See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/308277394

Cognitive Radio Testbeds: State of the Art and an Implementation

Chapter · September 2017

DOI: 10.1007/978-981-10-2254-8_7

CITATIONS READS
5 1,300

3 authors:

Selahattin Gökceli Gunes Karabulut Kurt

Tampere University Polytechnique Montréal
31 PUBLICATIONS   225 CITATIONS    332 PUBLICATIONS   1,530 CITATIONS   

SEE PROFILE SEE PROFILE

Emin Anarim
Bogazici University
239 PUBLICATIONS   1,270 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

User Keyboard Usage Behavior Analysis View project

Causality and Extreme Value Based DDOS Attack Detection in SDN Environmentment View project

All content following this page was uploaded by Emin Anarim on 09 November 2017.

The user has requested enhancement of the downloaded file.

 1

Cognitive Radio Testbeds: State of the Art and

an Implementation
Selahattin Gökceli, Güneş Karabulut Kurt and Emin Anarım

Abstract

Cognitive radio (CR) technology is one of the most popular techniques in the telecommunications field due to
its potential benefits on the solution of spectrum scarcity problem. According to principal of the CR, availability of
the licensed spectrum portion is monitored by the spectrum sensing process and strategies are applied to use this
portion without interfering the primary user (PU). CR provides a flexible system that secondary user (SU) can make
decisions about the spectrum at any time by configuring correspondent parameters easily. However, implementation
of CR systems cannot be done easily in most cases due to characteristic difficulties of the wireless channel such
as fading. Especially spectrum sensing process has these difficulties which should be overcome in order to create
a successful CR system. Robust spectrum sensing is essential for a CR system due to its vital role in the efficient
usage of the spectrum. Therefore, algorithms that are proposed for this issue should contain suitable properties for
realistic environments. Software defined radio (SDR) is a core component of the CR technology and it allows a
practical development process with modification of the software rather than hardware. Thus, SDR based approaches
to CR problems are quite effective and in recent term, SDR based studies of the CR are very wide. In this chapter,
state of the art of CR studies are explained in detail by demonstrating essential details of these studies and effective
approaches of the implementation of these SDR studies are mentioned. Moreover, an energy detection based spectrum
sensing implementation for 2.4 GHz ISM band is implemented by using SDR tools LabVIEW and NI USRP-2921
hardware in real-time. In order to help future studies and give insight about such implementations, implementation
details of our system which is very practical to regenerate, are explained in broad detail.

Index Terms

Software defined radio, cognitive radio, spectrum sensing, energy detection, wireless channel.

I. I NTRODUCTION

Capacity of the communication resources is decreasing significantly particularly due to rapid development of
the mobile communication technologies and related increase of the expected service quality. Licensed spectrum is
very limited in various countries and there are not too many alternatives of these resources. Most problematic issue
for authorities who govern communication services is management of current frequency spectrum and if increase
trend of the mobile devices will continue, this issue will be more problematic for deployment of next generation

Selahattin Gökceli, and Güneş Karabulut Kurt are with Wireless Communications Research Laboratory, Department of Electronics and
Communication Engineering, Istanbul Technical University, Turkey. Email: {gokcelis, alakoca, tedik, gkurt}@itu.edu.tr.
 2

mobile technologies. As highlighted in [1], Long Term Evolution (LTE) technology has spectral efficiency around
%20 that means it is at around full capacity in terms of Shannon capacity limit and efficiency cannot be increased
significantly. In recent term, cognitive radio (CR) technology has been used widely as an important candidate for the
management of dense spectrum. CR is a radio technology that can sense spectral activity even in the wideband and
use obtained information strategically in wireless networks in order to provide necessary service quality to the user
[2]. Important increase of spectrum utilization efficiency can be obtained by allocating spectrum to both primary
users (PUs) and secondary users (SUs) without harming to usage of PUs [3]. In order to provide this improvement,
SUs should be aware of the existence of PUs in the band. Spectrum sensing provides this functionality and it is
essential for the quality of CR system.

A. Importance of Spectrum Sensing for Cognitive Radio

CRNs operate over wireless networks and characteristics of wireless channels which transmissions are realized
must be understood in order to understand principals of the CRNs. Transmission channels are very important
especially for spectrum sensing. Physical transmission channel is usually defined as

y[t] = s[t] + n[t] (1)

where s(t), h(t) and n(t) represents transmitted signal, channel impulse response and noise component. According
to this model, signal is received by addition of the noise. Channel noise can be emerged from natural or artificial
sources. The main part of distortion occurs because of thermal noise and noise is usually modeled as a probabilistic
process. Parameters of this process are different for different channels. In general, noise is modeled as AWGN.
However, this simple channel model does not consider fading effect of the channel. Due to fading existence, power
of the channel takes different values at time and frequency.
In the literature, some channel models that consider fading effect have been proposed [4]. An important example
of these models is the Rayleigh channel model. Rayleigh model assumes that any direct path between transmitter
and receiver pairs does not exist, and signals are received by receiver after reflections from various reflectors.
As mentioned in [5], distorting effects of wireless channel caused by fading and shadowing, directly affect the
spectrum sensing performance and detection probability significantly decreases. Furthermore, it is explained in [6]
that presence of the PU cannot be detected and interference can be generated on licensed band due to multipath
fading existence. This kind of models are vital for CRN implementations, algorithms that are suitable to real-life
can be obtained.
As explained earlier, SDR implementation has an unique benefit on channel models compared to simulation
modeling. By using SDR platform, algorithm is evaluated in a realistic environment and effects of wireless channel
are experimented. In such experiments, cognitive radio’s essential element, spectrum sensing must perform well to
create a robust algorithm. In accordance with these details, we implemented a robust SDR base CRN system in
real-time by experimenting wireless channel results with results that are obtained by a spectrum analyzer.
 3

B. Software Defined Radio Technology

CR technology can provide various improvements on spectrum management with its software defined radio
(SDR) and intelligent signal processing (ISM) components [7]. SDR is a communication system that realizes some
processes which were processed by using some hardware in past, by using a software with the help of an embedded
system or a computer. SDR technology provides efficient usage and management of communication resources. In
classical hardware based communication systems, modifications on systems can only be applied physically and this
causes serious cost disadvantages. However, with SDR, devices can be upgraded by modification of software easily
and without adding any cost. In this way, quality of the system can be improved too. Due to requirement of usage
of protocols that changes dynamically in real-time, SDR technology has a huge potential in tactical and mobile
communication technologies. It is expected that SDR technology will be a crucial part of future communication
technologies.
In SDR, radio parameters can be defined with software in programmable processing units. Main examples of
these units are field-programmable gate array (FPGA) and digital signal processors (DSP). Implementation of
SDR technology decreases hardware requirements and provides performance improvement. As explained in study,
programming with SDR technology offers flexibility between unsuitable radio frequencies and modulation techniques
and allows combination of these techniques [8]. For example, a radio that operates at 900 MHz can communicate
with another radio which operates at 2.4 GHz. Therefore, a common communication structure can be created with
SDR technology.
SDR implements various processes before transmission; defines appropriate transmission channels, suitable mod-
ulation type to the channel, determines right transmit gain and then realizes transmission. Similarly implements
some processes before reception; can detect energy distribution in main channel and adjacent channels, defines
transmission quality, can cancel interference adaptively, decodes channel modulation and tries to fix bit errors
in accordance with desired BER level [9]. SDR devices support a wide variety of protocols and appropriate
improvements.

