SYNERGETIC RADIO COOPERATIVE AND COLLABORATIVE RADIO

Cyclostationary Feature Detection Based

Spectrum Sensing Algorithm under Complicated
Electromagnetic Environment in Cognitive Radio
Networks
Yang Mingchuan1, 2, 3, Li Yuan1, Liu Xiaofeng1, Tang Wenyan3
1
Communication Research Center, Harbin Institute of Technology, China
2
Key Lab. for Spacecraft TT&C and Communication under the Ministry of Education, Chongqing
3
Instrument Science and Technology Postdoctoral Station, Harbin Institute of Technology

Abstract: This paper focuses on improving I. INTRODUCTION

the detection performance of spectrum sensing
in cognitive radio (CR) networks under com- CR is an important solution to the shortage
plicated electromagnetic environment. Some of spectrum resource, and it has been widely
existing fast spectrum sensing algorithms can- investigated over years. In CR networks, the
not get specific features of the licensed users’ cognitive users (CUs) sense the spectrum all
(LUs’) signal, thus they cannot be applied the time for adaptation or for dynamic fre-
in this situation without knowing the power quency access to utilize the spectrum without
of noise. On the other hand some algorithms disturbing LUs, which leads to good spectrum
that yield specific features are too complicat- utilization [1, 2]. Thanks to its advantage the
ed. In this paper, an algorithm based on the concept of CR can be utilized in collaborative
cyclostationary feature detection and theory networks which are widely investigated in
of Hilbert transformation is proposed. Com- recent years [3, 4] to increase their spectrum
paring with the conventional cyclostationary efficiency. On the other hand, collaborative
feature detection algorithm, this approach is networks may provide CR networks with
more flexible i.e. it can flexibly change the the method of cooperative spectrum sensing,
computational complexity according to cur- which can boost its performance of detection.
rent electromagnetic environment by changing In CR system, CUs are required to detect
its sampling times and the step size of cyclic the presence of the LUs off hand in order to
frequency. Results of simulation indicate that react and avoid impacting their transmissions.
this approach can flexibly detect the feature So it is of great importance for CUs to get the
of received signal and provide satisfactory de- specific features of received signal, such as
tection performance compared to existing ap- frequency, power, sample rate etc. [5]. On the
proaches in low Signal-to-noise Ratio (SNR) other hand, the complexity of electromagnet-
situations. ic environment makes the features of signal
Keywords: cognitive radio; cyclostationary change quickly. Therefore, faster detecting
feature detection; Hilbert transformation algorithms are required in CR system.

