Improved Energy Detector Based Spectrum Sensing for Cognitive

Radio: An Experimental Study

Prachetos Sadhukhan
Department of Electrical Engineering
Indian Institute of Technology - Delhi
Hauz Khas, New Delhi 110016, India
E-mail: [email protected]
Naveen Kumar
Department of Electrical Engineering
Indian Institute of Technology - Delhi
Hauz Khas, New Delhi 110016, India
E-mail: [email protected]
Manav R. Bhatnagar
Department of Electrical Engineering
Indian Institute of Technology - Delhi
Hauz Khas, New Delhi 110016, India
E-mail: [email protected]
Abstract Cognitive radio (CR) is a technology
that addresses the problem of scarcity of spec-
trum and paves the way for ecient use of the
same. In recent times, it has been observed that
most research is focused on building theoretical
models which are backed mostly by simulation re-
sults. In order to successfully support any theo-
retically developed model, it is necessary that the
algorithm be tested in practical environment us-
ing hardware. Since some assumptions made to
satisfy simulation requirements may not hold un-
der practical circumstances, the implementation
and testing of a theoretical concept is quite chal-
lenging. In this paper, we present experimental
test results from our implementation of improved
energy detector for single node and co-operative
CR networks. The theoretical model is taken
from [1], while GNU Radio and universal soft-
ware radio peripherals 2 (USRP2) kits serve as
software and hardware, respectively, in the imple-
mentation. Shadowed fading (due to obstacles and
hidden node problem) makes single node sensing
sometimes unreliable. Use of cooperation among
secondary nodes, along with proper fusion rules,
not only exploits the advantage of spatial diversity
but also improves the reliability of spectrum hole
sensing, as evident from the test results. It has
been demonstrated by using the practical testing
that cooperative sensing is better than single node
sensing.
Keywords Cognitive Radio (CR), GNU Radio, im-
proved energy detector, primary user (PU), spec-
trum sensing, USRP2.
I. Introduction
In recent times, there has been an explosive growth in
wireless devices and applications, and they have created
a huge demand for bandwidth consuming, quality content
apart from other services that require spectrum. Most
easily usable spectrum bands have been allocated and
through studies, it has also been shown that most of
the time, the licensed spectrum bands are not fully uti-
lized, i.e., they may remain idle for substantial periods of
time. The demand pressure along with the underutiliza-
tion of unallocated spectrum has forced wireless industry
to look for technology that will enable intelligent access
to licensed spectrum without disrupting primary licensed
user. Cognitive radio (CR) oers us the promise of solu-
tion to all these problems. The CRs are wireless devices
those are fully programmable, can sense their environ-
ment, and can adapt their transmission waveform, chan-
nel access method, spectrum use, and networking pro-
tocols to oer the best possible network and application
performance [2]. By sensing of spectrum holes, the CRs
can use licensed spectrum bands not in use at any point in
time, frequency or space [3]. This ability can help to de-
liver optimal quality-of-experience (QoE) to users, with-
out sacricing quality of content or speed of applications.
There are plenty algorithms/solutions that show sec-
ondary users can exist harmoniously with primary users,
but the regulators concern cannot be waived away by the-
oretical frameworks alone. In CR, some major challenges
are how to maximize spectrum utilization while reducing
incidences of false alarm (constraint on false alarm prob-
ability [4]), how to keep the primary and secondary from
deteriorating each others performance and assured Qual-
ity of Service (QoS) of the whole network, and how to
detect signals in low Signal to Noise Ratio (SNR) condi-
tions (constraint on probability of missed detection [5]).
This calls for rigorous studies of signal pattern, character,
and test performance and benchmarking of algorithms in
real world environment, to show that the claims made
theoretically hold in practical conditions. Co-operative
sensing is usually preferred over single node sensing be-
cause of better immunity to fading, noise uncertainty, and
shadowing [3].
In this paper, we explain software dened radio (SDR)
based implementation of improved energy detection al-
gorithm, which is theoretically proposed in [1] for single
node spectrum sensing. We use this algorithm in a col-
laborative CR network and use suitable fusion rules at
the fusion center. It is shown by the experimental re-
sults that the improved energy detection scheme signi-
cantly outperforms the classical energy detection scheme.
These experimental results corroborate the theoretical re-
sults proposed in [1].
978-1-4799-2275-8/13/$31.00 2013 IEEE
2013 Annual IEEE India Conference (INDICON)
II. System Model
A. Basic Hypothesis and System Model Analysis
Spectrum sensing can be represented and analyzed as a
binary hypothesis testing problem with
H
0
: y[n] = w[n],
H
1
: y[n] = x[n] +w[n], (1)
where H
0
represents the hypothesis that the signal is
absent, H
1
represents the hypothesis that the signal is
present, y[n] represents the received signal, n = 1, 2, .., N,
where N is the observation interval; w[n] represents addi-
tive noise, and x[n] represents primary transmitters sig-
nal. Practical implementation of energy detector based
spectrum sensing CR depends neither on the nature of
w[n] nor on the nature of x[n], but for theoretical simu-
lation and analysis sake, the assumption that w[n] is ad-
ditive white Gaussian noise (AWGN), needs to be made
for the purpose of simplicity.
In conventional/classical energy detector, the energy of
the received signal is compared with a threshold to decide
the absence/presence of Primary User (PU):
=
1
N
N