II. BASIC C OMPONENTS

1) Requirements of Software Defined Radio Technology and Associated Tools in the Study: SDR devices offer
various commercial benefits, moreover, usage of SDR technology has also various benefits for research and
development process of the next generation technologies. By using SDR devices, a system can be implemented and
performance measurement can be obtained before real deployment of the correspondent system. Simulation based
performance analysis is the most common method in such efforts. However, simulations are usually very insufficient
in modeling real-time algorithms. For example, channel models like additive white Gaussian noise (AWGN) do not
match with realistic conditions. Some channel models can model the environment successfully, yet some techniques
such as multiple-input and multiple-output (MIMO) cannot be modeled easily [10]. Due to advantages like flexible
system modeling, configurable system settings, easy deployment of new generation communication algorithms, SDR
usage in research efforts is more preferable over simulation based approaches.
 4

In order to exploit the benefits of the SDR technology, some requirements should be considered. Especially for
real-time SDR implementations, these three factors should be considered [10]:
1.Hardware Abstraction: If abstraction between software source and utilized hardware can be provided, user
can understand necessary technical details of the hardware without dealing system complexity and design can be
simplified. But scope of the abstraction is very important, abstraction should be provided at all layers in wide range.
2.Comprehensive Software Development Environments : SDR system development tools should only contain
heterogeneous processing components, moreover, should be capable of joint operation with different computation
algorithms and software tools. In this way, development and optimization of a system can done by using more proper
tools and methods. Another requirement about the software tool is quality of the simulation service. Simulation in
used software tool should reflect implementation environment and data processing must be rapid.
3.Heterogeneous Multiprocessing : Software tool of the SDR system should support different hardware and
libraries. Multiple tools should be considered and system should be configurable. Joint usage of different hardware
and libraries should not cause any problem.
In accordance with mentioned requirements, LabVIEW software and NI USRP-2921 hardware is used in our
implementation. Technical details of these tools are explained in following part.
LabVIEW LabVIEW is a visual programming tool that uses visual elements in programming which is a different
approach compared to text based tools such as C, C++, Java and MATLAB. Visual programming feature makes
LabVIEW easily programmable and understandable. Because of this, LabVIEW is mostly used in SDR based
studies.
Programming in LabVIEW is managed in two separate parts which are front panel and block diagram. In
block diagram, programming is done by using elements from the functions palet and connecting them with virtual
cables. Core programming structure in LabVIEW is called as virtual instrument (VI). LabVIEW contains different
function categories like programming, measurement, signal processing and data communication. With content of
these functions, a communication system can be implemented in detail. LabVIEW has three element types, control,
indicator and constant. With control elements, values of some functions are controlled by inserting value in the front
panel. Indicator is responsible from demonstrating results in front panel. Constant elements have control element
properties, but these are managed in block diagram.
NI USRP-2921: Universal Software Radio Peripheral (USRP) is a hardware that has real-time data transmission
ability. USRP devices are used in various communication studies. USRP devices ease the learning, teaching
and research processes of communication systems. Due to suitability of the USRP-2921 device to LabVIEW,
implementation of SDR systems is quite easy. USRP-2921 device operates in 2.4 GHz-2.5 GHz and 4.9 GHz-5.9
GHz frequency bands. Moreover, it has half-duplex characteristic and has good hardware features, can support
instantaneous bandwidth up to 20 MHz and I/Q sampling rate up to 25 MS/s [10].

III. S TATE OF THE A RT ON C OGNITIVE R ADIO T ESTBEDS

In this section, importance of the SDR implementations for the CR studies is reviewed with an emphasis on
spectrum sensing. Moreover, some leading studies are detailed and importance of SDR technology to such CR
 5

studies is highlighted in order to emphasize their suitablity in real-life.

Spectrum monitoring in CR systems is mostly implemented based on three sensing methods which are energy
detector based sensing, waveform based sensing and cyclostationary based sensing [11]. Energy detector method is
based on the comparison of output of the energy detector with a threshold which is determined according to noise
properties. Waveform method is based on the comparison of the received signal with a copy of the signal when
receiver knows correspondent pattern. cyclostationary method contains the sensing of the primary signal by using
some cyclostationary features. Energy detector has more usage than the other methods because of its advantages
like low computational complexity and easy implementation. Moreover, receiver does not need any information
about primary user. In literature, energy detector method is widely used in spectrum monitoring applications. In
[12], implementation advantages of the enery detection are shown with LabVIEW and NI PXI hardware. However,
study does not include wideband sensing. Similarly in [13], in order to show flexibility of the SDR technology, a
simple spectrum monitoring implementation is done by using LabVIEW and USRP hardware. However, this study
does not cover ISM band sensing. In [14], spectrum measurements of 2.4 GHz ISM band which are obtained
separately by Anritsu MS2690A spectrum analyzer and USRP2 hardware are compared without channel based
sensing. Similarly in [15], real-time spectrum monitoring is implemented in 2.4 GHz ISM band by using LabVIEW
and USRP. Channel based sensing is not included in this study.
In accordance with the 5G requirements, problems of current spectrum sensing frameworks are tackled and a
novel method is proposed for these problems in [16]. Traditional frameworks assumes that all SUs should join
to sensing process and each one should have good properties which is usually not valid in practice. Moreover,
wide sensing range and frequent detection cause higher costs and increased energy consumption which are serious
handicaps for 5G. To this end, a novel framework is proposed and spectrum sensing is implemented by spectrum
agents. This approach reduces mentioned drawbacks and provides a performance that is suitable to 5G. In [17],
due to vulnerability of using a common channel between SU transmitter and receiver, sequential channel scanning
and quorum based rendezvous methods are implemented without using a common channel with a small testbed by
using NI USRP-2921 hardware and software tools LabVIEW and MATLAB. According to model, SU pairs scan
for common channel by knowing idle channel list initially. Success of model is demonstrated and supported with
numerical results. In [18], synchronicity between PU and SU nodes is targeted and issues as well as solutions related
to asynchronous spectrum sensing are explained. As highlighted, assuming asynchronous sensing is more suitable
to realistic scenarios. Due to lack of related studies, issues about asynchronous sensing are handled. Moreover,
two sensing schemes for non-cooperative and cooperative sensing are proposed. As shown with results, proposed
asynchronous approaches perform very well. As studied in [19], distributed implementation is more desirable
than using a control channel for multi-hop setting. Therefore, a distributed cognitive radio network architecture is
modeled and full stack implementation is done in real-time by using GNU Radio and USRP N210. Moreover, a
programmable emulation testbed which includes emulated channels obtained from RFnest emulator, is developed and
detailed performance measurements are implemented. As shown with results, study proposes an effective approach
with proper performance results.
In [20], a complete CR system is designed and implemented in real-time by using GNU Radio and USRP SDR
 6

