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
3Activity
0 of .
Results for:
No results containing your search query
P. 1
Performance Enhancement of High Resolution Multiple Widebandand Nonstationary Source Localization with Unknown Number of Sources

Performance Enhancement of High Resolution Multiple Widebandand Nonstationary Source Localization with Unknown Number of Sources

Ratings: (0)|Views: 38 |Likes:
Published by JournalofICT
Journal of Information and Communication Technologies, ISSN 2047-3168, Volume 2, Issue 11, December 2012

http://www.jict.co.uk
Journal of Information and Communication Technologies, ISSN 2047-3168, Volume 2, Issue 11, December 2012

http://www.jict.co.uk

More info:

Published by: JournalofICT on Dec 17, 2012
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

01/04/2013

pdf

text

original

 
JOURNAL OF INFORMATION AND COMMUNICATION TECHNOLOGIES, VOLUME 2, ISSUE 11, DECEMBER 20126
Performance Enhancement of HighResolution Multiple WidebandandNonstationary Source Localization withUnknown Number of Sources
Sandeep Santosh and O.P.Sahu
 
Abstract-
 
In this paper, Performance Evaluation of high-resolution multiple wideband and nonstationary sourcelocalization using a sensor array is presented. The sensor array’s received signals are first converted into time-frequency domain via short-time Fourier transform (STFT)and it is found that a set of power spectrum matrices atdifferent time instants havethe joint diagonalization structure in each frequency bin. Using such a joint diagonalizationstructure, a novel cost function is designed and a new spatial spectrum for direction of arrival(DOA) estimation isobtained.The algorithm in this paper obtains the DOA estimates via one –dimensional (1-D) search instead ofmultidimensional search.It’s computational complexity is much lower than the ML method.For this algorithm ,it is notnecessary to determine the number of sources in advance unlike the subspace based high- resolution DOAestimation techniques .The algorithm used in this paper is robust to the effects of reverberation caused by multipathreflections. It is suitable for multiple acoustic source localization in a reverberant room.
Index Terms-
Direction of Arrival(DOA) Estimation, Nonstationarity, Source localization, Joint diagonalization, Time-delay estimation.
 
 Sandeep Santosh,Astt. Prof., Department of Electronics and Communication Engineering 
 
 ,O.P.Sahu ,Prof., Department of Electronics and Communication Engineering, National Institute of Technology ,Kurukshetra, INDIA
1
 
INTRODUCTION
The received signals of the sensor array are firstconverted into time-frequency domain via short-time Fourier transform(STFT) and we find thatset of short –time power spectrum matrices atdifferent time instants have the jointdiagonalization structure in each frequency bin.Based on such joint diagonalization structure, anovel cost function is designed and a newspatial spectrum for direction of arrival (DOA)estimation at hand is derived. Compared tomaximum-likelihood (ML) method with highcomputational complexity ,the proposedalgorithm obtains the DOA estimates via onedimensional (1-D)search instead of multidimensional search. Therefore, itscomputational complexity is much lower thanML method .Unlike the subspace- based high-resolution DOA estimation techniques ,it is notnecessary to determine the number of sources inadvance for the proposed algorithm. The proposed method is robust to the effects of reverberation caused by multipath reflections.Hence, it is suitable for multiple acoustic sourcelocalization in a reverberant room. The results of numerical simulations and experiments in a realroom with a moderate reverberation are providedto demonstrate the good performance of the proposed approachSource localization has been an important topicthat arises in many fields such as sonar, radar,communications,speechprocessing,oceanography andnavigation[1],[2].Numerous high-resolution
 
