Upload
vgtu
View
0
Download
0
Embed Size (px)
Citation preview
A New Approach for Spectrum Sensing in
Wideband
Liudas Stasionis #1, Arturas Serackis #2
# Electronic Systems Department, Vilnius Gediminas Technical University
Naugarduko str., 41-413, Vilnius, [email protected]@vgtu.lt
Abstract—A new algorithm of the spectrum sensing is proposedin this paper. The spectrum sensing methods presented in thispaper are optimised to implement these in FPGA based embed-ded systems. The low power and highly parallelised architectureof FPGA requires low complexity in implementation of separateprocessing units – spectrum sensors (SpS). The widely used de-tection methods, based on analysis of signal spectrum energy andstandard deviation are integrated in proposed spectrum analysisalgorithm for cognitive radio spectrum sensing applications. Theexperimental investigation of proposed algorithm is performed insimulated and real RF environments. It is showed that proposedalgorithm increase the spectrum sensing efficiency and minimisethe miss-detection of licensed users to 12 % using only the energydetector in proposed algorithm and 5 % additionally addingthe standard deviation based spectrum sensing. Three channelverification techniques with proposed spectrum sensors weretested during our investigation. The Forward Consecutive MeanExcision (FCME) based technique showed the highest accuracyin channel verification task.
Index Terms—spectrum sensing, cognitive radio (CR), parallelcomputing
I. INTRODUCTION
The increasing number of communication devices, which
requires high data rates, makes current static frequency alloca-
tion schemes ineffective. The use of cognitive radio increase
the effectiveness of the use of various frequency bands not
occupied by licensed users [1].
The most challenging task for establishing the cognitive
radio is the determination of the frequency bands not occupied
by primary user. A number of spectrum sensing methods
are proposed to detect the absence of primary users in anal-
ysed frequency bands [2]. The most known spectrum sensing
methods are based on: energy detector [1], [3], [4], [5], [6],
[7], signal waveform analysis [8], detection of cyclostation-
arity features [9], transmission technology detection [10]. All
spectrum sensing methods are highly computational intensive.
Most of the proposed methods are not even tested in real-time
environment.
In this paper authors propose a new parallelized structure
of spectrum sensing algorithm based on energy detector and
estimation of modified standard deviation. The estimation of
signal spectrum energy and modified standard deviation are
made in parallel. This is a continuation of previous authors
work, related to the spectrum sensors implementation for
different RF environment [11]. The proposed spectrum sensing
algorithm is additionally tested in simulated and real RF
environments to investigate the performance of proposed mod-
ifications in evaluation of standard deviation, parallelisation of
energy and standard deviation calculations and modifications,
made for FCME algorithm implementation. Additionally pro-
posed spectrum sensor has been tested in three new RF envi-
ronments: one test signal with multiple carriers, one wideband
signal with dynamically changing carrier frequency and one
with combination of two burst signals and one narrowband
signal.
The use of energy detector-based spectrum sensing tech-
nique is reasoned by low complexity of this approach. The
parallelisation of spectrum processing threads is specially
designed for implementation of the algorithm in Field Pro-
grammable Gate Array (FPGA). The use of FPGA makes
possible to perform spectrum sensing in real-time. However
the resources on particular FPGA are limited by a number
of hardware multiply/divide units. Thus some additional spec-
trum analysis simplifications are introduced in the proposed
approach.
Three channel verification techniques are investigated in this
paper to compare the performance of direct, partial estimation
of signal spectrum parameters with adaptive FCME-based
threshold setting algorithm proposed by Lehtomaki et al. [12].
The paper is organized as follows: In Section II the structure
of the spectrum sensors is introduced ant estimated statistical
parameters are explained. Three channel verification tech-
niques are presented Section III. The parallelisation of spec-
trum sensing related calculations is introduced in Section IV. In
Section V the simulation results are discussed, which reflects
the processing speed and accuracy of proposed algorithms. The
results of experimental investigation in real RF environment
are presented and followed by conclusions in Section VII.
II. SPECTRUM SENSOR
Using ordinary averaging and squaring procedures for the
analysed signal may limit the SpS capabilities to detect
only wideband signals. The averaging of the signal spec-
trum estimates makes SpS unable to detect a low magnitude
narrowband signals. The signal standard deviation estimation
could solve this challenging task. The high values of standard
deviation for a narrowband signal makes able to create a
detector for this type of primary users.