hardware. Implementation consists of spectrum sensing, spectrum management and spectrum decision processes.
Study targets spectrum scarcity problem and dynamic spectrum access (DSA) approach is utilized to provide solution.
Design model is designed as cognitive radio (CR) user which behaves like secondary user, aims to use licensed
spectrum when primary user (PU) does not use correspondent part. Hence, CR user first performs spectrum sensing
and decides whether channel is free or not. Then spectrum management is applied to manage the frequency band
with other users. Spectrum decision part is responsible for usage decisions such as when to connect, duration of
usage. In the end, data transmission is realized which is the step that licensed band is used. For spectrum sensing,
energy detection based approach is utilized.Implementation setup consists of four USRPs which are used as SU
transmitter, SU receiver, spectrum analyzer and PU transmitter. During implementation, two channels which are
determined as 2.488 GHz and 2.492 GHz are used. With experiments, two different cases are observed which are
PU and CR users use same channel or different channels. Firstly, by utilizing GMSK and DQPSK modulations,
spectrum sensing is observed and CR user can identify PR user, afterwards changes channel. Then two mentioned
cases are measured. Performance is measured in terms of packet reception rate (PPR). According to first case, PPR
decreases around 91.6 % from the first case for SU when PU transmits with 50 % power transmission at the same
channel. In second case, regardles of power level, performance decrease is not valid. Therefore, study successfully
shows that spectrum sensing is very crucial DSA based solutions and licensed user can have robust performance
only when spectrum is utilized correctly. Because of these facts, study provides effective results. However, authors
do not focus on spectrum sensing because, the main target is a fully functional CR system. Limitation of sensing
with two channels is one example of this fact. Further details of spectrum sensing can be analyzed with channel
based sensing approach.
[21] focuses on noise variance uncertainty problem in the energy detector implementations and by utilizing some
sensing algorithms such as max-min SNR signal energy based spectrum sensing algorithms, real-time implementation
is realized with USRP hardware. As explained in the paper, experiments for energy detectors show the existence
of noise variance uncertainty which degrades detection performance. Novel max-min SNR signal energy based
spectrum sensing algorithms provide proper properties against this problem and minimize the effects of noise
variance uncertainty, adjacent channel interference, carrier frequency offset and symbol timing offset. Due to such
benefits, these algorithms are utilized in the study.
Linear combination of oversampled received signal is the core component of the models of the utilized sensing
algorithms. After solving some optimization problems and applying further solution steps where transmit filter is
known by the receiver, alarm probability equality is obtained. Parameters that play significant roles in detection are
determined and by using these results, experiments are carried out. Testbed includes USRPs as hardware components
and LabVIEW as software component. Carrier frequency of the transmission is considered as 433.5 MHz. For
transmission, OFDM waveform is used and both transmitter and receiver contains filtering steps. Performance of
the system is observed with different tests which measure different components. Firstly, experimental results of the
false alarm probability are compared with theoretical results and it is observed that both results match. Secondly,
change of detection probability results with respect to increase of SNR is experimented. As expected, increase
in SNR causes higher detection probabilities. Moreover, results show that experimented channel performs like an
 7

AWGN channel. Thus, study performs well and its suitability to real-life is valid due to asynchronicity between
hardware which makes study robust against some distortions such as CFO and timing offset.
A scenario is evaluated in [22] where a jammer perturbs the communication between two CR nodes and waveform
switch is aimed during the communication. This model is implemented in real-time by using USRP hardware and
CogWave software. CogWave is an open-source software which includes various waveforms defined. With CogWave,
CR node can change the waveform according to transmission quality. Model of study includes two waveforms
which are DAA-OFDM and DADS. If obtained throughput becomes lower than a threshold, DADS which has low
complexity, is changed with multichannel DAA-OFDM which implements energy-based sensing method. Testbed
includes 4 USRP B100 and 1 USRP N210 devices and these devices are managed remotely by using Arduino
module. Real-time results are obtained by using this testbed. According to results, DAA-OFDM performs well
than DADS in simulated channel. Furthermore, if threshold comparison results lower value, switch of waveform
is successfully implemented. It is clear that such realistic scenario can be implemented easily with SDR tools and
realistic properties can be developed.
Cooperative sensing scheme is implemented in real-time by using USRP N210 devices and GNU Radio software
in [23]. Similar to other approaches in the literature, this study also sees combination of CR and SDR technologies as
a functional tool to solve spectrum scarcity problem. Study includes the implementation of a sensing algorithm with
a model where two secondary users do cooperative sensing. Accordingly, Roy’s Largest Root Test (RLRT) algorithm
which is based on calculation of eigenvalues of the received signal covariance matrix, is used as base algorithm for
sensing decisions. Two scenarios are experimented, RLRT supported cooperation and without cooperation. Beyond
real-time experiments, scheme’s performance is also observed with simulations for all scenarios. Results show that
cooperative sensing has significant performance advantages. Cooperative sensing is based on combination of the
individual sensing measurements of each CR. With this approach, individual sensing limitations and difficulties
can be overcome and sensing performance can be increased clearly. Two methods are used in general which are
centralized and distributed methods. A centralized data base manages cooperative sensing process in centralized
case, and this management brings additional processing disadvantages to network due to necessity of a certain
control mechanism. In distributed case, mentioned limitations are not valid and a centralized data base is not used.
Utilized cooperative sensing method is based on eigenvalue calculation and has blind characteristic, any knowledge
related to primary user is not used. Planned system model is performed based on two hypotheses, first one handles
the case that primary user does not use correspondent channel and other one handles the case that this channel is
used by primary user. By using this binary hypotheses test, utilized sensing method implements sensing based on
eigenvalues of the covariance matrix. After finding these eigenvalues, a decision variable is defined and is compared
with a decision threshold. If considered threshold has a higher value, then it is decided that primary user uses the
channel, otherwise channel is stated as free. In order to manage wrong calculations which can arise because of
some destructive effects, binary test can be validated with false alarm and detection probability evaluations.
As mentioned earlier, utilized system model is also implemented in real-time with the help of some hardware.
According to implementation scenario, one primary user and two CR users are considered. PU transmits a 5 MHz
OFDM signal and its transmit power is kept constant throughout the experiments. A signal generator is used
 8

