Optimization of Energy Detection Approach of

Spectrum Sensing in Cognitive Radio Network

Niranjan Baral, Dibakar Raj Pant
Department of Electronics and Computer Engineering, IOE, Central Campus, Pulchowk, Tribhuvan University, Nepal
niranjan@ioe.edu.np

Abstract: In the cognitive radio, spectrum sensing is the first task to check the presence of licensed users.
Among the different approaches of spectrum sensing, energy detection (ED) is the easiest approach in terms
of computational complexity which requires less prior information. In this paper, energy detection (ED) is
considered under a typical fading unknown as channel and white Gaussian noises. Introducing an idea of
auxiliary noise variance estimation for the absence of prior knowledge of noise power, Hybrid Energy
Detection-1 (HED1) / Hybrid Energy Detection-2 (HED2) approach of signal detection is set forth. For
HED1, noise variance is estimated in S auxiliary noise only slots and for HED2 noise variance is estimated
in S auxiliary slots which are declared only noise signal slots by ED. The detection performance of the
considered methods are derived and expressed by analytical formulas. The impact of noise estimation
accuracy on the performance of ED is compared based on Receiver Operating Characteristics (ROC)
curves and Performance Curves. This study shows that the performance of ED can be increased in terms of
ROC performance even in low SNR by increasing the number of samples considered for noise variance
estimation.

Index Terms – Energy Detection, Receiver Operating Charactersitics

within the given spectrum and thus contribute to the

1. INTRODUCTION
Spectrum is a valuable resource. The rapid development
in communication and technologies evolving day by day
demands more spectrum. Those ranges of spectrum
which are highly demanded in terms of propagation
properties are already assigned. The spectrum assigned is
also underutilized. Cognitive radio is an evolving
technology that can provide dynamic spectrum access to
the unlicensed users whenever the licensed users are
inactive in that range. The main purpose of cognitive
radio is to obtain the best available spectrum through
cognitive capability and provide dynamic spectrum
access. Dynamic spectrum access techniques allow the
cognitive radio to operate in the best available channel.
The cognitive radio enables the usage of temporally
unused spectrum, which is referred to as spectrum hole or
white space. If this band is further used by a licensed
user, the cognitive radio moves to another spectrum hole
or stays in the same band, altering its transmission power
level or modulation scheme to avoid interference.
“Cognitive radio” is a fully programmable wireless
device that can sense their environment and dynamically
adapt their transmission waveform, channel access
method, spectrum use, and networking protocols as
needed for good network and application performance
[1]. Cognitive radio network is a combination of
Spectrum sensing, Spectrum management, Spectrum
mobility and Spectrum sharing. This network will
perform efficiently if all the functions of the Cognitive
radio network are attained at a desirable rate. Cognitive
radio gives the unlicensed users access to the licensed
spectrum whenever the primary users are not active
effective bandwidth utilization.
Spectrum Sensing is the first task needed to be performed
for Dynamic Spectrum access in a Cognitive radio
network which enables the cognitive radio to adapt to its
environment by detecting spectrum holes. The most
efficient way to detect spectrum holes is to detect the
primary users that are receiving data within the
communication range of a secondary user. In reality,
however, it is difficult for a secondary user to have
accurate channel information between a primary receiver
and a transmitter due to the inherent property of cognitive
radio. The most recent work engages in primary
transmitter detection based on local observations of
secondary users. The current spectrum sensing methods
can be classified as three categories: Non-cooperative
spectrum sensing, cooperative spectrum sensing and
interference temperature spectrum sensing [3]. In the
aspect of spectrum sensing, the sensing device located at
secondary user section measures certain characteristics of
the radio waveform and then decides if a primary user is
actively using the spectrum. Among different transmitter
detection spectrum sensing techniques such as the
matched filter detection (coherent detection through
maximization of the signal to noise ratio) and the
cyclostationary (exploitation of the inherent periodicity of
primary signals) “Energy detection” is the most popular
method addressed in literature [3].
Measuring only the received signal energy, the energy
detector is a non coherent detection device with low
implementation complexity. It doesn’t need any prior
information about the signal characteristics as in other
types of transmitter detection techniques of spectrum