The structure of SpS, optimised for FPGA implementation,
is shown in figure 1. Two statistical parameters are calculated
in this SpS: an average of signal spectrum estimates in selected
frequency range:
AV GH =1
N
N−1∑
k=0
|H (jω) |2, (1)
and signal modified standard deviation:
σ =
√
√
√
√
1
N
N−1∑
k=0
(AV GH − |H (jωk) |)2. (2)
The decision is made accordingly to both estimated statis-
tical parameters.
The main stages of the proposed SpS are given in Algo-
rithm 1. The structure diagram is shown in Figure 1. An
average of the spectrum estimates is calculated in upper part of
structure. Squaring is implemented by the use of multiplying
element, which multiples two same spectrum samples. To
multiply two signal samples we must use two multiplication
units: one for imaginary and one for real part of the complex
signal, received at the output of FFT block.
H (jω) = ℜ (jω) + jℑ (jω) . (3)
To estimate (|H (jω) |2), the following expression is used:
|H (jω) |2 = (ℜ (jω))2+ (ℑ (jω))
2. (4)
These operations can‘t be calculated directly, because sev-
eral bit manipulations should be performed. Xilinx FFT ip core
supports only signed 2‘s complement output. If imaginary and
real parts of this signal was multiplied from itself, then results
would be unpredictable. Not to distort the negative part of the
signal it must be converted to positive, therefore additional
signal sample preparations should be performed:
Bit manipulation operation reacts to signed bit, which is
most significant bit. If most significant bit bit is logical 1, then
all bits must be inverted, otherwise bits is left unchanged. This
operation must be done before squaring to real and imaginary
parts.
The result of the first operation (|H (jω) |2) is transferred
to accumulation part. The accumulation operation is controlled
by control signal C1. This signal appears when amount k of
accumulated samples reaches quantity N (see (1) equation).
Accumulation result SUMH is passed to multiplier element,
which is also used for standard deviation calculations. The load
of this multiplier element is low comparing with accumulation
and multiplication operations performed in previous steps.
To increase the efficiency of the algorithm several parallel
threads are sent to the multiplexer, implemented just before
the multiplier element.
Algorithm 1 The SpS algorithm
A. Average calculation of the spectrum estimates |H (jω) |.
if the number k of processed |H (jω) | values is below Nthen
set control signal C1 = 0;calculate the square of the spectrum value |H (jωk) |
2;
accumulate the calculated result: SUMH (k + 1) =SUMH (k) + |H (jωk) |
2;
increase counter value: k = k + 1;else
set control signal C1 = 1;reset counter: k = 1;
end if
calculate the multiplicative inverse: N−1;
multiply: AV GH = SUMH ·N−1.
B. Generation of the hypothesis
if AV GH > THD1 then
set hypothesis: HY Pwideband1 = 1;
else
set hypothesis: HY Pwideband0 = 1.
end if
C. Standard deviation estimation
take a negative value of |H (jω) |;calculate sum: g (n) = AV GH (n− 1)− |H (jω) |;take an absolute value |g (n) |;calculate the square of the difference: |g (n) |
2;
if the number k of processed |g (n) |2
values is below Nthen
set control signal C2 = 0;accumulate the calculated result: SUMg (k + 1) =
SUMg (k) + |g (n) |2;
root calculation: σ =√
SUMg;
increase counter value: k = k + 1;
else
set control signal C2 = 1;
reset counter: k = 1;
end if
D. Generation of the hypothesis
if σ > THD2 then
set hypothesis: HY P narrowband1 = 1;
else
set hypothesis: HY P narrowband0 = 1.
end if
Algorithm 2 Bit reconstruction
if most significant bit is 1 then
then invert all bits
else
then do nothing
end if
The multiplier element multiplies multiplexed threads of
SUMH, calculated for each sub-band, and N−1. To speedup
C1
Z-1
0
C2
1/N
SQRT
C3
|H(jω)|2 SUMH
TMPH(k)
AVGH(n)
AVGH(n-1)
-g(n) |g(n)| |g(n)|2
TMPg(k)
2(n) (n)
HYP1
HYP0THD1
THD2
MUX
MUX
MUX MUX
σ σ
∑ ∑
∑
|H(jω)|
Fig. 1. Spectrum sensor
the calculations and save system resources, the size of the sub-
band window could be selected equal to 2x samples. Then this
multiplication element can be replaced by shift operation to
the right side by x positions.