i=1
|
y[n]

w
|
2
H
1

>

ED
H
0
, (2)
where is the test statistic,
ED
is the threshold, and

w
2
is the variance of the additive noise. Taking cue from
Chen [1], to improve the performance of the spectrum hole
detection, we next discuss the model of improved energy
detector. The improved energy detector is represented as
follows:
=
1
N
N

i=1
|
y[n]

w
|
p
H
1

>

IED
H
0
, (3)
where p is a positive rational number and
IED
is a con-
stant threshold value. The probability of detection can
be stated, as
P
d
= P( > /H
1
) (4)
and probability of false alarm, as
P
f
= P( /H
1
). (5)
As only one threshold is used, we can compute probability
of missed detection as
P
m
= 1 P
d
. (6)
Using the above two formulas, we can write the expression
for total probability of error as
P
e
= P
m
+P
f
. (7)
The receiver operating characteristics (ROC) of CR node
give a visual representation of improved energy detector
Figure 1: Demonstration of how fusion center assimilates de-
cision data received from secondary receivers.
with varying p. But to quantify and optimize the per-
formance, total probability of error and area under ROC
curve can be useful. Our experiments are focused on nd-
ing optimal value(s) of p and
IED
, by plotting the total
error probability versus p and total error probability ver-
sus
IED
plots. The curves obtained are shown in Fig. 4
and Fig. 5, respectively. Using the optimal value of p
and
IED
, we implement and analyze the performance of
improved energy detector, in case of co-operative commu-
nication, for a collection of three and four CRs using OR,
AND, and majority based binary decision rules.
B. System Model for Decision at Fusion Center
In our implementation of co-operative sensing, the CR
nodes have been kept spatially distributed with the fu-
sion center located centrally. Each CR node senses the
spectrum and takes a binary decision which is indepen-
dent of the other CR nodes. The local binary decision of
each CR node is then conveyed to the fusion center which
takes the nal binary decision to determine whether the
channel is busy or idle, i.e., whether the primary user is
present or absent. The fusion center takes the decision ac-
cording to OR, AND, and majority fusion rules. In Fig. 1,
a CR network is shown showing how fusion center assim-
ilates decision data received from secondary receivers.
III. System Overview and Implementation
Background
In this experimental setup, we need to develop a sys-
tem for testing classical energy detection and improved
energy detection, both for single node and co-operative
network. We chose GNU Radio as software, Universal
software radio peripherals 2 (USRP2) kits as hardware,
Lenovo ThinkPad laptops as host computers, and Ubuntu
12.04 LTS as operating system. Detailed explanations
have been given in the following sections. Using the latest
version of each of the above mentioned software, smooth
working of the equipment can be insured, and at the same
time the updated library of signal processing blocks is
an added advantage. Fig. 2 shows the block diagram of
USRP2 interfaced with host computer for the purpose of
Figure 2: Block Diagram of architecture of setup: A host
computer running Ubuntu 12.04 Operating System, USRP,
and RFX2400 daughterboard (Operating range 2.3 GHz to
2.9 GHz).
spectrum sensing. In our experiments, we use four USRP2
kits with one daughterboard onboard each.
A. GNU Radio
GNU Radio is a free and open source software develop-
ment toolkit that is commonly used to build software de-
ned radio implementations [6].Using readily available RF
hardware in conjunction with GNU Radio, a wide variety
of wireless devices and algorithms can be implemented
and tested. GNU Radio also supports development of
signal processing algorithms on recorded signals, avoiding
need for actual RF hardware. In GNU Radio, non-critical