as noise generator and SNR values can be configured manually by configuring noise generator. In cooperative
sensing process, a centralized data base is realized with a personal computer. This node stores sensing data of
each CR users to be utilized in hypotheses test. It is noteworthy that these CR nodes collect data at 770 MHz
periodically. Moreover, nodes are connected to each other with wires in order to prevent undesirable signals.
Receiver Operating Characteristic (ROC) curve is used to demonstrate performance results. Accordingly, cooperative
sensing outperforms individual sensing if number of samples parameters is at a sufficient value. According to
performance comparisons of hardware and simulation measurements, hardware results show similar facts with
important performance degradation which is due to front-end related limitations.
In result, this study covers the real-time implementation of a RLRT algorithm based cooperative sensing method as
well as simulation validated observations. Study provides comprehensive implementation properties of this scheme
and an important performance difference is demonstrated. However, channel based sensing is not handled and
observed frequency range is limited. Moreover, implementation details are not mentioned in detail probably due to
scope of the study. By handling a more comprehensive spectrum sensing process and modification of some real-time
characteristics, cooperative sensing can be shown as a more suitable method to real-time.
In [24], methods such as frequency hopping, band hopping, transmission band selection for interference avoidance
are analyzed from cognitive radio perspective and a USRP and GNU Radio based testbed is created for implemen-
tation of these methods. In testbed, both transmission and reception processes are included. According to system
model, optimal channel is scanned by a device and if a channel matches expectations, user can operate in this
channel with good transmission performance. Afterwards, it is modeled that device continues to scan operation and
observes other signal interactions in the channel. If interference is valid, then node can change the channel. Four
algorithms are evaluated with testbed which are frequency hopping, band hopping, band selection, and new band
assignment. Frequency hopping algorithm is utilized to prevent interference from harmful users which could be
implemented in security aimed studies. With band hopping, interference signals from other users can be prevented
rather than preventing other users from the signals from the CR. Similar to energy detection methods, if signal
power has higher value than a threshold, band is changed. Note that during this process, transmission frequency of
the CR user is skipped in order to prevent confusion. Band selection algorithm includes first time operations such
as scanning the spectrum and deciding proper frequency area. When band is selected, frequency and band hopping
processes are followed with the help of new band assignment algorithm.
Real-time evaluation of system model is done by using a USRP, a signal generator, a spectrum analyzer and GNU
Radio software. USRP device is used as CR node, spectrum generator and analyzer are used as PU and monitoring
device respectively. In experiments, three separate scenarios are tested. Band selection scenario handles USRP’s
band preference after initial scan when PU is not active. In signal detection scenario, signal detection performance
of the USRP is tested with the operation of PU. In last scenario, it is tested whether USRP can yield to a PU
or not. According to obtained results, channel preferences are similar for first experiment. In second experiment,
signal strength of the PU is a crucial parameter and at 20 dBm gain, success rate of 91% can be obtained. Authors
mention that results are generally good, however, some performance limitations arise due to USRP’s hardware
characteristics. In order to improve study further, this fact should be considered as well as wideband spectrum
 9

sensing could provide more efficient data as mentioned in the paper.

This study includes a GNU Radio and USRP based implementation of spectrum sensing with an energy detection
method that is applied by calculation of magnitude square of the FFT output [25]. By implementation of this energy
detection method, a hypothesis test is evaluated and probability of detection errors are calculated. According to test
scenario, channels are sampled 100 times separately and values are compared with thresholds to decide whether
channel is free or not. Real-time experiments are evaluated with two USRPs which are used as a receiver that
implements spectrum sensing and PU transmitter which transmits data with GMSK modulation respectively. In
experiments, two different scenarios are measured. In first case, sensing ability of the receiver is measured without
the existence of PU. Accordingly, vacant channels can be found successfully throughout the time. In second scenario,
similar model is implemented with the PU transmitter. Accordingly, receiver is again able to detect PU. Note that for
all experiments, sensing is utilized to a fixed bandwidth which is the range between 2.5 GHz – 2.508 GHz. Study
provides essential observations, however, wideband channel based sensing is not included. Scope of the testbed can
be extended and more comprehensive results can be provided.
OFDM based spectrum sensing is analyzed and implemented in real-time by using GNU Radio and USRP in [26].
OFDM based sensing brings some difficulties which are mostly emerged from channel distortions. Due to these,
detection of OFDM signals is a hot topic in the literature and various detectors have been proposed for this process.
In this study, these detectors are analyzed and experimented in real-time in order to measure performances in a
realistic environment which also includes hardware related issues. Moreover, handled algorithms are also modified
to make them suitable to carrier frequency offset (CFO) problem.
CFO existence harms autocorrelation properties of the detectors, thus these algorithms are modified accordingly.
Algorithms such as WRAN, GLRT and averaging are evaluated in the paper. Testbed includes two USRPs which are
used as PU and SU respectively. SU is considered as inactive during the sensing. 2.4 GHz ISM band is considered
and a shielded test environment which does not contain interference from other sources, is chosen. According to
implementation details of SU, firstly SNR estimation is handled. Then time-variant auto-correlation function is
computed to be used in three algorithms. In the end, obtained values are compared with correspondent thresholds
and decisions are made. For proper SNR values and fair results, receiver gain is kept constant. In experiments, two
different scenarios are measured with different metrics. First one is based on the FFT length of 2048 and the other
is based on the length of 256. According to first scenario, GLRT and WRAN detectors are almost same in terms
of performances. With increased sensing time, these performances increase too. Averaging detector performs badly
in these experiments. However it is advantageous when SNR level is quite low, can outperform other detectors. In
second case, results show similar facts, except the difference that smaller FFT length with fixed CP size increases
performances. Here, the WRAN and GLRT detectors show much better performances than the averaging detector
compared to previous case. To sum up, study provides significant results in detection of the OFDM signal issue
and shows proper characteristics of the mentioned algorithms to realistic implementations. It is clear that real-time
brings additional issues which are hard to model in simulations and analyzing correspondent algorithms in this
environment provides more realistic observations.
This study targets link adaptation and performance improvement of the wireless networks when cognitive radio
 10