Received average AV GH of the sub-band spectrum es-
timates is compared to the threshold value THD1. Two
hypothesis could be generated accordingly made comparison:
• HY PAVG1 channel is occupied by wideband signal;
• HY PAVG0 channel is free.
The accuracy of primary user detection relies on the proper
selection of THD1 value. A simple strategy for selecting the
threshold THD1 is to set it higher, than receiver noise floor
[5]. This boundary can be found by estimating average of
input signal spectrum in analyzed bandwidth. The analysis
sub-band should be chosen wide enough to incorporate all
active noise components. The optimal sub-band width for
noise boundary estimation is whole operational bandwidth.
However the selection of the wider band for signal analysis
increases the computational load of the system.
Another strategy that could be applied for setting THD1
value is based on noise variance estimation [13]:
THD1 = σ2noise
(
1 +Q−1 (Pfa) /√
N/2)
, (5)
here Pfa is the probability of false alarm, N is the total
number of spectrum samples in the channel. Pfa value in
(5) equation must be chosen considering risk, because the
estimated threshold THD1 must guarantee non-interference
communications with primary user and utilization of unused
spectrum.
Together with signal energy estimation a standard deviation
σ is calculated for narrowband primary user signal detection.
To estimate σ, the average of the channel spectrum samples
AV GH should be calculated. The implementation of sepa-
rate AV GH estimation algorithm requires additional hardware
resources. Because the average of the spectrum samples is
already being calculated in Algorithm 1 stage A, this result
can be used to calculate the standard deviation σ. In order to
calculate the standard deviation, the AV GH value is compared
to signal spectrum estimates.
To speedup the calculation of σ, this process is implemented
in hardware as a thread, parallel to calculation of AV GH. In
order to parallelise the σ calculation, the energy average is
taken from previous spectrum analysis window. The use of
neighbouring spectrum estimates AV GH (n− 1) is possible
taking assumption that the changes of environmental noise is
not essential in small shift of spectrum analysis window. On
the other hand, the AV GH (n), which should be used for σ (n)calculation is different comparing to AV GH (n− 1), that
is used in proposed parallelised architecture. The difference
Err = AV GH (n) − AV GH (n− 1) ; increases the actual
calculated σ value. Therefore the selection of threshold THD2
should be adjusted for modified standard deviation value:
σ(n) =
√
√
√
√
1
N
N−1∑
k=0
(AV G(n− 1)− |H (jω) |)2. (6)
The procedure of σ estimation (C part of the Algorithm 1)
is shown in lower part of the diagram shown in Figure 1. The
standard deviation σ (n) part in hardware implementation is
inactive until AV G (n− 1) is calculated. These two threads
should work in parallel, but not sequential, so further modifi-
cation could be done.
After performed modifications the σ (n) is estimated by sub-
tracting signal spectrum components |H (jω) | from delayed
average AV G (n− 1). The sign of the result is changed to
positive, by changing the most significant bit into 0. The fol-
lowing operations are performed like in signal averaging part.
Before the decision is made, the square of the intermediate
estimate |g (n) | is calculated. The decision about presence of
the narrowband primary user signal is made by comparison of
(6) equation result with THD2.
The modification of standard deviation frees up: 7 slices, 26slice registers and 1 dsp48e1s. If there are 20 signal spectrum
components in analysed channel, modified calculation requires
164 less cycles and 84 cycles less if there are 10 spectrum
components.
From SpS structure (Figure 1) it is obvious, that the average
thread AV GH delay should be selected according to the
σ (n) estimation time. Therefore, if accordingly to the energy
detector estimate the hypothesis HY Pwideband1 is set to 1,
then standard deviation estimation is meaningless and it can
be interrupted.
III. CHANNEL VERIFICATION TECHNIQUES
Several techniques could be used for channel verification,
based on:
• direct estimation of statistical parameters;
• partial estimation of statistical parameters;
• FCME algorithm [12].
In the first channel verification technique statistical parame-
ters are calculated for all channel spectrum components. This
is an ordinary way to estimate system parameters, but it takes
more time to detect a wideband signal. This is because all
samples should be accumulated before the decision could be
made.
Accordingly to second channel verification technique (par-
tial estimation of statistical parameters) the signal spectrum
analysis is made in 2 stages (see Figure 2). A part of
the spectrum section is analysed in the first stage and the
remaining part in the second. This kind of channel processing
is chosen due to increase processing speed of the sections.