applications such as graph construction and user interface
are written in python, while the performance-critical, time
consuming signal processing path is written in C++ using
processor oating point extensions. Thus developer can
construct codes for application in user friendly language
python, and at the same time write optimized custom
signal processing blocks in C++. Applications written
in python later interconnects signal processing blocks us-
ing paths, both written in C++, and they operate in real
time, paving way for high throughput real-time radio sys-
tems. Recently, GNU Radio Companion (GRC) has been
added, which is GUI for designing ow graphs and con-
necting signal processing blocks. With GRC whole appli-
cations/radio systems can be made without even writing
one line of code, thus making it easy for those who nd
coding tedious. GNU Radio supports a wide range of
RF hardware platform and thus allows us to access wide
range of spectrum band. Among them USRP is the most
used platform (described in detail in the next section).
Our implementation setup used GNU Radio 3.6.2 run-
ning on Ubuntu 12.04 LTS operating system. The above
block diagram quite accurately depicts a software dened
radio setup [7]. The antenna of the transceiver daughter-
board picks up the signal, which passes through the subse-
quent modules and nally gets fed into the host computer.
The blocks are implemented independently and then con-
nected together in the main code. Data ows along the
arrows shown, and their format can be specied accord-
Figure 3: Actual test bed set-up in lab.
ing to need or according to the signal processing block
accepting it as input.
B. USRP2
USRP are computer hosted SDR manufactured by Ettus
Research, LLC, and its parent company National Instru-
ments [8]. A large range of kits are available with dierent
features and connectivity. We use USRP2 model kits for
our experiments. It has a Xilinx Spartan 3-2000 FPGA
motherboard for signal processing, extension sockets for
connecting daughterboard and a Gigabit Ethernet inter-
face for connection to host computer. The signal pro-
cessing motherboard contains analog-to-digital convert-
ers (ADC) and digital-to-analog converters (DAC) which
help in converting signal from continuous to discrete do-
main at receiver, and vice versa at transmitter respec-
tively. The data (signal samples) are transmitted from
kit to host computer through high speed Gigabit Ethernet
connector. The wideband ADC can capture all the chan-
nels supported by the daughterboard. There are two 100
Mega-sample per second (MS/s), 14 bit ADCs (LTC2284)
with limits of 72.4 dB SNR, and 85 dB SFDR for signals
at Nyquist frequency. There are two 400 MS/s, 16 bit,
DACs (AD9777) with limits of 160 MS/S w/o interpo-
lation, and up to 400 MS/s with 8x interpolation [9]. It
also has a SD-Card based bootloader system. Though the
product has been discontinued by the parent company, the
latest versions of GNU Radio and USRP Hardware Driver
(UHD) support USRP2; RFX2400 was used as the daugh-
terboard. It supports operation in the range of 2.3 GHz to
2.9 GHz, and hence is most suited for carrying out exper-
iments in the 2.4 GHz free band. There are limits on gain
at receiver and transmitter as well as maximum power
of transmitted wave (20 mW). USRP2 is superior com-
pared to USRP1, but there are still bottlenecks in maxi-
mum throughput. USRP2 kits interact with GNU Radio
or host computer via UHD; UHD provides built in com-
mands to perform simple operations on received signals,
e.g. open a GUI based spectrum analyzer by using uhd t
command and record in phase and quadrature phase val-
ues of the received signal using uhd rx cle command.
UHD can perform basic functions, but the main power of
signal processing lies with GNU Radio. Gigabit Ethernet
0 1 2 3 4 5 6 7 8 9 10
10
3
10
2
10
1
10
0
p
T
o
t
a
l

