Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
Detection and Estimation of multiple far-field primary users using sensor array in Cognitive Radio Networks

Detection and Estimation of multiple far-field primary users using sensor array in Cognitive Radio Networks

Ratings: (0)|Views: 61 |Likes:
2013, Journal of Computing, eISSN 2151-9617, http://www.journalofcomputing.org/, Call For Computer Science Papers
2013, Journal of Computing, eISSN 2151-9617, http://www.journalofcomputing.org/, Call For Computer Science Papers

More info:

Categories:Types, Research
Published by: Journal of Computing on Apr 29, 2013
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

04/29/2013

pdf

text

original

 
Detection and Estimation of multiple far 
-
fieldprimary users using sensor array in CognitiveRadio Networks
 
Kiran Sultan, Ijaz Mansoor Qureshi, Muhammad Zubair 
Abstract
— The field of spectrum sensing faces a lot of challenges in terms of reliability and accuracy in gathering information for detectionand estimation of primary transmissions in Cognitive Radio Networks (CRNs). We propose an efficient, reliable and low
-
complexity spectrumsensing scheme for CRNs which not only detects the number of sources but also estimates their parameters such as frequency, Direction
-
of 
-
 Arrival (DOA) and power strength. It is based on Genetic Algorithm (GA) as global optimizer hybridized with Pattern Search (PS) as localoptimizer. Fitness function is derived from Maximum Likelihood (ML) principle and defines the MSE between actual and estimated signals. Itseffectiveness under low SNR conditions is proved. Our proposed system model constitutes a uniform linear array (ULA) of sensors. Bestestimates of the parameters of the active primary users are obtained by minimizing the fitness function. We detect signals in the frequencyband of 80MHz
-
108MHz and assume far 
-
field approximation and the snapshots are available to us after 10
-
15 seconds.
Index Terms
— Cognitive Radio Network, Direction
-
of 
-
 Arrival, Spectrum Sensing
 
——————————
 
 
——————————
 
1 INTRODUCTION
S
pectrum sensing [1] is a process conducted to becomeaware of the status of the spectrum usage which involvesdetection of active signals then estimation of the signalparameters, followed by decision but it has revamped as avery active area of research with the advent of cognitiveradio technology [2]. In Cognitive Radio (CR), spectrumsensing is a decision making technique in which secondaryusers (SUs) are required to dynamically detect spectrumholes to become aware of the presence of the primary users
(PUs) which have high priority being the licensed users.
Beingthe core component of Cognitive Radio Network (CRN),spectrum sensing faces many challenges [3] in terms ofhardware requirements, hidden terminal problem, detectionof spread spectrum primary users, data/decision fusion inscenarios of cooperative sensing, multipath fading, noisepower uncertainty, implementation complexity, security etc.In order to meet these challenges efficiently, spectrumsensing requires innovative techniques for not onlydetecting the number of PUs but also estimating theiramplitudes, DOAs and frequencies to avoid interference between primary and secondary transmissions. A number ofspectrum sensing methods to detect spectrum holes in CRshave been proposed in literature which have been broadlycategorized into three main classes: Non-cooperativespectrum sensing [4], cooperative spectrum sensing [5]-[6]and interference based spectrum sensing [7].Non-cooperative spectrum sensing also known astransmitter detection is further classified into EnergyDetection (ED), Matched Filter Detection (MFD) andCyclostationary based Detection (CBD). Energy Detector [8]is the most widely studied spectrum sensing technique because of its less complexity and no requirement of priorknowledge of PU signal, but it is accompanied by a numberof shortcomings which include noise power uncertainty,poor performance under low SNR and inefficient to detectspread spectrum signals. MFD [9] is considered as theoptimum method of signal detection when perfectknowledge of PU is available otherwise it performs poorly.Implementation complexity of MF is impractically large because it demands CR to have dedicated receivers for allsignal types. CBD [10] relies on the prior knowledge of PUsignals and exploits cyclostationary features of the receivedsignals, hence it is capable of differentiating PU signals andnoise. Its implementation complexity lies between energydetector and matched filter.
 ———————————————— 
 
 
Kiran Sultan, Department of Electrical Engineering, Air University,Islamabad, Pakistan, 46000.
 
 
Ijaz Mansoor Qureshi, Department of Electrical Engineering, Air University, Islamabad, Pakistan, 46000.
 