technology is used [27]. As explained in paper, cognitive radio has various parameters that affect the system perfor-
mance, due to this fact, multi-objective optimization is a suitable approach. Various machine learning techniques are
used with cognitive radio systems in order to determine efficient parameters. Similar to these approaches, an adaptive
parameter adjustment method which is based on generic algorithm is used in this study to provide performance
improvement in terms of bit error rate, power, bandwidth and data rate. As different from previous studies on the
same issue, real-time implementation of utilized system is realized by using SDR hardware which supports MIMO
configuration and realistic performance results are obtained.
Implementation of the transmitter contains three blocks which are adaptive modulation, pilot inserter and genetic
algorithm blocks. In adaptive modulation block, selection of parameters are decided based on the some performance
criteria. Selection of utilized modulation type is the main process. M-ary Quadrature Amplitude Modulation
(MQAM) is used and decision of the M is made in accordance with the quality of the link, if quality is high,
higher M such as 16 or 64 is used in order to increase the data rate. Otherwise, M is selected as 4. In pilot
inserter block, pilots are inserted to the transmitted data to correct noise correlation and frequency offset. In generic
algorithm block is the essential part of the transmission and genetic algorithm is realized by following necessary
steps. According to procedure, some parameters are taken as inputs and a score is given as output by the fitness
function. Determined input parameters are bit error ratio, bandwidth, band efficiency, transmission power and data
rate. These are modeled with some other parameters due to their dependencies which are transmit power, modulation
type, pulse shaping roll-off factor and symbol rate. By using the parameters, algorithm is run by following necessary
optimization steps of the generic algorithm. As said earlier, real-time experiments are implemented in order test
performance of the algorithm in realistic environment. NI RF Signal Generator 5673, NI RF signal Analyzer 5663
modules with NI PXle-1075 chassis are used as hardware components. Channel is modeled as AWGN and BER
is used as performance metric. According the results, higher M values result in higher BER values. Moreover,
system is able to optimize the initial selection and improve the performance. Thus, study successfully shows the
good performance of the adaptation algorithm in real-time and provides feasible results for dynamical parameter
adaptation necessity of CR systems by using SDR technology.
In [28], as an interesting topic, design of the cognitive radio network (CRN) is discussed and core details
are explained. Moreover, created model is experimented in real-time by wireless open-access research platform
(WARP) which is a functional SDR platform. According to authors, usage of machine learning algorithms in CRN
has an important future in this research field. However, algorithms such as matrix completion and decomposition,
kernel adaptive filtering bring significant processing overhead, thus this usage requires powerful hardware for these
computations. However, utilized hardware in CRN cannot be supported with correspondent properties easily and
possible cost of such action is quite high. Thus, approaches for solution of this issue are valuable to deploy
proper models. In this study, CRN testbed is modeled with connection of all nodes to the cloud and leaving all
computations to the cloud. With this configuration, limitations of the implementation of learning algorithm are
overcome. In testbed implementation, nodes are connected to a local network. Network is modeled with some
computers and computations are processed by these computers which are also in local network. As software tool,
MATLAB is used due to its ability of computation various algorithms and its suitability to such cloud based
 11

applications. However, MATLAB causes some computation delays. In order to mange this and ethernet connection
related delays, a latency detection algorithm is created. With this usage, problems related to these delay are mostly
decreased. Frequency band shifting which is a core feature of CR is evaluated in this study with one transmitter SU
and one receiver SU. This feature covers the process that SU looks for vacant bands by sensing the spectrum. An
important problem that band shifting process brings is providing a robust communication link between two SUs.
In order to overcome this, authors propose an algorithm where SU applies spectrum sensing, if sensing fails, same
procedure is implemented for new frequency band which is determined according to a previously created band set.
As a control mechanism, receiver checks incorrectly received packets and if number of these packets become higher
than a threshold, transmission is shifted to new band.
In real-time implementation, one wireless baby monitor is used as PU and two WARPs are used as SUs. Baby
monitor is used in frequency hopping monitor with 70 MHz bandwidth from 2.4 GHz to 2.47 GHz. According to test
scenario, firstly two WARP nodes operate at Wi-Fi channel 4. When PU starts transmission, transmitter node changes
the channel after detecting PU with sensing. In this case, incorrect packets increase and if number of these exceeds
the threshold, receiver also changes the channel. Due to Wi-Fi activity, channel 4 is used initially and channel 9
is used as secondary channel rather than channels 1, 6 and 11. As shown with figures, process outputs successful
detection results and transmission quality can be kept at desirable level. As future work, security in CRN testbed is
described as an important issue and SDR testbed provides necessary data as tested with some minor experiments.
SDR structure is very suitable for CRN based studies as explained in this study. After theoretical modeling, with
a SDR testbed, comprehensive analysis could be done and success of such algorithms can be measured in detail.

IV. A N E NERGY D ETECTION BASED S PECTRUM S ENSING I MPLEMENTATION

Spectrum monitoring is one of the main components of management of the frequency spectrum. With spectrum
monitoring, dynamically changing spectrum can be monitored in real-time and free areas can be detected. Because
of this fact, usage of spectrum monitoring method increases significantly. In future, it is expected that devices will
include this method and will be able to arrange the usage of spectrum by monitoring the spectrum dynamically. In
this study, spectrum monitoring method is implemented by using NI USRP-2921 device and LabVIEW within the
scope of energy management system design and implementation for WLAN networks. With this implementation,
usage density of the 2.4 GHz industrial, scientific and medical (ISM) band is investigated.

A. Energy Detection Spectrum Monitoring System Model

In this section, model of spectrum monitoring method that is used in this study is explained. In our spectrum
monitoring system, energy detection method that has growing usage in literature, is utilized because of its practical
applicability and efficiency. Energy detection is a spectrum detection method that decides to presence of signal in
channel by calculating signal power of received signal. At the energy detector, power outputs are compared with a
threshold which is determined according to noise variance and decision is made. Energy detection method can be
modeled with following hypothesis test []. Signal model can be shown as

y[n] = s[n] + w[n] (2)

 12

where y[n], s[n] and w[n] denotes received signal, transmitted and desired signal and additive white Gauss noise
(AWGN) component respectively. Moreover, n represents sampling index. Test statistic for energy detector can be
represented as
N
X −1
M= |y[n]|2 (3)
n=0

where N denotes length of the observation vector. By comparing determined threshold λE and M value that is
obtained after test statistic, usage statistics of the correspondent frequency band can be obtained. This decision
process can be modeled with two different hypotheses as

H0 : y[n] = w[n] (4)

H1 : y[n] = s[n] + w[n]. (5)

Algorithm for measuring detection performance can be reduced to two probabilities; PD , detection probability and
PF , false alarm probability. PD is the probability that whether a signal is actually at correspondent frequency band
or not and high values of this probability are desired in general. This probability can represented as

PD = P r(M > λE |H1 ) (6)

Furthermore, PF is the probability of detection of a signal at a frequency band which does not include this signal
actually. Opposite to PD , low values of PF are desired. PF can be shown as

PF = P r(M > λE |H0 ) (7)

Threshold λE should be defined consistent in order to balance PD which should be high and PF which should
have low value. However, it is not easy to provide this balance because this process requires noise knowledge and
power properties of detected signals. Signal power is hard to be estimated because of changes in its properties in
accordance with transmission characteristics. Therefore, determination of threshold targets proper estimation of false
alarm probability []. Noise variance is sufficient for determination of threshold. AWGN component can be modeled
as zero mean and σw 2 variance, w(n) = N (0, σw 2 ). For easier analysis, signal component can also be modeled as
zero mean Gaussian distribution with σs 2 variance, s(n) = N (0, σs 2 ). As a result of these models, test statistic,
M(2) is obtained as chi-square distribution with 2N degree of freedom. However, channel characteristics cannot
be obtained properly in real-life due to dynamic changes. Thus chi-square distribution is not a feasible model for
real-time applications. In accordance with these practical limitations, spectrum monitoring system which is utilized
in this study covers calculation of M values in 3 for channels 1, 6 and 11. Moreover, these M values are compared
with determined thresholds in order to decide whether channel is empty or not.

