《频域自适应算法2003》由会员分享,可在线阅读,更多相关《频域自适应算法2003(14页珍藏版)》请在金锄头文库上搜索。
1、IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 51, NO. 1, JANUARY 200311A Class of Frequency-Domain Adaptive Approaches to Blind Multichannel IdentificationYiteng (Arden) Huang, Member, IEEE, and Jacob Benesty, Member, IEEEAbstractIn this paper, we extend our previous studies on adaptive blind channel
2、 identification from the time domain into the frequency domain. A class of frequency-domain adaptive approaches, including the multichannel frequency-domain LMS (MCFLMS) and constrained/unconstrained normalized mul- tichannelfrequency-domainLMS(NMCFLMS)algorithms, are proposed. By utilizing the fast
3、 Fourier transform (FFT) and overlap-save techniques, the convolution and correlation operations that are computationally intensive when performed by the time-domain multichannel LMS (MCLMS) or multichannel Newton (MCN) methods are efficiently implemented in the fre- quency domain, and the MCFLMS is
4、 rigorously derived. In order to achieve independent and uniform convergence for each filter coefficient and, therefore, accelerate the overall convergence, the coefficient updates are properly normalized at each iteration, and the NMCFLMS algorithms are developed. Simulations show that the frequenc
5、y-domain adaptive approaches perform as well as or better than their time-domain counterparts and the cross-relation (CR) batch method in most practical cases. It is remarkable that for a three-channel acoustic system with long impulse responses (256 taps in each channel) excited by a male speech si
6、gnal, only the proposed NMCFLMS algorithm succeeds in determining a reasonably accurate channel estimate, which is good enough for applications such as time delay estimation.Index TermsBlind channel identification, frequency-domain adaptive filtering, least mean square, multichannel signal pro- cess
7、ing.I. INTRODUCTION SYSTEM identification is the technique of building a math- ematical model of an unknown dynamic system by ana- lyzing its input/output data. This problem of fundamental in- terest arises in a variety of signal processing and communica- tions applications. The ability to identify
8、a system facilitates a better understanding of how an input signal is transmitted/pro- cessed/distorted by the system and, therefore, enables a prac- tical attempt to equalize the dispersive effect introduced by the fading channels and/or to design a more efficient communica- tions system. Since, in
9、 practice, the system is generally nonsta- tionary andusually hasalongimpulseresponse,determining its characteristics is not easy, even when the input signal is known a priori, such as in the case of acoustic echo cancellation. How- ever, in many other cases, e.g., acoustic dereverberation, wire- le
10、ss communications, time delay estimation, etc., the input is either unobservable or very expensive to acquire; the choice in-Manuscript received November 7, 2001; revised August 19, 2002. The asso- ciateeditorcoordinatingthereviewofthispaperandapprovingitforpublication was Dr. Inbar Fijalkow. Theaut
11、horsarewithBellLaboratories,LucentTechnologies,MurrayHill,NJ 07974 USA (e-mail: ardenresearch.bell-; jbenestybell-). Digital Object Identifier 10.1109/TSP.2002.806559evitablycomesdowntoablindmethod,andasaresult,thechan- nels are more difficult to estimate. Theinnovativeideaofblindchannel identificat
12、ionandequal- ization was first proposed by Sato in 1. Since then, many algorithms have been proposed. Broadly, one can dichotomize these approaches into the class of second-order statistics (SOS) methods and the class of higher order statistics (HOS) methods. Because HOS cannot be accurately compute
13、d from a small number of observations, slow convergence is the critical draw- back of all existing HOS methods. In addition, a cost function basedontheHOSisbarelyconcave,andanHOSalgorithmcanbe misledtoalocalminimumbycorruptingnoiseintheobservations. Since it was recognized that the problem can be so
14、lved in the lightofonlySOS2,thefocusoftheblindchannelidentification research has shifted to SOS methods, behind which, the motiva- tionisthepotentialforfastconvergence.Thereisarichliterature on SOS blind channel identification. Many batch methods have had good success to some extent, such as the sub
15、space (SS) algorithm3,thecrossrelation(CR)algorithm4,5,theleast squares component normalization (LSCN) algorithm 6, the linear prediction based subspace (LP-SS) algorithm 7, and the two-stepmaximumlikelihood(TSML)algorithm8(see9for areviewontheSOSmethodsandthereferencestherein). According to 10, a s
16、atisfactory blind channel identification algorithm needs to satisfy three design requirements: 1) quick convergence; 2) adaptivity; 3) low complexity. Most SOS batch methods can converge quickly, but unfortu- nately, they are difficult to implement in an adaptive mode 9 and are in general computationally intensive. In an earlier study 11, we developed an adaptive eigenvalue decomposition algo- rithm to blindly identify a single-input two-output acousti