, the
cyclostationary detection will be employed. In this hybrid method,
the noise is differentiated from the signal, by checking whether the
SCF is 0 for a specific CF.
All the papers mentioned above only consider the detection
performance in an AWGN channel and do not include the adaptable sensing time.
A recent study in [27] shows that the test statistics could be
obtained using the Cyclic Autocorrelation Function (CAF), which
is directly derived from the autocorrelation function. This makes
the detection algorithm more computationally efficient than such
that utilize the SCF because it is the Fourier transform of the CAF.
The effectiveness of the method could be further increased by implementing the process as a moving average (sliding correlation)
rather than the classic one. For that reason, the authors termed the
method "single-cycle detector with sliding correlation" (SCD-SC).
The closed-form expressions for the Pfa and Pd are derived. By integration over the distribution of the channel coefficient, the expression for the Pd is applied for the flat-fading case. Due to these
advantages in comparison to the other alternatives ([36], [37],
[38]), we have chosen the same algorithm for our implementation
of the cyclostationary detection scheme.

TRADE-OFF BETWEEN THE SPECTRUM SENSING
EFFICIENCY AND ITS ACCURACY
There are some works ([22], [23], [24]) which examine the potential
problems related to how long the sensing period should last so that
an opportunity to utilize an unused chunk of the spectrum may not
be lost, and also what the length of the transmission period of the
SU should be, so that the chance that it will create interference to
22

the returning PU, is minimal. All of these works consider computer
simulations of a traditional ED, which operates within a scheme
which incorporates the sensing and the transmission periods of the
SU. An expression, which takes into consideration the Pd and Pfa,
the probability that the PU is active/absent from the band, and the
distribution of the states of the PU (present/absent in the band) has
been developed. In [22], the proposed framework uses the result
given by the detector and combines it with the distribution of the
states of the PU, as a renewal process function model. The optimal
value of this function was found for a defined length of the frame
of the SU. Another alternative is presented in [23]. The distribution of the PU states is not included into the efficiency-accuracy
trade-off expression, only the assumed a priori probabilities that the
PU does or does not occupy the band. The conclusion is similar
to the one made in [22]-the results yield one optimal value for
the sensing time for the specific parameters (e.g., the number of
slots, into which the frame is divided and the duration of the frame).
In [24], the efficiency-accuracy trade-off is obtained by introducing a bound, taking into account the expected SU interference and
whether it would be tolerable or not, as well as the expected lost
opportunity for the SU to utilize the unused spectrum due to false
alarms. A spectrum sensing framework is built around these parameters but the essence here is that they are mostly dependent on
the characteristics of the PU states. Therefore, as evident from the
results, there would exist only one optimal sensing time for each
set of PU parameters and the performance of the detector does not
affect the result. This outlines the importance of the definition of
the PU states and their estimation at the SU. It also suggests that
there may not be a need to solve the trade-off expression during the
spectrum sensing process but that can be done a priori. Thus, the
already known optimal sensing time can be defined as a constant if
the parameters of the PU do not change, or it can be chosen from a
set of values, in the case when these vary (which is likely in a realworld scenario). The adaptive approach proposed in this article is
based on [22] and it appoints the optimal sensing time depending
on the estimated probability of the PU being present in the spectrum
band as described in detail in later sections of this article.

PRACTICAL IMPLEMENTATIONS OF ENERGY AND
CYCLOSTATIONARY DETECTORS
A large number of the existing reports on practical realizations of
spectrum sensing algorithms have only emerged in the last few
years. In general, the SDR platform used for such experiments is
the USRP one because of its flexibility in producing signals with
various modulations and bandwidths. Similar flexibility coupled
with processing speed can be achieved using field-programmable
gate array (FPGA) platforms, which is why they have also been
used for spectrum sensing applications in the recent years [12]-
[16]. Several USRP-based implementations of energy and cyclostationary detectors as reported in [4]-[9] examine the scenario
where one USRP is the PU transmitter, while the other is the SU
receiver, which implements the spectrum sensing function.
The solution in [4] specifies the desired decision threshold and
the number of samples for the test statistic manually from the ob-

IEEE A&E SYSTEMS MAGAZINE

MAY - JUNE 2018

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