JOURNAL OF INFORMATION AND COMMUNICATION TECHNOLOGIES, VOLUME 2, ISSUE 11, DECEMBER 20127
algorithms like maximum-likelihood (ML)method, MUSIC,ESPIRIT and subspace fittingmethod have been developed for estimating thedirection of arrival (DOA) of narrowband sources. The energy of narrowbandsource is concentrated in a frequency bandwhich is much smaller than the centrefrequency. However, the wideband signals arewidely encountered in many applications such ashigh data rate communications, passive sonar,seismic signal processing and speech signal processing. Unfortunately, these DOAestimation methods designed for narrowbandsignals are no longer applicable to widebandsources since the phase difference between thesensor outputs is not only dependent on DOAalone but also depends on temporal frequencywith a wide range[7].A variety of techniques have been proposed todeal with the wideband source localization problem. The ML method is extended towideband situations in [8] and [9].Since the MLmethod involves to maximize the likelihoodfunction with high nonlinearity andnonconvexity, it requires multidimensionalsearch and its computational complexity is high.A class of representative high resolution DOAestimation approaches for wideband sources isCoherent signal subspacemethod(CSSM)[7].CSSM converts thewideband signal subspaces into predefinednarrowband subspace by focusing matrices andsubsequently the narrowband subspace basedDOA estimation methods like MUSIC can beapplied. However, CSSM requires initialfocusing angles which are not far from trueDOAs to find such focusing matrices , and theestimation performance of CSSM is sensitive tothese initial angles.There are two intrinsic limitations of subspacedecomposition based methods. One limitation isthat they require the number of sources to beknown or to be exactly estimated in advance.The information theoretic criteria such as AICand MDL are most important methods for determining the number of sources. In practice,the rate of correctly detecting the number of sources can be somewhat low with moderatesignal –to-noise ratio (SNR)and its performancemay not be satisfying. The other drawback of subspace based methods is that they demand thenumber of sources be strictly less than thenumber of sensors.Two other classes of source localizationapproaches are based on time-delayestimation(TDE). One class is composed of twostages :first ,estimate the time delays amongspatially separated sensors, and then localize thesources by exploiting the estimated time delaysand the array geometric information.[14],[18]. Atypical and popular TDE algorithm in practice isthe generalized cross-correlation (GCC) method proposed by Knapp and Carter. An extremeexample of GCC is the phase transform GCCwhere only the crosspower-spectrum phase isused for TDE.GCC has good performance inmoderate SNR and nonreverberantenvironment. The performance of GCC willdegrade when multipath reflections andreverberations appear. The GCC method canonly obtain the integer part of time-delay. All of the source localization algorithmsaforementioned assume a stationary signalmodel .However, nonstationary signals arefrequently encountered in many fields such asacoustic and speech signal processing, seismicsignal processing etc. One important techniquewidely used in many practical applications e.gvideoconferencing, hands-free voicecommunication, acoustic targetdetection andsurveillance is the acoustic source localizationusing microphone arrays. In speechenhancement ,separation etc, localization is alsoused to help to form a fixed beamforming to provide an initial guess for those separation andadaptive beamforming algorithms. Notice thatthe speech signals are nonstationary in nature .Therefore, it is expected that the performance of the localization algorithm will improve if thenonstationarity of signals is taken into account.We propose an efficient multiple wideband andnonstationary source localization algorithm byexploiting both the nonstationarity of the signalsand the array geometric information. The proposed algorithm is based upon jointdiagonalization structure of a set of short time power spectrum matrices at differenttimeinstants of each frequency bin. It can be used
 
JOURNAL OF INFORMATION AND COMMUNICATION TECHNOLOGIES, VOLUME 2, ISSUE 11, DECEMBER 20128
forquick and accurate multiple nonstationarysource localization . A new spatial spectrum for multiple wideband and non-stationary sourcelocalization is proposed . The DOAs of thesources can be estimated by 1-D search of thespatial spectrum and multidimensional search isavoided. The computational complexity isgreatly reduced compared with the ML method.The proposed algorithm is one stage i.e. itdirectly searches the DOAs over continuouslocation parameter space. Thus the proposedalgorithm does not suffer the error accumulation problem. Unlike the subspace based methods, itis not necessary to determine the number of sources before computing the spatial spectrum.Most existent high-resolution DOA estimator demand that the number of sources must bestrictly less than the number of sensors.However, it requires that the number of sensorsis not less than number of sources for the proposed method. The simulation andexperimental results validate that the proposedalgorithm is robust to reverberation caused bymultipath propagation effect. Therefore,it isespecially suited for multiple speakerslocalization in a reverberant room.
2
DATA MODEL
 
The N statistically uncorrelated wideband andnonstationary sources impinge on a uniformlinear array (ULA) of M sensors with inter-sensor spacing d. We further assume that thesize of acoustic array is much smaller comparedwith the distance from sources to the array.Hence, a plane wave propagation model can beadopted .In sub-space based high resolutionDOA estimators, N must be strictly less than M.In the proposed approach we only need toassume N
M. In other words N can be equal toM for proposed method. Let the first sensor bethe reference point. We use s
n
(t
c
) to represent thewaveform of the nth source with t
c
denoting thecontinuous time index. Then the signal received by the mth sensor is ,x
m
(t
c
) =
n=1 N
α
n
s
n
(t
c
 – 
τ
nm
)+z
m
(t
c
)m=1,2 ,. ,M
(1)
where
α
n
is an attenuation factor due to propagation effects , z
m
(t
c
) is the additive whiteGaussian noise and
τ
nm
denotes propagation timedelay from the mth sensor to the first sensor of the nth source which can be written as
τ
nm
=(m-1)dcos
θ
n
/v (2)where v is the propagation speed of signals and0
≤θ
n
 
π
is the DOA of the nth source.The continuous signals are sampled at samplingfrequency f 
s
= 1/T
s
. Then the discrete timereceived signals are ,x
m
(t) =
n=1 N
s
n
(t-(m- 1)k cos
θ
n
) + z
m
(t),m=1,2 ,….. ,M (3)where t denotes the discrete time index andk=f 
s
d/v. (4)is a constant. Note that the time-delay (m- 1)k cos
θ
n
in (3)may be non integer multiples of sample period T
s
, and s
n
(t-(m- 1)k cos
θ
n
heredenotes the copy of s
n
(t) with fractional delay(m- 1)k cos
θ
n
.
3
NEW MULTIPLE SOURCELOCALISATION ALGORITHM.3.1 Short –time Fourier Analysis of theReceived Signals
It should be pointed out that most existing timedelay and source location algorithms assume astationary signal model . The speech signals arenonstationary in nature. The short-time Fourier Transform is adopted to process nonstationaryacoustic signals. The short-time discrete timeFourier Transform of a signal s(t) is defined as ,s(
ω
,t)=
τ
=1T
s(t-
τ
)w(
τ
)e
-j
ωτ
(5)where 0
≤ω≤
2
π
is the digital frequency and j=
-1 is the imaginary unit and w(
τ
) is an analysiswindow function with length T.

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)//-->