Fig. 2. Illustration of partial estimation of statistical parameters using onesub-band processing
First stage of this channel verification technique is used
to calculate the statistical parameters only in narrow the part
of the analysed channel. If it is decided after this step, that
the channel is occupied, then the second stage of the channel
processing is skipped and algorithm proceeds with another
channel. If it is decided, that the channel is free, the second
stage of analysis is initiated. At the second stage of channel
verification technique the statistical parameters for the rest part
of section are calculated.
If channels are occupied by the wideband signals, then it
is a big probability that the primary user will be detected in
the first stage. Second step is used mostly for detection of
narrowband signals.
In the suggested channel analysis scheme SpS sub-band
processing speed depends on the quantity of narrowband and
wideband signals. If in the analysed RF spectrum part are
many wideband signals, then this segment will be processed
sufficiently faster than using direct analysis scheme.
The third technique is based on FCME algorithm. An
assumption is made, that the first sample of the spectrum
in the channel is caused by noise. Accordingly the following
comparison is used:
H (ωk+1) > Tk
N−1∑
k=0
H (ωk), (7)
here Tk is scaling factor, given in (8) equation, which defines
coefficients used by selected method, N is the number of
channel spectrum samples.
Tk = FINV (1− Pfa, 2N, 2Nk) /k. (8)
If the comparison given in equation (7) is confirmed, then
it is decided that the channel has components of the primary
user signal. Otherwise the comparison is repeated for the rest
channel samples until the boundary is reached.
Algorithm 3 FCME algorithm
A. It is decided, that the first sample of channel is caused by
noise.
B. The scaling factor for kth sample is calculated (see (7)
equation).
C. The comparison of signal spectrum samples is performed:
if k = 0 then
if H (ω1) > T0H (ω0)) then
primary signal is detected.
set hypothesis: HY P1 = 1;else
continue primary signal search. Go to part B.
end if
else if k ≤ N then
if H (ωk+1) > Tk
N−1∑
k=0
H (ωk) then
if sum of previous samples is below signal spectrum
(i. e. for k = 10, H (ω11) > T10
10∑
k=0
H (ω10))
primary signal is detected.
set hypothesis: HY P1 = 1;else
continue primary signal search. Go to part B.
end if
else
all samples are tested.
set hypothesis: HY P0 = 1;end if
The implementation of this algorithm in embedded system
is complicated by scaling factor estimation procedure. It
uses F inverse cumulative function and implementation of
this function requires a lot of computational resources. The
structure proposed in Figure 3 makes possible to efficiently
implement FCME algorithm using FPGA.
T0
Tn-1
Tn
T1
T2
ctrl
Z-1 ACC
MUX
Hk+1
>_
Fig. 3. Proposed structure of FCME algorithm implementation in FPGA
The scaling factor used in FCME algorithm depends on: Pfa
false alarm ratio, N channel capacity (number of spectrum
samples in the channel) and k number of current spectrum
sample in the channel (see (8) equation). All these parameters
can be predefined before applying equations, so all Tk values
can be calculated in advance. The FCME algorithm can use
these pre-estimated scaling factor values from lookup table.
All other operations listed in this channel verification
strategy can be implemented by using simple elements like:
summation by ACC (accumulation element), comparison by
comparator, etc.
The weak part of this algorithm is the assumption, which is
made in the first stage. The probability, that the first spectrum
sample in the channel can be caused by noise or primary user,
is near in the densely used RF spectrum parts. Therefore for
the first algorithm stage it is better to use a small value, which
is near to noise level, instead of first sample of the channel.
IV. WIDEBAND SPECTRUM DIVISION TO PARALLEL
SENSORS
Energy detector based spectrum sensing methods requires
performing a certain arithmetic operations for analysis of the
selected signal frequency rage. Main arithmetic operations
should be parallelized in order to process a wideband spectrum
signal in real-time. In this situation, while one thread is used
for low frequency range spectrum processing, system is able
to analyse the high frequencies simultaneously.
In the proposed sensing method the wideband spectrum is
divided into equal width sub-bands with no overlap. For energy
detection based spectrum sensing methods the overlapping of
analysed sub-bands is inessential. Each sub-band (processing
thread) is analysed separately in parallel (diagram shown in
Figure 4).
