Aerospace and Electronic Systems Magazine May 2018 - 31

Ivanov et al.
limited mobility of the SU does not cause a substantial degradation
in the performance of the detectors. Therefore, the implementation
of these methods in non-stationary devices is feasible.
As for the efficiency-accuracy trade-off, it has been noted that
there is no necessity for it to be obtained during each detection
instance, but it can be calculated preliminary for specific PU parameters. That is because the result of the trade-off is not influenced by the measurements of the SU receiver but by the pattern
of transmission of the PU. Therefore, accurate definition of that
pattern is critical. In a scenario like ours, the parameters of the
transmitter are known but if a realistic arbitrary PU is employed,
this will make the estimation problem much more complex. Thus,
there are two questions which can be posed. Should the estimation
of the transmission parameters be done on the basis of preliminary
CONCLUSIONS AND FUTURE WORK
measurements of the PU activity? Or should we use an estimation
algorithm which is integrated into the detectors framework (like
In this article, we proposed an adaptive approach in the spectrum
the one utilized in this study)? If the former approach is adopted,
sensing process for improving the performance of existing cyclothen the means of measuring the active/idle periods of the transmitstationary and energy detectors based on a trade-off between efter have to be determined because the parameters of the PU (P(ON)
ficiency (sensing time) and accuracy. We introduce adaptivity in the
−1
−1
and P(OFF), and μON
spectrum sensing process as a means to optimize it. We evaluate it
and μOFF
) are based on them. In the second
in a practical implementation scenario by studying the performance
case, the receiver will have also to try to predict them based not
of a cyclostationary detector and an energy detector, both integrated
only on the detection statistics but on the choice of sensing time as
into a framework, which determines the optimal sensing time and
well. Therefore, it will be required to perform the trade-off calculawas realized in practice through the USRP2 hardware platform. The
tion together with the estimation of the probabilities that the PU is
results we obtain are very close to or better than those from the
active/idle.
simulations described in the papers which proposed the implementAs an alternative, on the focus of the spectrum sensing research
ed spectrum sensing methods ([28], [27]). It is also shown that the
could be the optimization of the detection solution itself so that
it is executed as fast as possible and the
trade-off between its accuracy and sensing time can be found empirically and
set as a standard. In this way the estimation of the PUs pattern of transmission
will not be necessary.
Concerning the processing time of
the solution, it also plays a role as we
have concluded earlier. The cyclostationary detector processes the obtained
samples several times slower than the
energy detector in the equivalent scenario. However, it provides much greater
detection accuracy in the DSS signal
case but not when OFDM signal is applied. The energy detector improves its
probability of detection with the number
of samples but the time to process them
also increases significantly (it is, on average, about 30 times greater than the
computation time of the cyclostationary
detector). Nevertheless, it fulfills the requirements of the standard [57] in terms
of speed as well, because sensing periods for up to 2 seconds are permitted.
It is then clear that both of the sensing
methods perform well in low SNR but
Figure 7.
do not reach the specifications in [57]
Cumulative distributions functions of the estimated probability that the primary user is active in compari(detection probability of 0.9 at −20 dB
son to that of the transmitter.
considered in-depth. Even though, measures of reducing it have
been taken (as explained in the Experimental Study section), the
processing time causes the detector to miss sometimes the idle
period of the PU and thus, it will detect the transmitter once
more in the next sensing iteration. This will result in increas ( ON ). The opposite can also occur. Another thing to be
ing of Pr
noted is that the transmitter program does not change from one
state to the other instantly. The delay is due to the fact that the
GNU Radio flow-graph which conducts the transmission process requires between a few and several tens of milliseconds to
execute and to stop. Therefore, the transmitter also contributes
to the inaccuracies in the PU states estimation.

MAY - JUNE 2018

IEEE A&E SYSTEMS MAGAZINE

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