Proceedings of IOE Graduate Conference, Vol. 1, Nov 2013 26

sensing. The energy detector measures the energy of the
input wave over a specific period and compares it with
the threshold for deciding the presence/absence of
spectrum usage by primary user. Absence of primary user
provides opportunity for secondary users to access the
spectrum until the former use that spectrum. Once the
primary user is active in the spectrum, the secondary user
should vacate that range of frequency.
Energy detection was first discussed in [3] which
evaluates the closed form expressions of the performance
parameters Pd(probability of Detection) and
Pf(Probability of false alarm rate) based on the scenario
of unknown signal of known amplitude transmitted over a
flat-band limited Gaussian noise channel. The result of
this work was extended in [5] which derived the closed
form expression of Pd and Pf of the signal with random
amplitude for different types of distributions. [6] studied
about the problem of Unknown signals over different
fading channels. Starting with no diversity case, it
presented the closed form of expression of system
performance when different diversity techniques are
employed. Based on the assumption that we cannot
estimate the actual noise variance of the channel, which
has a direct effect in the estimation of Pf, [7] analyze the
performance of spectrum sensing based on energy
detection using an estimated noise variance to calculate
the threshold where noise variance was estimated using a
spare channel dedicated for analyzing noise
characteristics. Performance of ED in AWGN and
different fading channels has been studied in many works
including [8, 9, 10]. These works assumed a perfect
knowledge of the noise power at the receiver, which
allows for the perfect threshold design. In that case ED
can work with arbitrarily small value of False Alarm
Probability and Miss-detection Probability (PMd), by
using sufficiently large observation time, even in low
SNR environment [11]. However, in real systems the
detector does not have a prior knowledge of the noise
level. Variation and unpredictability of the precise noise
level at the sensing device came as a critical issue, which
is also known as noise uncertainty. With the motive of
reducing the impact of noise uncertainty on the signal
detection performance of ED, several researches have
been proposed. Hybrid Spectrum Sensing Algorithms
based on the combination of ED and Feature Detection
techniques are put forwarded for the reduction of the
effect of noise variance uncertainty [13, 14]. In [11] the
fundamental bounds of signal detection in presence of
noise uncertainty are analyzed. This study showed that
there are some SNR thresholds under noise uncertainty
known as SNR Wall that prevents achieving the desired
performance even if the detection interval is made
infinitely large. It concludes that the robustness of any
detector can be quantified in terms of the SNR Wall
giving the threshold below which weak signals cannot be
detected reliably no matter how many samples are taken.
In [12] author performed the asymptotic analysis of the
Proceedings of IOE Graduate Conference, Vol. 1, Nov 2013
estimated noise power (ENP) to derive the condition of
SNR Wall phenomenon which suggested that the SNR
Wall can be avoided if the variance of the noise power
estimator can be reduced while the observation time
increases. [14] proposed an uniform noise power
distribution model for the noise uncertainty study of ED
in low SNR regime.
This paper formulates the detection performance of
Hybrid approach of ED which is named as Hybrid
Energy Detection (HED) and Hybrid Energy Detection 2
(HED2) to optimize the performance of Energy detection
techniques with high probability of detection.
2. SYSTEM MODELING
In our system model, we consider a single sensor Energy
Detector which senses and decides the presence or
absence of the primary signal within a defined spectrum
band W. In a given sensing time interval T, the Energy
Detector calculates its detection statistic taking N
samples of received signal . Let
(1)
be the received vector at an arbitrary sensing
interval T, where the element is the discrete
baseband complex sample at the receiver at time . Now
the spectrum sensing problem using Energy Detection
can be viewed as a binary detection problem. The
decision of the Energy Detector is the test of the
following hypothesis.
(2)
Where,
 n = 1, 2, 3…N, represents the samples (detection
period).
 is the received signal vector at the sensor and is
a channel vector which is assumed to be constant and
memory less within the sampling interval,
 is the primary signal vector,
, which is assumed
to be Complex Gaussian Distributed signal with zero
mean and variance having flat band limited power
spectrum density PSD:
 is noise vector,
also assumed to be
Complex Gaussian Distributed Noise signal with
zero mean and variance : .
[Note: is the notation for Complex Normal
Distribution Random Variable and is the Real Normal
Distribution Random Variable]
Using the information of the received signal vector to
develop a test statistic , which is the measure of the
average energy of the received signal over a sensing
27
interval N, the detector compares against a (10)
predefined threshold . If then it decides in favor
of Null Hypothesis otherwise in favor of Alternate
Hypothesis . The average energy of the received signal B. Hybrid Energy Detection
vector normalized by the noise variance can be
represented as, In HED, Noise Variance is estimated on L auxiliary noise
only slots and supposing the noise variance is constant
(3) over the adjacent slots, we perform independent noise
estimation in auxiliary “noise only” slots where we are
sure that the primary signal is absent. Since we are sure
A. Energy Detection
that there is only noise in each slots, noise variance can
In energy detection we compare the calculated decision be estimated by using all the samples. Consider a
statistic of the sample with the predefined threshold. The sampling window of length M before and adjacent to the
detection statistic is obtained by integrating the energy of detection window which is containing only the noise
the signal over certain time interval T. samples for sure. Then the estimated noise variance from
the noise only samples using a Maximum Likelihood
noise power estimate can be written as,