Results combination
1 2 3 n-1 n
FFT
SPS SPS SPS SPS SPS
ADC
Fig. 4. Spectrum processing parallelization diagram
The ability to implement several parallel SpS defines poten-
tial band width that could be analysed by the system in real-
time. For example: if there are 10 parallel signal processing
threads, the sub-band dedicated for one spectrum sensor (SpS)
is 20 MHz, the relevant band of analysed signal is 200
MHz. To increase the whole wideband spectrum processing
speed, embedded system must have resources for more parallel
SpS. The spectrum analysis simplifications introduced in this
paper are applicable to different FPGA devices with various
parallelisation abilities.
Each SpS in dedicated sub-band search for free channels.
The selection of channel width should be carefully performed.
The minimum width of the sub-band should be selected
according to the possible bandwidth of the primary user signal.
The signal spectrum analysis algorithm is summarised in
Algorithm 4.
Algorithm 4 The parallelisation of the spectrum for SpS
A. Estimation of the signal spectrum.
1) Analog-to-Digital conversion.
2) Calculation of the Fast Fourier Transform.
B. Wideband spectrum division into n number of sub-bands.
C. Each sub-band is analysed using SpS.
D. The decision on the presence of primary user is made.
V. SIMULATION RESULTS
Two simulations were performed in this experimental inves-
tigation. The aim of the first simulation is to investigate the
spectrum processing speed by using 2 stages and FCME algo-
rithm for channel verification. Second simulation is performed
to investigate the efficiency of proposed SpS for narrowband
signal detection.
For the first simulation the 10 MHz spectrum band was
occupied by different wideband signals from 0.05 % to 50 %of analysed band width. The relative duration of spectrum
processing (see Figure 5) was registered and results compared
to ordinary calculation (channels calculated by using first ver-
ification strategy), by using different 1 stage channel windows
(0.2, 0.25, 0.4, 0.5, 0.6 of channel width). Best results for
Fig. 5. Processing time ratio dependency from 10 MHz spectrum occupancyby wideband signals
2 stage channel verification strategy were achieved by using
0.2 window. It was most efficient than regular processing,
when spectrum occupancy by wideband signals was 5 %. The
algorithm was most inefficient using 0.6 window, and it has
reached ”Efficiency boundary“ just when occupancy was 25 %.
Better results were achieved by using FCME algorithm
based technique (see Figure 5). The received efficiency ra-
tio was close to ordinary estimation result, when channel
occupancy was just 0, 05 %. FCME algorithm has showed
excellent performance, when channel occupancy was 50 %,
and it performed more than 2 times faster than first channel
verification technique. Third analysis method has showed
∼ 20 % better results than 2 stage algorithm best scenario,
in all occupancy situations.
Fig. 6. Mismatch ratio dependency from SNR
The second simulation were performed for generated fre-
quency hopping examples in 1 MHz spectrum band. Hoppers
band various from 30 kHz to 50 kHz. One hop length was fixed
at 500 µs, to model GSM signal hoping [14]. This simulation
was made because GSM band is occupied just by 50 % even in
densely populated areas. Channels who are not being used by
cellular phones, can be exploited by other services. Therefore
the capability of detecting this kind of signals must be tested.
Five detectors (energy, standard deviation, modified stan-
dard deviation with direct verification strategy and energy,
standard deviation based on FCME algorithm) were tested on
various SNR, changed from 40 dB to 0 dB (see Figure 6). An
additive white Gaussian noise (AWGN) was used as a noise
source.
Worst results were achieved by using simple energy detector
at 0 dB SNR. Using this detector acceptable results were
achieved, when SNR was 10 dB and there error ratio was
below 5 %.
The use of other 4 detectors showed similar results, but
these detectors are significantly more efficient comparing to
simple energy detector. The acceptable performance of SpS
was reached with SNR equal to 5 dB and error ratio was less
than 3 %. Worst results were achieved at 0 dB SNR with error
range from 10 % to 14 %.
VI. EXPERIMENTAL RESULTS
Five SpS (energy, standard deviation, modified standard
deviation detectors with direct channel verification technique
and energy, standard deviation detectors based on FCME al-
gorithm) were tested in this section using real RF environment
data.
Fig. 7. Spectrograms of four recorded test signals of different type
For this purpose four RF environment records were recorded
with low cost SDR (see Figure 7). These records were used
to test the primary user detection capabilities of different
signals types, which are common for RF environment [14].