B. Structure of the Measurement System

An energy detection method based spectrum monitoring system is implemented by using LabVIEW and NI
USRP-2921 node in order to observe spectrum activity and understand the usage density of 2.4 GHz ISM band. In
this section, details of this implementation are explained with elementary details.
 13

1) LabVIEW Spectrum Monitoring Code: Algorithm for energy detection method based spectrum monitoring
implementation is prepared by using LabVIEW. In code, wideband spectrum of the 2.4-2.48 GHz frequency band
is obtained by shifting the carrier frequency. Moreover, energy in non-overlapping WLAN channels 1, 6 and 11 is
calculated separately and decisions for channel usages are made by comparing energy results with correspondent
thresholds. Each step is detailed in following parts.
In code, firstly VI sets which are responsible for management of USRP node are placed and connected to each
other. Because of spectrum monitoring function, USRP node only receives data from channels and transmits nothing,
thus only VI set that is used for receiver functionality of USRP is used. This USRP library and correspondent VI
set is shown in Figure 1. Properties of VI set can be summarized as:

Fig. 1: LabVIEW USRP library receiver VI set.

niUSRP Open Rx Session: This VI is necessary to start the RF signal reception process of USRP node. In this
way, USRP becomes active.
niUSRP Configuration Property: With this VI, USRP’s RF signal reception process can be managed. Various
parameters such as IQ sampling rate, carrier frequency, transmit gain, active antenna or channel configurations for
usage of multiple USRPs can be configured by using this VI and RF signal that has desired characteristic can be
obtained.
niUSRP Initiate: This VI gives feedback to USRP about parameter configurations and informs that samples can
be received.
niUSRP Fetch Rx Data: Starts IQ data reception by USRP node and informs that IQ data can be processed.
Without this VI, processing of IQ data cannot be done.
niUSRP Abort: This VI gives close session message to USRP and new session with different parameters can
be started after this message.
niUSRP Close Session: As last component of reception set, this VI ends data reception and allocates a memory
section for this session.
After configuration of this VI set, related parameters for wideband spectrum monitoring application are configured
with niUSRP Configuration Property component. These parameters are configured as in Figure 2. Afterwards, shift
of carrier frequency parameter which is crucial for wideband spectrum analysis is arranged. As shown in Figure
3, USRP-2921 has instantaneous bandwidth up to 20 MHz and USRP cannot operate in bandwidth higher than
20 MHz. However, targeted ISM band has bandwidth of 80 MHz and USRP node cannot support this application
without modification. In order to overcome this limitation, carrier frequency can be changed during reception and
 14

Fig. 2: Parameters utilized in LabVIEW code.

observation frame can be shifted. To this end, a structure to shift carrier frequency is configured as in [29] and
USRP node becomes able to observe whole bandwidth.

Fig. 3: Hardware structure of the USRP device.

As first step of carrier frequency configuration, initial frequency is set which is 2.4 GHz as shown in Figure 2,
and this control element is connected to Ramp Pattern VI as input. Ramp Pattern VI accepts a start value, a delta
value and an value; creates an array that contains elements which take value from start value to end value with
increase by delta value. By setting delta value as multiplication of "IQ Sampling Rate" parameter with 0.25 and
end value as 2.48G, an array that includes all carrier frequency values from 2.4 GHz to 2.48 GHz with 0.25 MHz
increase.
Furthermore, a cluster with name of "Frame Parameters" is created as in Figure 4 to be used in Aggregate
Spectrum VI where wideband spectrum is created by combining smaller spectra. In this cluster, "Frames" value
represents total number of carrier frequencies, "Length of Subset" value represents number of frequency bins that
are used for taking samples from spectrum interval and "Start Index of Subset" value represents start point of sample
 15

Fig. 4: Carrier frequency configuration in LabVIEW code.

in frequency domain. Numerical values of these parameters are demonstrated in Figure 5.

Fig. 5: Values in the Frame Parameters data cluster.

Later, mentioned parameters that are used as controls are sent to two nested for loop structure. In order to process
each carrier frequency separately and measuring each small band consistently, two nested for loops are used. Here
each element of carrier frequency array is processed by exploiting the "Auto-Indexed Tunnel" property of for loop
due to requirement of changing the carrier frequency at each step. With some other values which will be detailed
in following steps, this nested loop structure is created as in Figure 6. Contents of this structure are the core parts
of this code. Reception and processing of data are realized in this structure.
As second step, RX I/Q SubVI is placed into nested loop structure. In LabVIEW, SubVI can be defined as small
VI which also has similar properties just as in function structures of text based programming languages. Contents
of this SubVI are shown in Figure 7.
As it can be understood from the figure, some commands are defined in the structure called "Case Structure"
which is used for realization of some operation in the case of an input matches expected value. As long as these
values match each other, structure continues to run. As first component, niUSRP Property Node is placed. As
explained earlier, USRP parameters are configured according to controls connected as inputs. "Active Antenna"
Fig. 6: Nested for loops and correspondent SubVI.

16
 17

Fig. 7: Block diagram of RX I/Q SubVI.

parameter determines USRP’s antenna to be used, "Number of Samples" parameter determines number of samples,
"Carrier Frequency" parameter determines carrier frequency and spectrum area to be used for RF signal reception.
Afterwards, USRP VI set which is responsible for signal reception are added and obtained data from USRP is given
as output of SubVI by using niUSRP Fetch Rx Data component. After this SubVI, Aggregate Spectrum SubVI is
prepared which has same block diagram view as shown in Figure 15. Accordingly, power spectrum is obtained by
using input data and Power Spectrum and PSD VI which is used for measurement of power spectrum in accordance
with sampling rate. Windowing property of this VI is also configured by connection of Window Parameters" control.
This control includes various windowing alternatives and windowing configuration is also demonstrated to user by
Window Properties VI.
Then, frequency resolution information of the power spectrum which is obtained with the Power Spectrum and
PSD VI is sent to a SubVI where spectrum details are extracted. If SubVI gets data for the first time, an empty
array to keep spectrum measurements is defined and this array is filled iteratively with new spectrum information
which is obtained for correspondent carrier frequency. In the end, combined wideband spectrum is obtained. For
the first-time data case which is valid when data is received for the first time by SubVI, empty array is created with
Initialize Array VI by connection of multiplication of "Frames" and "Length of Subset" controls and 0 to value
control. With these connections, length of array is defined and is totally filled with 0s. Then by combining this
array and frequency resolution information, latest spectrum data is created.
In order to exploit amplitude knowledge obtained from power spectrum, firstly redundant portion is removed with
the help of "Start Index Subset" parameter and then remaining data is inserted into correspondent index of spectrum
data array after removal of previous data which starts from this index. Index value is defined as multiplication of
"Iteration" control which shows the number of iterations of for loops and "Length of Subset" parameter. In this
way, spectrum information is managed correctly in accordance with processed carrier frequency. In result, obtained
Fig. 8: Block diagram of Aggregate Spectrum SubVI.