Figure 1: Energy based detector
For Null Hypothesis rearranging the above equation (3)
using , we get,
(4)
The decision statistic for null hypothesis can be
summarized as Chi Square Distribution with 2N degrees
of freedom scaled by the factor . Now the Energy Detection Test statistics using equation
(5)
Similarly for Alternate Hypothesis, considering the
channel coefficient as a constant value rather than a
vector and rearranging the equation (3) using
, we summarize the decision statistics as,
(6)
A numerical study shows that Energy Detection ROC
curve based on chi-squared distributions may be
accurately represented by binomial receiver operating
characteristics (ROC) curves. This allows the detector
accuracy and the ROC shape assymetry to be expressed
simply in terms of distribution parameters[5]. The
probabilities of false alarm Pf and probability of detection
Pd for a given threshold energy is given by,
(7)
(8)
(11)
denotes that the noise only samples are adjacent to the
detection window. If the estimated variance is constant,
the estimation can be averaged over S successive noise-
only slots. Thus, above equation (11) can be modified by
averaging over S successive noise-only slots as,
(12)
(12) becomes,
(13)
Here denotes the detection statistics for the Hybrid
Energy Detection scheme and the statistical distribution
of depends upon the particular estimation technique
considered. Moreover equation (13) can be considered as
the parametric likelihood ratio test when the signal to be
detected is assumed to be Gaussian with zero mean and
variance . Now for Null Hypothesis rearranging the
above equation (13) using and equation
(11), we get,
(14)
Summarizing the result of decision statistics , it
follows the F-Distribution for both the hypothesis with
different scaling factor.
(15)

According to the Central Limit Theorem, when N and M

Based on the statistics, Pd and Pf for Gaussian distributed
TED can be evaluated finally as a Q function given as,
(9)
are made sufficiently large, the F-Distributed Random
Variable in equation (15) converges to a Gaussian
distribution [23]. It gave two approximation models
where both transform the C-CDF of F-Distributed
Random Variable to a Q-function with different

Proceedings of IOE Graduate Conference, Vol. 1, Nov 2013 28

parameter. In figure 2, we have plotted the Mean Square
Error (MSE) of the approximation considering of (19)
distributions as a comparison criteria for varying N and
found that the approximation shows perfect result for N
greater than 40. It also shows that the MSE is nearly zero,
C. Hybrid Energy Detection-2
i.e. the Gaussian approximation is perfectly true for N >
40, which indicated that the degree of freedom greater In HED2, Noise Variance is estimated on S auxiliary
than 40 is sufficient for approximating Chi Squared noise only slots which are declared noise only by ED.
Distribution of the detection statistic to a Gaussian Supposing the noise variance is constant over the
distribution. adjacent slots, we perform independent noise estimation
in auxiliary “noise only” slots which are declared by
-4