35 China Communications • September 2015

 However, the power of noise usually re- netic environment by changing sampling times
mains unknown in complicate electromagnetic and the step size of cyclic frequency. Detec- In this paper, a sta-
environment. So how to distinguish signal tors that use less sampling times can quickly tistic cyclostationary
from noise in this case has become a crucial detect the frequency of received signal, while detection algorithm
is proposed after
problem [6]. In order to find a balance be- increasing the sampling times makes detec-
investigating the con-
tween the speed of detection and accuracy, tor yield better detection performance when ventional method of
many algorithms have been proposed, such as SNR is low. Meanwhile, the smaller step size cyclostationary feature
matched filter detection, energy detection and of cyclic frequency is set in the procedure of detection.
cyclostationary feature detection algorithm [7, detection, the bigger spectrum resolution ratio
8]. Matched filter detection is one of the best detectors can yield.
methods based on the correlative detection.
But it requires prior information of LUs’ sig- II. THEORY OF CYCLOSTATIONARY
nal, and it also requires the received signal to FEATURE DETECTION
be precisely synchronized [9]. Energy detec-
tion is based on non-relative detection, and it The periodicity of signals is generally caused
is much less complicated than matched filter by the sine waves or pulse sequences during
detection. However, the threshold of this ap- modulation. Although data transmitted in the
proach may be easily affected by noise, so this system follows a stationary stochastic process,
algorithm cannot be applied in the low-SNR some statistical features of modulated signal
situation where the power of noise remains can be periodic e.g. correlation functions.
unknown [10]. In addition, energy detection Moreover, noise is usually assumed to be
can only provide the information about wheth- wide-sense stationary with no correlation. So
er the LU is absent or not, but cannot yield detectors can distinguish it from modulated
specific features of LU’s signals. signal by analyzing the spectral correlation
At present, method of direct sequence function of the signal.
spread spectrum or frequency hopping spread The received signal can be called cy-
spectrum is usually used in complicated elec- clostationary if it satisfies the following condi-
tromagnetic environment, and which makes tions:
the SNR of received signal too low for CUs to (1)
yield good detection performance [11]. There-
fore, spectrum sensing algorithms which can (2)
yield better detection performance are required
where is the period of signal, and the op-
in this case. Cyclostationary features detection
erator E(⋅) means calculating the average of
algorithm is one of the solutions to this prob-
signal. Function is the autocorrelation
lem, while it is quite complicated as compared
function of , and represents for the time
to energy detection [12].
offset.
It can be known from the above discussion
Next, we will discuss the feature of auto-
that, many fast algorithms cannot be applied
in low-SNR situations, moreover they usually correlation function and investigate
cannot yield specific features of received sig- the theory of cyclostationary feature detection.
nals, but other algorithms which can do this Assume there is a complex sine wave signal
are quite complicated. Aiming at this contra- defined as:
diction, the existing cyclostationary feature de- (3)
tection algorithm and theory of Hilbert trans- where represents the amplitude, and ,
formation are investigated in this paper, and f0 is its frequency and θ represents for phase.
then a flexible detection algorithm is proposed. Additive white Gaussian noise (AWGN) is
This algorithm can change its computation added to the signal. Therefore, the received
complexity according to current electromag- signal in CUs can be given as:

China Communications • September 2015 36

 (4) be infinite. So even if , the results of
CAF and CSD are not zero sometimes.
As the autocorrelation function is
It can be concluded from the above dis-
periodic, we can define its Fourier series
cussion that, by analyzing the CAF and CSD,
as:
conventional cyclostationary feature detection
(5) algorithm can distinguish signal from noise
and get some specific features. This method of
where is defined as cyclic frequency, and detection can mitigate the influence caused by
is called cyclic autocorrelation function noise, and it can provide better detection per-
(CAF). formance when SNR is low.
If we ignore the noise in the received signal However, the disadvantage of this algo-
and combine Eqs. (2), (3) and (5), then we get rithm is also obvious: Firstly, it requires the
that: signal transmitted to be cyclostationary, or the
detection performance may not be satisfactory.
Secondly, from the procedure of this algorithm
it can be known that it’s quite complicated,
and which makes the detection last longer.
Aiming at its disadvantage, we propose an
approach of statistic cyclostationary detection.
(6)
This algorithm is flexible and can be applied
Therefore, by calculating the value of CAF
in sophisticated electromagnetic environment.
on different cyclic frequency, we can know
the frequency of received signal. However,
the received signal can be influenced by noise III. THEORY OF STATISTIC
generally; furthermore, usual electromagnetic CYCLOSTATIONARY DETECTION
signals transmitted in space are not in complex
In this algorithm, it’s not necessary to cal-
form. In order to simplify the algorithm and
culate the cyclic autocorrelation function of
make sure it can work under the influence of
signal. Instead, we add a factor , which is
noise, we usually apply the theory of Fourier
transformation on CAF and get that: related to cyclic frequency , to the received
signal. And then calculate its average on a cer-
(7) tain cyclic frequency to get the corresponding
where is cyclostaionary spectral density result.
(CSD). Assume that is the signal received by
As for the stationary and stochastic signal CU, and because of the influence caused by
like AWGN, its average of power is zero, electromagnetic environment, may con-
i.e. . And note that , tain both signal and noise . So there
consequently for any cyclic frequency ,
are two hypotheses of given as follows:
the results of CAF and CSD are both zero,
i.e. and . On the contrary, (8)
for cyclostationary signal, there must be a
nonzero cyclic frequency to make sure that where represents for the assumption of
and . So in this way, CUs absence of LU when the signal contains noise
can distinguish cyclostationary signal from only; while represents for the hypothesis of
noise by calculating CAF and CSD on differ- presence of LU, and in this situation both sig-
ent cyclic frequencies. nal and noise are received by CU.
Actually, with the limit of memory and de- First, we add a factor to and
tecting time, the number of sampling cannot get that:

37 China Communications • September 2015

 (9) it can be known that can also reach
Next we calculate the average of . its peak when cyclic frequency equals to f0.
Assume the length of detecting time is t0, we The feature of variable is illustrat-
define a variable for a certain cyclic ed in figure 1. In this figure, the frequency of
frequency as follows: received signal f0 is set as 100Hz while noise
is ignored.
Figure 1 indicates that, when cyclic fre-
(10) quency equals to the frequency of LUs’ signal,
If we assume that the received signal is in result of detection reaches its peak,
complex form without noise as described in
no matter how long the detecting time t0 is. So
Eqs. (3), then the result of detection in CU can
if we ignore the value of the peak, which is
be given as follows:
 t   t  supposed to be infinite theoretically, then the
 s(t) × e− j2παt dt
0
 Ae j(2π f t+θ) × e− j2παt dt
0
0

M(α, t0 ) =
0
=
0
result can be simplified to a variable ,
t t0
 t 0  which is related to cyclic frequency only. After
|Ae jθ |  0
 |Ae jθ |
= e j2π( f −α)t
dt = |δ( f0 − α)|
setting a certain and proper detecting time t0,
0

t0  0  t0
(11) detector can acquire the frequency of received
where the operator δ(⋅) represents for Dirac signal by changing cyclic frequency to find the
function. peak value.
It can be known from Eqs. (11) that, if the In other words, CUs can detect the received
length of detecting time t0 is set as an integer signal for certain and proper detecting time,
multiple of the interval [0,2π], then the result and set different cyclic frequencies in a band.
of detection reaches its peak only According to the peaks of the result ,
when cyclic frequency equals to f0. So we can the probable spectrum of that signal can be
get the frequency of received signal by contin- yielded.
ually changing the value of to find the peak However, noise in channel can affect the
of that result. result of detection practically. In order to miti-
H o w e v e r, g e n e r a l l y e l e c t r o m a g n e t - gate the influence caused by noise, we consid-
ic signals are transmitted in real form, i.e. er using the complex form of received signal
. In this case, we should in detection as shown in Eqs. (11). According
reconsider the Eqs. (11). If we ignore the noise to the theory of Hilbert transformation, the
in channel, and simplify it as follows:
M(α, t0 ) = t1  0 x(α, t)dt = t1  0 r(t) × e− j2παt dt
 t 0
  t 0

0 0

= t1  0 A cos(2π f0 t + ϕ) × e− j2παt dt

 t 0

0

|A|
= 2t |e− jϕ Ψ1 ( f0 , α) + e jϕ Ψ2 ( f0 , α)|
0

(12)
where both and are func-
tions related to frequency f0 and cyclic fre-
quency . And they are given as follows:

(13)

(14)
It’s obvious that cyclic frequency , so
if the integral interval [0, t0] is an integer mul-
tiple of the interval [0,2π], then ,
. Therefore, from Eqs. (12) Fig.1 The feature of variable

China Communications • September 2015 38

 procedure of getting the complex signal from In the algorithm of statistic cyclostationary
its real component xr(n) can be illustrated in detection, sampling times N and the step size
Figure 2. of cyclic frequency will dramatically affect
In figure 2, xi(n) is the imaginary compo- the amount of calculation. Large amount of
nent of complex signal x(n). And the impulse calculation can provide good detection perfor-
response of Hilbert transformer is given as mance, but it will take longer time and more
follows: energy for detection. In practical application,
there should be a balance between detection
(15) performance and its speed of detection.