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
threshold = 0.8
threshold = 0.9
threshold = 1.0
threshold = 1.1
threshold = 1.2
threshold = 1.3
threshold = 1.4
Figure 4: Total error probability versus p for dierent values
of threshold (
IED
).
interface allows USRP2 kits to be connected via Gigabit
Ethernet network switch to host computer. They have
to be supplied with their own IP addresses, and hence
can be detected over a network They have an additional
connector called multiple-input multiple-output (MIMO)
connector, which allows two USRP2 kits to be connected
together and share the same Ethernet interface to host
computer. It performs synchronization between the two
USRP2 kits, and can be used for the purpose of cooper-
ative communication. Benchmarking can be done using
scripts supplied with GNU Radio, to observe the maxi-
mum symbol transfer rate that can be practically attained
by USRP2 kit. A large number of sample scripts are sup-
plied with GNU Radio, which can be extended or used
in their default form, .e.g. SNR, BER measurement over
a noisy channel can be done using digital bert tx.py to
specify symbol rate, spectrum band and modulation, and
digital bert rx.py to demodulate received signal and cal-
culate necessary parameters. UHD is quite stable, while
the hardware kits might be prone to some error due to
thermal noise, voltage uctuations, etc.
C. Ubuntu 12.04 LTS
The preferred OS for development of open source soft-
ware is Linux and GNU Radio and UHD are no excep-
tion. Ubuntu is one of the most well maintained, updated
avor of Linux, and GNU Radio Project community has
released installers and images for the long term support
version of the latest Ubuntu. Although GNU Radio and
UHD are available for windows OS, they are not quite
stable. Matlab and Simulink can also be used to process
live or recorded signal in windows. Though GNU Radio
and UHD are available for older versions of Ubuntu, the
latest is recommended to avail the benet of bug xes and
more ecient codes.
IV. Experimental Results
Our experiment is summarized as follows:
Step1: Collect the samples y[n] of the signal using
USRP2 kits under hypothesis H
0
and H
1
.
0 0.5 1 1.5 2 2.5 3
10
3
10
2
10
1
10
0
Threshold
T
o
t
a
l

E
r
r
o
r

P
r
o
b
a
b
i
l
i
t
y
p = 0.1
p = 0.2
p = 0.3
p = 0.4
p = 0.5
p = 0.6
p = 0.7
p = 0.8
p = 0.9
Figure 5: Total error probability versus threshold (
IED
) for
dierent values p.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of False Alarm
P
r
o
b
a
b
i
l
i
t
y

o
f

D
e
t
e
c
t
i
o
n
p = 2.0
p = 0.7
Figure 6: ROC curves for p = 2 (conventional energy detec-
tion) and p = 0.7 (optimal value).
Step2: Estimate the signal power from the samples col-
lected under hypothesis H
0
. This provides additive noise
variance (
2
w
).
Step3: Calculate the test statistic , given in (3).
Step4: Compare with predened threshold values
(
IED
) and calculate the value of P
d
, P
f
, P
m
, and P
e
as dened in equation (4), (5), (6), and (7), respectively.
We have plotted total error probability versus p in Fig. 4.
This gure shows that as the value of p increases, to-
tal error probability rst decreases, reaches a minimum
value (global minima), and then increases. As we further
increase p, the total error probability decreases, reaches
another minimum value (local minima) and then increases
again. Therefore, from Fig. 4, we conclude that the total
error probability is minimum for p = 0.3 (
IED
= 1.1), p
= 0.5 (
IED
= 1.2) and p = 0.7 (
IED
= 1.3), i.e., spec-
trum hole detection performance is maximum for these
values. There is another minima around p = 7 but at
this point the total error probability is greater than the
previous minimum value. For this minima, the value of
p varies from 6.5 to 7.5 depending on the threshold value
chosen. Fig. 4 shows the variation of total error probabil-
ity with threshold for dierent values of p. This plot con-
rms that the optimal values are p = 0.3 (
IED
= 1.1),
p = 0.5 (
IED
= 1.2) and p = 0.7 (
IED
= 1.3). In
Fig. 5, we compare the ROC curves for conventional en-
0 0.1 0.2 0.3 0.4 0.5 0.6
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
Probability of False Alarm
P
r
o
b
a
b
i
l
i
t
y