First record was taken with center frequency FC = 430MHz and bandwidth BW = 0.3 MHz. This record has
two narrowband signals with constant parameters (like carrier
frequency, band) and one transmitter, which makes long bursts
(with approximate duration of 0.5 s). Second record param-
eters were equal to FC = 479 MHz, BW = 1.3 MHz, it
was one wide-band signal (approximate 0.4 MHz) with fixed
carrier frequency. Third record was made with FC = 947MHz and BW = 1.8 MHz. In this record there are several
wide-band signals (in the range from 0.35 to 0.5 MHz) with
changing carrier frequency (frequency hopping, burst time few
milliseconds). Fourth record was taken with FC = 1095 MHz,
BW = 0.2 MHz, one narrow band signal and two similar band
burst signals (the duration of burst is several milliseconds).
These four signals reflects four main types of signals met
in real environment: the narrow-band, wide-band, frequency
hoping and burst signals. The detection capabilities of pro-
posed SpS algorithm were experimentally tested using these
recorded signals to estimate the performance of SpS in various
situations.
Fig. 8. Test signals processed by energy detector
Energy detector with direct channel verification strategy
showed good performance for wide-band signal detection with
constant carrier frequency (see Figure 8). The received Pfa is
just 2 %. Narrowband signals were detected without mistakes,
but the Pfa was received at 4 % measurements and it is higher
than result, received for the wideband signal. All bursts of
signals with changing and constant carrier frequency were
detected, but Pfa was equal to 2 % and 5 % accordingly.
For all these records separate threshold THD1 was selected
manually.
Fig. 9. Test signals processed by standard deviation detector
Excellent performance standard deviation detector have
achieved for narrowband signals detection – only 1 % of Pfa
(see Figure 9). Similar to energy detector results were achieved
for wideband signal with fixed carrier frequency. False alarm
ratio was less than 2 %. For wideband and narrowband burst
signals the received Pfa was just 2 %. All bursts were detected
for signal with frequency hoping and fixed carrier frequency.
For both ordinary and modified standard deviation detectors
the threshold THD2 was selected manually too.
Fig. 10. Test signals processed by modified standard deviation detector
For all test records best results were achieved by using
modified standard deviation detector (see Figure 10). False
alarm ratio for narrowband and wideband signals was received
near to 0 %. For wideband signal with changing carrier
frequency the Pfa was about 1 %.
Fig. 11. Test signals processed by energy detector with FCME channelverification strategy
Significantly worse results were achieved by using energy
detector with FCME channel verification strategy (see Fig-
ure 11). Narrowband signals were detected with 5 % false
alarm ratio. Wideband signals were detected with 12 % of
miss-detection ratio and the Pfa reached 14 %. Signal with
frequency hopping detection was inefficient (only 65 % of
bursts were detected). Worst results were achieved in burst
detection of narrowband signals (only few burst detected). For
both detectors (energy and standard deviation) with FCME
channel verification algorithms the threshold was selected
automatically.
Fig. 12. Test signals processed by standard deviation detector with FCMEchannel verification strategy
Better results were achieved by using standard deviation
detector with FCME algorithm (see Figure 12). Narrow band
signals were detected with 3 % false alarm ratio. Miss-
detection ratio for the wideband signal was less than 5 %and Pfa = 6 %. All bursts of wideband signal with changing
carrier frequency were detected with achieved 3 % false alarm
ratio. Excellent performance for this detector was received for
narrowband burst signals. These signals were detected with
false alarm ratio near to 0 %.
VII. CONCLUSION
This article discusses spectrum parallelization importance
for wide-band spectrum sensing in real time. System which
uses just one SpS requires high clock rate. This problem can
be solved by using several parallel SpS.
For SpS implementation the energy, standard deviation and
modified standard deviation based detectors were selected.
Better results in simulation and experimental investigation
were achieved by using ordinary and modified versions of
standard deviation detectors. During the simulation the energy
detector achieved acceptable results (with Pfa ≤ 5 %) with
10 dB S/N ratio and 5 dB for standard deviation detector.
The experimental investigation has showed that the standard
deviation detectors may achieve lower false alarm ratio by few
percents.
The shortest processing time for channel verification were
achieved by using FCME algorithm. This algorithm is 2 times
faster comparing to standard channel verification technique
with 50 % spectrum occupancy by wideband signals. This ver-
ification technique has shown good detection performance in
simulation results where it perform effectively (Pfa ≤ 3 % for
5 dB SNR) comparing to standard energy detector (Pfa ≤ 5 %for 10 dB SNR). Unfortunately with real RF spectrum data
FCME algorithm based channel verification technique was less
efficient. The miss-detection ratio for wideband signals using
energy detector was 12 % and 5 % using standard deviation
detector.