Therefore, Hilbert transformation, which IV. IMPLEMENTATION OF THE STATISTIC

shows the relationship between the imaginary CYCLOSTATIONARY ALGORITHM
component and the real component of com-
plex signal, can be defined as follows: Actually, it’s quite difficult for CUs to deal
with analogue signals, so these signals are
(16)
usually transformed into digital form before
where the operator represents for Hilbert processing. In this subsection, we will discuss
transformation. the statistic cyclostationary algorithm for digi-
Furthermore, in order to alleviate the im- tal signals.
pact of noise, CUs can sample the signal for
4.1 Procedure of the statistic
certain times and calculate the average of the
cyclostationary detection
results to improve detection performance.
As AWGN is wide-sense stationary, the Assume that the number of samples for each
procedure of calculating the average of several procedure of detection is N0, and according to
sampling results can mitigate the influence the theory of statistic cyclostationary detec-
caused by noise. Therefore CUs can work in tion, the procedure of one single search can be
low SNR situations and provide better detec- illustrated in figure 3.
tion performance in this way. In order to mitigate the influence caused by
From above discussion the principle of noise, detectors sample the received signal at
statistic cyclostationary detection can be con- contiguous points in time for N times. There-
cluded as follows: CU samples the received fore, the final result of detection can be
signal at contiguous points in time for N times, given as follows:
and detect each for a certain and proper
detecting time t0, and then average the results
of all the detections. (17)
If CU has several antennas, it can detect
every signal received by each antenna and av-
erage all these results to get the final result of where rk(m) is the kth received digital signal
detection. and represents for its detecting result.
After completing one single search, CU can
compare the final result on every cyclic
frequency with some certain thresholds. If the
value of on one cyclic frequency is
higher than the threshold, it indicates that the
signal contains this frequency component. On
the other hand CU can also put out the final
result on all different cyclic frequencies,
Fig.2 Procedure of getting the complex signal

39 China Communications • September 2015

and the probable spectrum can be directly giv-
en in this way.

4.2 The computation complexity

of the statistic cyclostationary
detection
It can be indicated from the procedure that,
the step size of cyclic frequency directly af-
fects the spectrum resolution ratio of system.
And it’s obvious that the smaller step size of
cyclic frequency used when detecting, the e  j 2π m
bigger spectrum resolution ratio can be yield-
ed. However, the amount of calculation will
increase at the same time. So there should be a 1 N0

balance between spectrum resolution ratio and N0

( )
m 1
computation complexity.
On the other hand, sampling times N can ( )
also affect the amount of calculation. If sam-
pling times is increased then detectors can
provide better detection performance and
work in the low-SNR situation. Meanwhile,
the amount of calculation will also increase
Fig.3 The procedure of one single search
and the procedure of detection will last longer.
In conclusion, step size of cyclic frequency Table I simulation parameters
and sampling times N can affect computation parameter description value
complexity. In the low-SNR situation, detec- T sampling time 0.001s
tors can increase the sampling times to yield
Fs sampling frequency 1 MHz
better detection performance. Moreover, in the
S step size of cyclic frequency α can be changed
procedure of detection, detectors can use large
N sampling times can be changed
step size of cyclic frequency to search for the
probable band of the signal at the beginning,
and once it is located, they can decrease the Assume that received signal is combined
step size to get bigger spectrum resolution ra- with two simple sine waves with different
tio. In this way, CU can flexibly control speed phases. The frequency of these sine waves is
of detection by changing these two parameters 150 kHz and 400 kHz respectively, and they
according to current electromagnetic environ- are transmitted in AWGN channel. Parameters
ment. used in simulation are listed in Table I.
At first, this approach is used to detect the
simple signal for only one sampling time with
V. PERFORMANCE SIMULATION
different spectrum resolution ratios at SNR=
-10dB. Two of the results in this condition
5.1 Simulation of statistic are illustrated in figure 4, where the values
cyclostationary detection on simple of statistic M are normalized to [0,1], i.e. the
signal minimum value among them is subtracted
In order to verify the availability of statistic from each value and then it is divided by the
cyclostationary algorithm, it’s utilized to de- maximum value. Results of simulation in
tect simple signal with some single frequen- figure 4 indicate that different spectrum reso-
cies in this subsection. lution ratios can be yielded via changing the