o
f

D
e
t
e
c
t
i
o
n
OR
AND
Majority
Figure 7: ROC curves for n = 3 according to OR, AND, and
majority decision rules.
ergy detection (p = 2) and improved energy detection (p
= 0.7). These plots clearly show that the improved en-
ergy detection method outperforms the conventional en-
ergy detection method. Figs. 7 and 8 are ROC curves
for cooperative spectrum sensing using OR, AND, and
majority binary decision rules. Fig. 7 shows that for
three CR nodes OR binary decision rule gives the best
performance. Fig. 8 shows that for four CR nodes the
performance for OR and majority binary decision rules is
almost equal and better than that of AND binary decision
rule.
V. Conclusion
In this experiment, we studied spectrum sensing with im-
prove energy detection to nd the optimal value(s) of p for
CR. It is established by using the experiments that the im-
proved energy detection scheme outperforms the classical
energy detection scheme and there exists optimal value(s)
of p for which the performance is maximum. To nd these
values we perform signicant amount of experiments for
dierent values of p and detection thresholds, and con-
clude that the optimal values for the pair (p,
IED
) are
(p = 0.3,
IED
= 1.1), (p = 0.5,
IED
= 1.2), and (p =
0.7,
IED
= 1.3). The total error probability for these
optimal values is approximately equals to 2 10
3
. It is
also demonstrated by experiments that the performance
of cooperative spectrum sensing scheme depends on the
number CR nodes present in the secondary system. Fur-
ther, we also conclude that OR and majority binary de-
cision rules are better than the AND binary decision rule
for the CR networks considered in the experiments.
References
[1] Y. Chen, Improved energy detector for random signals in Gaus-
sian noise, IEEE Trans. Wireless Commun., vol. 9, no. 2, pp.
558-563, Feb. 2010.
[2] P. Steenkiste, D. Sicker, G. Minden, and D. Raychaudhuri, Fu-
ture directions in cognitive radio network research, in NSF
Workshop Report, 9-10 March, 2009.
[3] G. Staple and K. Werbach, The end of spectrum scarcity,
IEEE Spectrum, vol. 41, no. 3, pp. 4852, Mar. 2004.
0 0.1 0.2 0.3 0.4 0.5 0.6
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
Probability of False Alarm
P
r
o
b
a
b
i
l
i
t
y

o
f

D
e
t
e
c
t
i
o
n
OR
AND
Majority
Figure 8: ROC curves for n = 4 according to OR, AND, and
majority decision rules.
[4] S. Haykin, Cognitive radio: Brain-empowered wireless com-
munications, IEEE J.Select. Areas of communications, vol.23,
no.2, pp. 201-220, Feb. 2005.
[5] R. Tandra and A. Sahai, SNR walls for signal detection, IEEE
J. Sel. Topics Sig. Proc., vol. 2, pp. 4-17, Feb. 2008.
[6] The GNU Software Radio Project. Further information is avail-
able at http://gnuradio.org/redmine/projects/gnuradio/wiki.
[7] K. Kim, Y. Xin, and S. Rangarajan, Energy detection based
spectrum sensing for cognitive radio: An experimental Study,
in Proc. IEEE Globecom, Miami, Florida, 6-10 December, 2010.
[8] Universal Software Radio Peripheral (USRP) by Ettus Research
LLC. Further information is available at http://www.ettus.
com/.
[9] Universal Software Radio Peripheral (USRP) wiki. Further in-
formation is available at http://en.wikipedia.org/wiki/Universal
Software Radio Peripheral