REFERENCES
[1] S. ElRamly, F. Newagy, H. Yousry, and A. Elezabi, “Novel modifiedenergy detection spectrum sensing technique for fm wireless microphonesignals,” in Communication Software and Networks ICCSN, 2011 IEEE
3rd International Conference on, may 2011, pp. 59–63.[2] T. Yucek and H. Arslan, “A survey of spectrum sensing algorithms for
cognitive radio applications,” Communications Surveys Tutorials, IEEE,vol. 11, no. 1, pp. 116–130, quarter 2009.
[3] F. Penna, C. Pastrone, M. Spirito, and R. Garello, “Energy detectionspectrum sensing with discontinuous primary user signal,” in Communi-
cations, 2009. ICC ’09. IEEE International Conference on, june 2009,pp. 1–5.
[4] B. Shent, L. Huang, C. Zhao, Z. Zhou, and K. Kwak, “Energy detectionbased spectrum sensing for cognitive radios in noise of uncertain power,”in Communications and Information Technologies, 2008. ISCIT 2008.
International Symposium on, oct. 2008, pp. 628–633.[5] S. Imani, A. Dehkordi, and M. Kamarei, “Adaptive sub-optimal energy
detection based wideband spectrum sensing for cognitive radios,” inElectrical, Control and Computer Engineering (INECCE), 2011 Inter-
national Conference on, june 2011, pp. 22–26.[6] H. Rasheed, N. Rajatheva, and F. Haroon, “Spectrum sensing with
energy detection under shadow-fading condition,” in Wireless Pervasive
Computing (ISWPC), 2010 5th IEEE International Symposium on, may2010, pp. 104–109.
[7] S. Srinu, S. Sabat, and S. Udgata, “Wideband spectrum sensing basedon energy detection for cognitive radio network,” in Information and
Communication Technologies (WICT), 2011 World Congress on, dec.2011, pp. 651–656.
[8] S. Mishra, S. ten Brink, R. Mahadevappa, and R. Brodersen, “Cognitivetechnology for ultra-wideband/wimax coexistence,” in New Frontiers in
Dynamic Spectrum Access Networks, 2007. DySPAN 2007. 2nd IEEE
International Symposium on, april 2007, pp. 179–186.[9] M. Oner and F. Jondral, “Cyclostationarity based air interface recog-
nition for software radio systems,” in Radio and Wireless Conference,
2004 IEEE, sept. 2004, pp. 263–266.[10] M. Gandetto, M. Guainazzo, and C. S. Regazzoni, “Use of time-
frequency analysis and neural networks for mode identification in awireless software-defined radio approach,” EURASIP J. Appl. Signal
Process., vol. 2004, pp. 1778–1790, Jan. 2004. [Online]. Available:http://dx.doi.org/10.1155/S1110865704407057
[11] L. Stasionis and A. Serackis, “Experimental study of spectrum sensingalgorithm with low cost sdr,” in Proceedings of the 22nd International
Conference, September 20-21 2012, R. Rinkeviciene, Ed., VGTU. Vil-nius: Vilnius Gediminas Technical University Press Technika, September2012, pp. 117–120.
[12] J. J. Lehtomaki, J. Vartiainen, M. Juntti, and H. Saarnisaari, “Spectrumsensing with forward methods,” in Proceedings of the 2006 IEEE
conference on Military communications, ser. MILCOM’06. Piscataway,NJ, USA: IEEE Press, 2006, pp. 2217–2223. [Online]. Available:http://dl.acm.org/citation.cfm?id=1896579.1896914
[13] S. Ramly, F. Newagy, H. Yousry, and A. Elezabi, “Novel modified energydetection spectrum sensing technique for fm wireless microphone sig-nals,” in Commun. Software and Networks (ICCSN), 2011 International
Conference on, may. 2011, pp. 59–63.[14] M. A. McHenry, P. A. Tenhula, D. McCloskey, D. A. Roberson,
and C. S. Hood, “Chicago spectrum occupancy measurements &analysis and a long-term studies proposal,” in Proceedings of the
first international workshop on Technology and policy for accessing
spectrum, ser. TAPAS ’06. New York, NY, USA: ACM, 2006.[Online]. Available: http://doi.acm.org/10.1145/1234388.1234389