6
x 10
ED by considering a sampling window of length M
before and adjacent to the detection window containing
5 only the noise samples. Optimum performance of Energy
detection technique can be achieved if there is prior
4
knowledge of nose samples. But in absence of that also,
Mean square error

ED can be optimized. HED2 optimize the performance of

3
ED by estimating the noise power knowledge from a
2
large number of noise samples. Also, it is difficult to
guarantee the availability of signal free samples to
1
estimate the noise variance in real time. The average
noise variance or estimated noise variance from the noise
0
0 10 20 30 40 50 60 70 80 90 100
only samples declared by ED can now be modified as,
Degree of Freedom

Figure 2: Mean Square Error of the approximation considering

as a comparison criteria
(20)
For the result in equation (15), using Approximation 2
for approximating F-Distribution Function to a Normal
Distribution Function, we get, Where and
Now the Energy Detection Test Statistic in equation (3)
becomes,
(16)
(21)

Here denotes the detection statistics for the Hybrid

For Gaussian distributed following above
approximation based on the statistics of , can be
evaluated as,
Where is the Q-function which is also known as the
tail probability of the Standard Normal Distribution.
Similarly, for the same threshold level
probability of detection is given by,
Following the same line of reasoning, we get the
expression of as shown below.
Energy Detection-2 scheme and the statistical distribution
of depends upon the particular estimation technique
considered. After applying the equation in case of Null
and Alternate hypothesis, we obtain the sum of square of
Standard normal random variable following a Chi Square
Distribution with degree of freedom equal to the
(17) summation order.
Noting the result we get,
the (22)
(18)
For large N and M, equation (22) with Chi Square
Distributions in numerator and denominators can be
approximated with their Normal approximates given by
the approximation formula . Thus,
simplifying the expression of for Null hypothesis in
equation (22) using the normal approximates, we get,

Proceedings of IOE Graduate Conference, Vol. 1, Nov 2013 29


Based on the statistics of
, can be evaluated as,
ranges from 1 to 10). It can be realized from Figure 6.4
that the analytical and the numerical curves are perfectly
matching which validates the analytical expressions.
(23) Also, it can be noted that, the increase in number of slots
for noise variance estimation correspondingly increases
the performance of HED and approaches closer to the
optimal one (ED with known noise variance). Under the
considered scenario with M = N = 400 which is the
number of noise samples in each slot, just S = 10(i.e.,
(24) 4000 samples) provides very near convergence to ideal
performance. For each numerical curve, its analytical
shown in equation(23,24) counterpart is superimposed to evaluate the accuracy of
the model for different values of S.

Pd versus Pf for SNR = -10 dB and N = 20, 50, 100 and 200
1

0.9
After certain mathematical procesiing using equation
0.8
(24), we get
0.7

Detection probability Pd
(25) 0.6

0.5

0.4
Similarly, for the same threshold level the
0.3
probability of detection is given by,
0.2 N= 20
N= 50
0.1 N= 100
N= 200
Following the same line of reasoning and using 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
equation(24), we get the expression of as shown False alram Probability Pf

below.
Figure 3: ROC plot of Energy Detection with SNR = -10dB for varying
number of samples
(26)
N = 400, SNR = -10dB
1

3. RESULTS AND DISCUSSION

0.8

This section implements the simulation of Energy

Detection in single sensor environment. Signal, channel 0.6
ED (Known variance)
PD

and the noise environment is set in such a way it matches

HED Analytical (S = 1)
the scenario explained in previous sections. 0.4
HED Numerical (S = 1)
Figure.3 illustrates the ROC plot of single sensor Energy HED Analytical (S = 2)
Detector with its detection statistic following Normal HED Numerical (S = 2)
0.2
distribution. Analytical result of ROC for Energy HED Analytical (S = 10)
Detection computed for SNR = -10dB, number of sensors HED Numerical (S = 10)
K = 1 and N=20, 50, 100 and 200 respectively where N 0
0 0.2 0.4 0.6 0.8 1
represents the number of samples taken to calculate the PFa
decision statistics. This figure shows high detection
probability with higher number of samples (i.e. N=200) Figure 4: ROC curve of Single Sensor Hybrid Energy Detection
for N = 100, M = 100, S = {1, 3, 5 and 10}, ,
taken to calculate decision statistics and the proper
, SNR = -10dB
detection rate decreases simultaneously with lower
number of samples considered. For given probability of false alarm PFa = 0.05 and
considered parameter (K = 1;N = 50;M = 50), the
The accuracy of the closed-form expression of the performance of Hybrid Energy Detection is also
theoretical formula is compared against simulated evaluated in terms of probability of detection against
detection performance over S auxiliary noise only slots (S different values of SNR as shown in figure 5. This