China Communications • September 2015 40

 step size of cyclic frequency. In practical ap-
1 plication, the computational complexity of this
0.9 approach and its spectrum resolution ratio can
0.8 be flexibly changed by the step size of cyclic
0.7 frequency.
However, the influence caused by noise can
Value of statistic M

0.6

0.5 make the result of detection unstable. There-

0.4 fore, if detectors only sample one time in the
0.3 low-SNR situation, the detection performance
0.2 will dramatically deteriorate. In this case, de-
0.1 tectors can increase the sampling times N to
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
yield better detection performance. Figure 5
Cyclic frequency(Hz)
illustrates the peak to average ratio of results
5
x 10

a) The result of simulation with spectrum resolution ratio is 5 kHz in detection versus SNR for different sampling
times N, where 10000 samples are detected for
1 every SNR value in each procedure of statistic
0.9 cyclostationary detection.
0.8 It can be known from figure 5 that the peak
0.7 to average ratio of results increases along with
Value of statistic M

0.6 the increase of SNR. Generally, higher peak to

0.5 average ratio will make it easier to distinguish
0.4 signals from noise, and the peak to average
0.3 ratios of the results in figure 4 a) and b) are
0.2 respectively 6.033 and 6.040 for reference.
0.1 Results in figure 5 also indicate that detectors
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
may mitigate the influence caused by noise
Cyclic frequency(Hz)
and boost their detection performance in low-
5
x 10

SNR situation by increasing sampling times.

b) The result of simulation with spectrum resolution ratio is 1 kHz
On the other hand, increasing sampling
times means larger amount of calculation
Fig.4 Results of simulation when N=1, SNR=-10dB
and longer time for detection. So CUs should
consider the balance between detection perfor-
mance and the speed of detection. If the condi-
tion of channel is good, then detectors can de-
16
crease the sampling times; otherwise they can
statistic cyclostationary detection N=1
14
statistic cyclostationary detection N=4 increase it to get better detection performance.
statistic cyclostationary detection N=8
12 statistic cyclostationary detection N=16
5.2 Simulation of statistic
cyclostationary detection on BPSK
Peak to average ratio

10

8 signal
6
In this subsection, statistic cyclostationary de-
4 tection algorithm is used to detect modulated
2
Binary Phase Shift Keying (BPSK) signal in
different SNR for testing its performance in
0
-35 -30 -25
SNR(dB)
-20 -15 -10
detecting cyclostationary signals.
It should be noted that although BPSK
signal is cyclostationary, the phase of carrier
Fig.5 Peak to average ratio vs. SNR for different sampling times N can be abruptly changed due to the procedure

41 China Communications • September 2015

of modulation. Therefore, its frequency spec-
trum will extend on both sides of the carrier
1
frequency. Assume that the SNR of detector is
0.9
-5dB and the modulated BPSK signal whose
0.8
carrier frequency is 800MHz is transmitted in
0.7
AWGN channel. The parameters used in simu-

Value of statistic M
0.6
lation are shown in Table I while the step size
0.5
of cyclic frequency is set as 1 MHz, sam-
0.4
pling times N=200. The result of simulation
0.3
which is normalized to [0,1] is shown in figure
0.2
6. From the result of simulation, probable
0.1
spectrum of BPSK signal can be illustrated.
0
If the transmission condition deteriorates, 0 2 4 6 8 10
Cyclic frequency(Hz)
12 14 16
8
x 10
which leads to decrease of SNR at detectors,
performance of detection will be influenced
and this approach may not yield probable Fig.6 Result of simulation on BPSK signal when N=200, SNR=-5
spectrum of the received signal. In this case,
increasing the sampling times N can mitigate
the influence cause by noise. 1