 
 Muhammad Zubair, Department of Electronics Engineering, InternationalIslamic University, Islamabad, Pakistan, 46000.
JOURNAL OF COMPUTING, VOLUME 5, ISSUE 2, FEBRUARY 2013, ISSN (Online) 2151-9617https://sites.google.com/site/journalofcomputingWWW.JOURNALOFCOMPUTING.ORG7
 
The focus of interference-based spectrum sensing is todesign the CRNs to operate in underlay spectrum sharingenvironment. In this method, SUs do not perform spectrumsensing to find spectrum opportunities rather they identifyspectrum occupancy status of PUs and an interferencepower threshold is set up for SUs towards PUs for aparticular frequency band and location. In cooperativespectrum sensing, SUs collaborate and share sensinginformation to solve problems like hidden terminal problem,receiver uncertainty and multipath fading at the cost ofincreased detection delay and high implementationcomplexity due to requirement of control channels efficientinformation sharing algorithms.With new challenges and dimensions in CRNs, sensingfrequency only may not be enough. Thus it requiresexploration of new dimensions of direction of arrival (DOA),frequency, strength of signal, range and a critical parameterwhich is number of active PUs. All these parametersformulate a hyperspace which may be called as transmissionhyperspace [11] or radio spectrum space. Knowledge of thishyperspace will provide more comprehensive view of theradio environment which has to be shared by multiple users.In order to ensure secure, reliable and efficientcommunication keeping in view the privilege of PUs,advanced spectrum sensing algorithms capable ofidentifying occupancy in all of the above dimensions ofspectrum space to locate spectrum holes need to bedeveloped which have not been considered simultaneouslyin CRNs yet according to the best of our knowledge.Source localization by means of sensor arrays has been oneof the fundamental and effective ways to estimateamplitude, frequency, DOA and range estimation of both farand near field sources upto high accuracy in many systemsincluding radar, navigation and wireless communicationsystems. In order to achieve optimum performance of asensor array [12], array geometry, the number of sensors andthe physical separation between the sensors are criticaldesign parameters in addition to the number of other factorsincluding signal-to-noise ratio. Many algorithms havealready been proposed in array signal processing for sourcelocalization which can be categorized into far-field sourcelocalization and near-field source localization on the basis ofrange between the radiating source and the array of sensors.Far-field source localization algorithms make assumptionthat sources are located in the far-field region of the array.Thus each signal arriving at the array has planar wavefront.ESPRIT algorithm [13] and MUSIC algorithm [14] are amongthe widely studied far-field source localization algorithms.However, the far-field assumption is no longer valid whenthe sources are located close to the array and are described by spherical wavefront assumption, thus range parameter isalso included in addition to amplitude, DOA and frequencyto characterize the sources. A number of techniques have been proposed in this area such as 2-D MUSIC [15], LinearPrediction method [16], higher order ESPRIT-method [17]etc but most of these algorithms are computationallycomplex.This paper addresses the problem of detecting the number ofactive PU signals and then estimating their signalparameters to ensure interference free communication inCRNs. Most of the existing techniques to determine thenumber of sources are based on the Singular ValueDecomposition SVD of the covariance matrix of thesnapshots which yields M distinct eigenvalues, where M isthe number of signals present and the remainingeigenvalues are either zero or non-zero repeated eigenvalues[18] or non-zero eigenvalues less than threshold. However,SVD has high uncertainty in terms of decisions about settingof the threshold and so different schemes [19] have beenproposed for threshold setting to detect the presence ofsignals. These include Maximum Eigenvalue Detection(MED), Maximum Minimum Eigenvalue (MME), MaximumEigenvalue to Trace (MET) etc. Unfortunately, most of theexisting methods are either problem specific orcomputationally complex due to exhaustive comparisons oftest hypothesis involved to achieve high accuracy.In [20], atechnique is proposed to detect number of signals in order tosolve problem of DOA.In this paper, we propose a generalized method to firstdetect the number of possibly active primary users located inthe far field region of the array and then estimate theiramplitudes, DOAs and frequencies. Our proposed spectrumsensing scheme is not application specific. It can be used forcooperative as well as non-cooperative spectrum sensing.We use Mean Square Error (MSE) as fitness function whichdefines an error between actual and estimated signals atdifferent sensors of the uniform linear array ULA and isderived from Maximum Likelihood (ML) Principle. MSE isone of the easy and optimum fitness functions to beminimized using array of sensors and fairly good results areobtained even in the scenario of low signal to noise ratio(SNR). We employed heuristic optimization techniques tominimize the error in which Genetic Algorithm (GA) [21] being one of the most popular evolutionary algorithm because of its reliability, efficiency and robustness is used as
JOURNAL OF COMPUTING, VOLUME 5, ISSUE 2, FEBRUARY 2013, ISSN (Online) 2151-9617https://sites.google.com/site/journalofcomputingWWW.JOURNALOFCOMPUTING.ORG8
 