Proceedings of IOE Graduate Conference, Vol. 1, Nov 2013 30

validates our result showing that the increase in number used for the estimation of the noise variance, HED and
of slots for noise variance estimation correspondingly HED2 curves approximate the ED with known variance.
increases the performance of HED. The probability of
detection is increased with increase in number of slots 4. CONCLUSION
and plot for S=10 almost coincide with the optimal one
(known noise variance) Different methods of existing spectrum sensing were
studied and the performance of different channels is
PFa = 0.05 analysed in terms of Receiver Operating Characteristic
1
(ROC) curves. Objective of every spectrum sensing
0.9
scheme is to find out the detection statistic which can be
0.8 used in the decision making by comparing the detection
0.7
statistic with the threshold value. In context of Energy
Probability of Detection PD

Detector, the detection statistic can be obtained by

0.6
integrating the energy of the signal over certain time
0.5 interval T. The analysis of semi-blind spectrum sensing
0.4 algorithms, especially, ED is carried out in context to
CRN. The analysis is then extended to hybrid approaches
0.3
of ED with analytical expressions for the performance
0.2 S=1 parameters, PD and PFa is derived for each algorithms.
S=2
0.1 S = 10 Impact of noise variance estimation on ED was carried
0
out based on ROC curves and Probability of detection vs
-16 -14 -12 -10 -8 -6
SNR (dB)
-4 -2 0 2 4
SNR curves. The results showed that the effect of
fluctuation of noise variance estimate from nominal value
Figure 5: Probability of detection vs SNR for Hybrid Energy is severe in case of small number of auxiliary slots used
Detection for the estimation of noise variance. High detection
probability is attained with higher number of samples
Mis-Detection Probability (Pm)

Mis-Detection Probability (Pm)

1 1
HED2 Analytical HED2 Analytical taken to calculate decision statistics and the proper
0.8 0.8
HED Analytical HED Analytical detection rate decreases simultaneously with lower
0.6 ED Analytical 0.6 ED Analytical
number of samples considered to calculate the decision
0.4 0.4 statistics. Increase in SNR also increases the detection
0.2 0.2 probability but it is not always possible to have higher
0 0 value of SNR in real scenario. Even at low SNR, the
-10 -5 0 5 10 -10 -5 0 5 10
SNR in dB SNR in dB
performance of ED, HED and HED2 is better in regards
to probability of detection if higher number of samples is
Mis-Detection Probability (Pm)

Mis-Detection Probability (Pm)