If the spectrum of the received signal is 0.9

not required in the system, i.e. CU only wants 0.8

to know whether LU is present or not. In this 0.7

Probability of Detection

case, we can set a proper threshold to make 0.6

decisions: if the average value of results is 0.5

higher than that threshold CU decides that the 0.4

LU is present; otherwise CU predicates that 0.3

the received signal is noise. Figure 7 shows 0.2

statistic cyclostationary detection N=8
statistic cyclostationary detection N=4
the probability of detection Pd of statistic cy- 0.1 statistic cyclostationary detection N=2
conventional energy detection
clostationary method vs. SNR for different 0
-20 -18 -16 -14 -12 -10 -8 -6 -4 -2 0
SNR(dB)
sampling times N at a fixed probability of
false alarm Pf=0.005. For comparison, the per-
formance of conventional energy detection is Fig.7 Pd vs. SNR for BPSK signal at Pf =0.005
illustrated as well.
It can be indicated from figure 7 that the
statistic cyclostationary detection can pro- 1

vide satisfactory detection performance when 0.9

SNR of detector is low. Moreover, the perfor- 0.8

mance of statistic cyclostationary detection 0.7

Probability of Detection

with sampling times N=2 is similar to that 0.6

of conventional energy detection, while this 0.5

performance can become better along with 0.4

the increase of sampling times. Furthermore, 0.3 conventional energy detection at SNR=-15dB
conventional energy detection at SNR=-10dB
in order to exhibit the relationship between 0.2 statistic cyclostationary detection N=4 at SNR=-15dB
statistic cyclostationary detection N=4 at SNR=-10dB
sensitivity (Pd) and specificity (Pf) of the sta- 0.1 statistic cyclostationary detection N=8 at SNR=-15dB
statistic cyclostationary detection N=8 at SNR=-10dB
tistic cyclostationary detection algorithm, the 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Probability of False Alarm
receiver operating characteristic (ROC) curves
are shown in figure 8, where the ROC curves
of conventional energy detection are also illus- Fig.8 ROC curves at different SNR values