global optimizer hybridized with Pattern Search (PS) as localoptimizer. This simple and elegant technique simplydemands a passive sensor array whose snapshots should bereadily available to us for calculation after every 10-20seconds.
2 SYSTEM MODEL AND PROBLEM FORMULATION
We have an array of sensors that is sensing the signals fromdifferent base stations of primary users. If the array is almostat the same height as that of the base station transmitters, wedo not have to detect the elevation angle
 
. So, consider auniform linear array as shown in fig.1 consisting of Lomnidirectional sensors observing M far-field primarysignals radiating with different unknown carrier frequencies.The distance
 between two consecutive elements is keptone-quarter of the minimum wavelength of received signalsi.e.
4
min
 
.
Fig.1. The System Model
The composite signal
i
received by the
th
i
sensor isexpressed as,
i M mi jk mi
ea
mm
1sin)1(
 
 
 Li
1
 
)1(
 where
m
a
and
m
 
represent the amplitude and DOA of the
th
m
source impinging on the array,
m
is the propagationconstant and
c  f  
mmm
/2/2
   
with
m
 representing the frequency of the
th
m
signal incident on thearray and
i
 z 
is the AWGN added to the output of
th
i
 sensor. Thus the parameters to be estimated for M incidentsources are expresses in a vector
 
 as,where M is the number of active PUs and is also unknownand has to be detected first.The received signal vector at the L-element ULA isexpressed as,
 L Li
x x x x x
],.,,..................,,[
121
 where superscript T denotes the transpose.Thus the problem in hand is to develop a novel technique fortwo purposes, first detecting the number of active PU signalsimpinging on ULA and second, performing joint estimationof amplitude, DOA and frequency of the detected sourcesconsidering the sensor array as reference. We also considerthe effect of variation in SNR on the detection and estimationresults. The fitness function can be expressed mathematicallyas,
2,,,
ˆ
min
 X 
 f  a
 g 
 
 
)2(
 where defines the estimated signal vector at the sensorarray and is given as,
 L Li
x x x x x
]
ˆ
,
ˆ
.,,.........
ˆ
.........,
ˆ
,
ˆ
[
121
 
i
ˆ
is the estimated output at the i
th
sensor and is expressedas,
 g mm
 M mi j mi
ea
1
ˆ
sin)1(
ˆ
''''
ˆˆ
 
 
 Li
1
 
)3(
 where
 g 
is the number of sources randomly selected todetect the possibly active PUs.Thus the elements of the estimated vector
'
 
 obtainedthrough optimization algorithm are given by,
3 Proposed Algorithm for Detection
In this section, we give an overview of the procedural stepscarried out in GA optimization, parameter settings for GAand hybrid scheme PS, and pseudo code for the proposedalgorithm. We solve the problem of detection first. Toachieve this purpose, we randomly select
 g 
number ofsources in the estimated signal vector
 
 and calculate meansquare error MSE given in eq. (2). The value of
 g 
 is thenincreased or decreased aiming to decrease the MSE in eachselection. This process is repeated until minimum mean
,......,,,......,,,......,
111
 M  M 
aa
   
]
ˆ
,......,
ˆ
,
ˆ
,......,
ˆ
,
ˆ
,......,
ˆ
['
111
 g  g  g 
 M  M 
f    f  aa
   
JOURNAL OF COMPUTING, VOLUME 5, ISSUE 2, FEBRUARY 2013, ISSN (Online) 2151-9617https://sites.google.com/site/journalofcomputingWWW.JOURNALOFCOMPUTING.ORG9

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->