1 1
HED2 Analytical HED2 Analytical
considered to calculate the decision statistic.
0.8 0.8
HED Analytical HED Analytical
0.6 ED Analytical 0.6 ED Analytical
REFERENCES
0.4 0.4
[1] I.F. Akyildiz, W.Y Lee, C.V Mehmet, M. Shantidev , Broadband
0.2 0.2 and Wireless Networking Laboratory, School of
0 0
Electrical and Computer Engineering, Georgia Institute of
-10 -5 0 5 10 -10 -5 0 5 10 Technology, Atlanta, GA 30332, United States, “Next
SNR in dB SNR in dB generation/dynamic spectrum access/cognitive radio wireless
Figure 6: plot of Single Sensor Hybrid Energy networks: A survey” Received 2 January 2006; accepted 2 May
Detection-2 for Gaussian Approximation of the Decision 2006
Statistics for N = 10, S = (2, 10, 20, 50), M = 10, SNR = -10dB [2] J. Mitola & G.Q. Maguire, Royal Institute of Technology
to 10 dB, and false alarm probability pf = 0.1. “Cognitive Radio: Making Software Radios More Personal”,
IEEE Personal Communication, August 1999
Figure 6 illustrates the simulation of Single Sensor HED2
ROC curves when the noise variance is independently [3] H.Urkowitz, Senior Member, IEEE “Energy Detection of
Unknown Deterministic Signals”, Proceedings of IEEE, Vol. 55,
estimated by applying the obtained equation over S No 4, April 1967
auxillary only noise slots determined by ED and the
[4] C.Selami & M. Torlak, Dept. of Electrical Engineering, University
estimate is used in equation (3) for all other slots of Texas at Dallas,”A Comparison of Energy Detectability Models
recursively. For generating the HED2 performance curve, for Spectrum Sensing” , 2008, IEEE
ED parameters within HED2 are, N = 10, pd_ed = 0.5,
[5] V.I Kostylev, Voronezh State University, Universitetskaya,”
assuming signal probability 0.5 and noise probability 0.5. Energy Detection of a Signal of Random Amplitude”, 2002 IEEE
Since, there is a chance of misdetection in case of HED2,
performance of HED2 is slightly lower than HED but still [6] F.F. Digham, M.S Alouni and M.K. Simon” On the Energy
Detection of Unknown Signals over Fading Channels”,
no visible difference can be noted in extreme high or low Department of Electrical and Computer Engineering, University of
SNR values. With the increase of the number of slots Minnesota

Proceedings of IOE Graduate Conference, Vol. 1, Nov 2013 31

[7] Y. Zhuan, M. Gokhan, G.John , Wireless Solution Research
Center, Motorola Labs, Schaumburg, II, USA Department of
EECS, Northwestern University, Evanston, II, USA,” Energy
Detection using Estimated Noise Variance for Spectrum Sensing
in Cognitive Radio Networks”, 2008 IEEE
[8] V.I Kostylev, “Energy detection of a signal with random
amplitude,” Communications, 2002. ICC 2002. IEEE International
Conference on , vol.3, no.pp.1606,1610 vol.3, 2002 doi:
10.1109/ICC.2002.997120
[9] A. Mariani; A. Giorgetti, M. Chiani, “SNR Wall for Energy
Detection with Noise Power Estimation,” Communications (ICC),
2011 IEEE International Conference on , vol., no., pp.1,6, 5-9
June 2011 doi: 10.1109/icc.2011.5963367
[10] Kalamkar, S. Sanket, Banerjee, Adrish, “On the performance of
generalized energy detector under noise uncertainty in cognitive
radio,” Communications (NCC), 2013 National Conference on ,
vol., no., pp.1,5, 15-17 Feb. 2013 doi:
10.1109/NCC.2013.6487927
Proceedings of IOE Graduate Conference, Vol. 1, Nov 2013
[11] Y. Zeng, Y.C. Liang, A.T Hoang, Y. Peh, “Reliability of Spectrum
Sensing Under Noise and Interference Uncertainty,”
Communications Workshop, 2009. ICC Workshops 2009. IEEE
International Conference on, vol., no., pp.1,5, 14-18 June 2009
doi: 10.1109/ICCW .2009.5208033
[12] V. Valenta, R. Maršálek, G. Baudoin, M. Villegas, M. Suarez, and
F. Robert, "Survey on spectrum utilization in Europe:
Measurements, analyses and observations," in Fifth International
Conference on Cognitive Radio Oriented Wireless Networks &
Communications (CROWNCOM 2010), 2010
[13] Z. Khalaf, A. Nafkha, J. Palicot , “Enhanced hybrid spectrum
sensing architecture for cognitive radio equipment,” General
Assembly and Scientific Symposium, 2011 XXXth URSI , vol.,
no., pp.1,4, 13-20 Aug. 2011 doi:
10.1109/URSIGASS.2011.6050528
[14] F. Moghimi, R. Schober, R.K Mallik, “Hybrid Coherent/Energy
Detection for Cognitive Radio Networks,” Wireless
Communications, IEEE Transactions, vol.10, no.5, pp.1594,1605,
May 2011 doi:10.1109/TWC.2011.030411.100973
32