China Communications • September 2015 42

 trated for comparison. CTTC-FX201305).
The ROC curves in figure 8 indicate that
the performance of statistic cyclostationary References
detection algorithm is better than that of con- [1] Y. Liang, K. Chen, G Li, and P Mahonen, “Cog-
nitive Radio Networking and Communications:
ventional energy detection. Moreover, this An Overview,” IEEE Trans. vehicular technology,
performance can be improved by increasing vol. 60, no. 7, pp. 3386-3407, Sep. 2011.
sampling times N, but the amount of calcula- [2] M Heng. “Research on Spectrum Detection
tion will increase at the same time. Technology in Cognitive Radio,” Radio Engi-
neering Journal, vol. 44, no. 03. pp.77-79, 2014.
[3] W. Feng, N. Ge, and J. Lu, “Hierarchical trans-
VI. CONCLUSION mission optimization for massively dense dis-
tributed antenna systems,” IEEE Commun. Lett.,
In this paper, a statistic cyclostationary detec- vol. 19, no. 4, pp. 673-676, Apr. 2015.
[4] W. Feng, Y. Wang, N. Ge, J. Lu, and J. Zhang,
tion algorithm is proposed after investigating “Virtual MIMO in multi-cell distributed antenna
the conventional method of cyclostationary systems: coordinated transmissions with large-
feature detection. This approach is based on scale CSIT,” IEEE J. Sel. Areas Commun., vol. 31,
cyclostationary feature, and uses the theory no. 10, pp. 2067-2081, Oct. 2013.
[5] Taherpour A., Nasiri-Kenari M.,and Gazor S.
of Hilbert transformation. Comparing with “Multiple Antenna Spectrum Sensing in Cogni-
the cyclostationary feature detection, this tive Radios,” IEEE Trans. on Wireless Commun.,
approach is more flexible, i.e. it can flexibly vol. 9, no. 2, pp. 814-823, Feb. 2010.
control the computational complexity and the [6] A. Mohammadi, M. R. Taban, J. Abouei, and H.
Torabi. “Cooperative spectrum sensing against
speed of detection according to current elec- noise uncertainty using Neyman-Pearson
tromagnetic environment. Results of simula- lemma on fuzzy hypothesis test,” Applied Soft
tion indicate that this approach can be used to Computing Journal, vol. 13 ,no.7, pp. 3307-
detect both simple signals and modulated cy- 3313, Jul. 2013
[7] S. Haykin. “Cognitive radio: brain-empowered
clostionary signals, and yield satisfactory per- wireless communications,”. IEEE J. Sel. Areas
formance comparing to conventional energy Commun., vol. 23, no. 2, pp. 0733-8716, Feb.
detection algorithm. Moreover, this algorithm 2005.
can get specific features of received signal, [8] S. Xie, Y. Liu, Y. Zhang, and R. Yu, “A Parallel Co-
operative Spectrum Sensing in Cognitive Radio
while it doesn’t require the priori information. Networks,” IEEE Trans. vehicular technology, vol.
Therefore, it can work in the situation where 59, no. 8, pp. 4079-4092, Oct. 2010.
the power of noise remains unknown for CUs [9] Y. C. Eldara, A. V. Oppenheimb, and D. Eg-
and provide detectors with satisfactory de- nor, “Orthogonal and projected orthogonal
matched filter detection,” Signal Processing
tection performance when the SNR is low. In Journal, vol. 84, no.4, pp. 677-693, Apr. 2004.
conclusion, via flexibly controlling its amount [10] Ashish Bagwari and Geetam Singh Tomar
of calculation, this algorithm can get specific “Adaptive double-thresholdbased energy de-
features of the licensed users’ signal while ex- tector for spectrum sensing in cognitive radio
networks,” International Journal of Electronics
ecuting the detection as fast as possible. Letters, vol. 1, no. 1, pp. 24-32, Apr. 2013.
[11] C. Hwang, S. Chen, and T. Hsinchu, “Spectrum
ACKNOWLEDGMENT sensing in wideband OFDM cognitive radios,”
IEEE Trans. Signal Processing, vol. 58, no. 2, pp.
The paper is sponsored by National Basic 709-719, Feb. 2010.
[12] Gill, R and Kansal, A, “Comparative Analysis of
Research Program of China (973 Program,
the Spectrum Sensing Techniques Energy De-
No. 2013CB329003), National Natural Sci- tection and Cyclostationary Feature Detection”
ence Foundation of China (No. 91438205), International Journal of Advanced Research
China Postdoctoral Science Foundation (No. in Electrical, Electronics and Instrumentation
Engineering, vol. 3, no. 7, pp. 10601-10607, Jul.
2011M500664). Open Research fund Program
2014.
of Key Lab. for Spacecraft TT&C and Com-
munication, Ministry of Education, China (No.

43 China Communications • September 2015

Biographies Li Yuan(1992-), is studying for a master’s degree
in Information and Communication Engineering at
Yang Mingchuan(1977-), is currently a research
Harbin Institute of Technology, China. His research
associate of Communication Research Center Harbin
interests are in the fields of the quick reconstruction
Institute of Technology, China. He received a B.Sc. de-
of the wireless networks under complicated electro-
gree in Mechanical Engineering from Harbin Institute
magnetic environments.
of Technology, China, in 2001, and a M.Sc degree and
Ph.D degree in Information and Communication En-
Liu Xiaofeng(1961-), is currently a professor of
gineering from Harbin Institute of Technology, China,
Communication Research Center Harbin Institute
in 2004 and 2010 respectively His research interests
of Technology, China. His research interests are in
are in the fields of space information network and
the fields of space information network and mobile
mobile propagation channel modeling, especially
propagation channel modeling.
mobile satellite and High Altitude Platforms (HAPs).
He also has done some research on the quick recon-
Tang Wenyan(1962-), is currently a professor of
struction of the wireless networks under complicated
Communication Research Center Harbin Institute of
electromagnetic environments. *The corresponding
Technology, China. His research interests are in the
author Email: [email protected]
fields of Sensors and Signal Processing.

China Communications • September 2015 44