18
 19

power spectrum is combined with previous spectrum data and this process continues similarly whenever VI is called
again.
After Aggregate Spectrum SubVI, "Case Structure" is generated for energy detection application and band energies
of non-overlapping WLAN channels 1, 6 and 11 are calculated separately. This structure is composed of different
cases which are defined with some frequency intervals and energy calculation method is run for the input carrier
frequency value depending on which interval contains this input. 7 different cases are defined in this structure. Three
essential cases are created for measurements of channels 1, 6 and 11 where frequency intervals are characterized as
2.402-2.422 GHz, 2.427-2.447 GHz and 2.452-2.472 GHz respectively. For measurements of other channels which
are located in remaining parts of ISM band, three more cases are also defined. 7. case which does not contain
energy calculation method is created in order to manage other possibilities where input value is not included in all
six cases. Each main case has same content and an exemplary block diagrams is shown in Figure 9 for first case.

Fig. 9: Case Structure content for three main cases.

Other three cases which are not located in channels 1, 6 and 11 have similar contents and one more exemplary
diagram is shown in Figure 10 for these cases.

Fig. 10: Case Structure content for overlapping channels.

Lastly, content of case 7 which does not have energy calculation method is shown in Figure 11.
In order to plot the obtained spectrum from "Case Structure", Plot Spectrum SubVI is created. This SubVI is
connected after Aggregate Spectrum SubVI and spectrum data that is given as output in Aggregate Spectrum SubVI
is connected as input to Plot Spectrum SubVI. Moreover, windowing setting and "Frame Parameters" control are
also connected as inputs. Spectrum array includes multiple components and two subarray which contain frequency
resolution and amplitude values are extracted. These values are used for start point of the spectrum plot. Similarly,
 20

Fig. 11: Block diagram of case 7.

"Start Carrier" parameter which represents first carrier frequency value and "Start Index of Subset" parameter are
extracted from "Frame Parameters" data set. "Start Index of Subset" parameter is multiplied with frequency resolution
value and subtracted from "Start Carrier" parameter to determine the starting point of the spectrum. Furthermore,
Spectrum Unit Conversion VI in the Spectral Measurement Toolkit is added for unit setting of amplitude values in
the spectrum. Diagram of this VI is shown in Figure 12.

Fig. 12: Block diagram of Plot Spectrum SubVI.

Block diagram of LabVIEW code which is created after all previous steps, is demonstrated in Figure 13.

C. Real-Time Results

In this section, real-time performance measurement results of created energy detection based spectrum monitoring
system are shared and explained. Testbed used during these experiments contains one computer with LabVIEW
software, one NI USRP-2921 node and one switch for ethernet connections. Physical view of this testbed is shown in
Figure 14. Before experiments, necessary antenna and cable connections of USRP node are provided for reception
of spectrum information in ISM band and USRP device is connected to the computer with ethernet cable for
processing of data. After physical preparation, antenna, carrier frequency and number of samples parameters are
configured in LabVIEW. These configured values are shown in Table I. As explained in system model details, band
energy values (M ) for each channel are calculated separately. Then, these values are compared with correspondent
thresholds (λE ). Virtual LEDs in LabVIEW are used to visualize the case that M value is higher than the threshold.
Fig. 13: Block diagram of the spectrum monitoring code.

21
 22

Fig. 14: Physical view of the testbed

After necessary configurations, experiments are evaluated. USRP device is configured to receive 5000 sample
per second and signals from ISM band are received and processed. An exemplary result which was obtained after
one run of the system is shown in Figure 15. As it can be understood from the figure, at that moment, channels
1 and 6 are in high usage, but channel 11 is not dense. In order to understand stability of the system, ISM band
was observed with Anritsu Spectrum Master MS2711E spectrum analyzer. According to this test, same results were
obtained, thus we can claim that system has robust performance.
For further consistency, ISM band was observed in long-time experiments. At different moments, system was
run for one hour and performance results were observed. According to these experiments, usage density of each
channel has decreased compared to previous measurement, and ISM usage has been low in general. As result of
these experiments, an exemplary plot is shown in Figure 16.
In result, real-time and channel based measurements show that energy detection based spectrum monitoring system
operates successfully. With these experiments, non-overlapping ISM channels 1, 6 and 11 have been observed and
empty channels have been found successfully.
 23

TABLE I: Configuration parameters of the USRP.

Parameter Value
First carrier frequency 2.4 GHz
Last carrier frequency 2,48 GHz
I/Q rate 1 MS/sec
Gain 1 dB
Number of carrier frequencies 301

Fig. 15: Exemplary plot obtained after a random experiment.

V. C ONCLUSIONS

CR technology is very crucial for the future of wireless networks and related algorithms would be very helpful
on the spectrum scarcity problem. As supported and explained with various details by this chapter, SDR based
studies are very practical and provide comprehensive results. Experimenting algorithms in real-time with realistic
 24

Fig. 16: Results obtained after longtime experiments

effects of wireless channel and hardware distortions, more robust CR algorithms can be proposed which reduces the
possibility of unsuccessful deployments that affect PU negatively. In this chapter, by comparing and detailing the
important components of spectrum sensing and prominent studies in the literature, we provide crucial insights for
the future studies about SDR based implementation of a practical CR algorithm. We provided a detailed explanation
an practical implementation of the energy based spectrum sensing of 2.4 GHz ISM band by using SDR tools. As
shown, such studies can be implemented in this way without any complexity and robust algorithms can be proposed
by following mentioned structures.

R EFERENCES

[1] G. Sun, G. Liu, and Y. Wang, “Sdn architecture for cognitive radio networks,” in Cognitive Cellular Systems (CCS), 2014 1st International
Workshop on, Sept 2014, pp. 1–5.
[2] D. Cabric, S. Mishra, and R. Brodersen, “Implementation issues in spectrum sensing for cognitive radios,” in Signals, Systems and
Computers, 2004. Conference Record of the Thirty-Eighth Asilomar Conference on, vol. 1, Nov 2004, pp. 772–776 Vol.1.
[3] Y. C. Liang, K. C. Chen, G. Y. Li, and P. Mahonen, “Cognitive radio networking and communications: an overview,” IEEE Transactions
on Vehicular Technology, vol. 60, no. 7, pp. 3386–3407, Sept 2011.
[4] A. Goldsmith, Wireless communications. Cambridge university press, 2005.
[5] R. Kishore, C. K. Ramesha, V. Sharma, and R. Joshi, “Performance evaluation of energy based spectrum sensing in multipath fading
channel for cognitive radio system,” in Communication, Signal Processing and Networking (NCCSN), 2014 National Conference on, Oct
2014, pp. 1–6.
[6] S. Ahmed, M. S. Hossain, M. Abdullah, and M. A. Hossain, “Cooperative spectrum sensing over rayleigh fading channel in cognitive
radio,” International Journal of Electronics and Computer Science Engineering, vol. 1, no. 4, pp. 2583–2592, 2012.
 25

[7] A. Shukla, E. Burbidge, and I. Usman, “Cognitive radios - what are they and why are the military and civil users interested in them,” in
Antennas and Propagation, 2007. EuCAP 2007. The Second European Conference on, Nov 2007, pp. 1–10.
[8] R. M. Hickling, “New technology facilitates true software-defined radio,” RF Design Magazine Apr, 2005.
[9] J. Mitola, “The software radio architecture,” Communications Magazine, IEEE, vol. 33, no. 5, pp. 26–38, 1995.
[10] N. Instruments, “Web sayfasında erişime açıktır: http://www. ni. com,” Feb 2015.
[11] 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, First 2009.
[12] J. Xu and F. Alam, “Adaptive energy detection for cognitive radio: An experimental study,” in Computers and Information Technology,
2009. ICCIT ’09. 12th International Conference on, Dec 2009, pp. 547–551.
[13] J. J. Yang, M. Huang, J. Yu, L. Li, and L. Li, “Usrp: A flexible platform for spectrum monitoring,” in Applied Mechanics and Materials,
vol. 610. Trans Tech Publ, 2014, pp. 233–240.
[14] D. Denkovski, M. Pavloski, V. Atanasovski, and L. Gavrilovska, “Parameter settings for 2.4ghz ism spectrum measurements,” in Applied
Sciences in Biomedical and Communication Technologies (ISABEL), 2010 3rd International Symposium on, Nov 2010, pp. 1–5.
[15] E. Najafzadeh, D. George, and M. P. Green, “Labview-based spectrum occupancy measurements,” in Applied Radio Systems Research and
Smart Wireless Communications, 2012. ARSR-SWICOM 2012. The First Conference on, May 2012, pp. 1–6.
[16] Z. Zhang, W. Zhang, S. Zeadally, Y. Wang, and Y. Liu, “Cognitive radio spectrum sensing framework based on multi-agent arc hitecture
for 5g networks,” IEEE Wireless Communications, vol. 22, no. 6, pp. 34–39, December 2015.
[17] N. Sharma, D. B. Rawat, B. B. Bista, and S. Shetty, “A testbed using usrp(tm) and labview(r) for dynamic spectrum access in cognitive
radio networks,” in 2015 IEEE 29th International Conference on Advanced Information Networking and Applications, March 2015, pp.
735–740.
[18] C. Jiang, N. C. Beaulieu, L. Zhang, Y. Ren, M. Peng, and H. H. Chen, “Cognitive radio networks with asynchronous spectrum sensing
and access,” IEEE Network, vol. 29, no. 3, pp. 88–95, May 2015.
[19] S. Soltani, Y. Sagduyu, Y. Shi, J. Li, J. Feldman, and J. Matyjas, “Distributed cognitive radio network architecture, sdr implementation
and emulation testbed,” in Military Communications Conference, MILCOM 2015 - 2015 IEEE, Oct 2015, pp. 438–443.
[20] M. A. Sarijari, R. A. Rashid, N. Fisal, A. Lo, S. Yusof, and N. Mahalin, “Dynamic spectrum access using cognitive radio utilizing gnu
radio and usrp,” in 26th Wireless World Research Forum (WWRF26). Citeseer, 2011.
[21] T. E. Bogale and L. Vandendorpe, “Usrp implementation of max-min snr signal energy based spectrum sensing algorithms for cognitive
radio networks,” in 2014 IEEE International Conference on Communications (ICC), June 2014, pp. 1478–1482.
[22] V. L. Nir and B. Scheers, “Description of a cognitive radio testbed based on usrp platforms and cogwave,” in 2014 9th International
Conference on Cognitive Radio Oriented Wireless Networks and Communications (CROWNCOM), June 2014, pp. 514–519.
[23] R. S. Yoshimura, F. S. Mathilde, J. P. Dantas, V. A. de S Jr, J. H. da Cruz Jr, J. J. Bazzo, and D. C. Melgarejo, “A usrp based scheme
for cooperative sensing networks,” in Anais do 4 Workshop de Redes de Acesso em Banda Larga - WRA 2014, Brazil, pp. 1–5.
[24] H. Shahid and Y. D. Yao, “Algorithm and experimentation of frequency hopping, band hopping, and transmission band selection using a
cognitive radio test bed,” in 2014 23rd Wireless and Optical Communication Conference (WOCC), May 2014, pp. 1–5.
[25] M. A. Sarijari, A. Marwanto, N. Fisal, S. K. S. Yusof, R. A. Rashid, and M. H. Satria, “Energy detection sensing based on gnu radio and
usrp: An analysis study,” in Communications (MICC), 2009 IEEE 9th Malaysia International Conference on, Dec 2009, pp. 338–342.
[26] A. Blad, E. Axell, and E. G. Larsson, “Spectrum sensing of ofdm signals in the presence of cfo: New algorithms and empirical evaluation
using usrp,” in 2012 IEEE 13th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), June 2012,
pp. 159–163.
[27] P. V. Kumar and M. L. N. Sai, “Sdr based mimo link adaptation for cognitive radio application,” in Communications and Signal Processing
(ICCSP), 2014 International Conference on, April 2014, pp. 577–581.
[28] Z. Chen, C. Zhang, F. Lin, J. Yu, X. Li, Y. Song, R. Ranganathan, N. Guo, and R. C. Qiu, “Towards a large-scale cognitive radio network:
Testbed, intensive computing, frequency agility and security,” in Computing, Networking and Communications (ICNC), 2012 International
Conference on, Jan 2012, pp. 556–562.
[29] N. Instruments, “Spectrum Monitoring with NI USRP , https://decibel.ni.com/content/docs/DOC-34781 ,” April 2016.

View publication stats