文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > A Gauss-Newton method for blind source separation of convolutive mixtures
侵权投诉

A Gauss-Newton method for blind source separation of convolutive mixtures

In this paper we present several Gauss-Newton algorithms for blind source separation of convolutive mixtures. The algorithms can be interpreted as generalizations of two previous approaches due to Gerven-Compernolle [7] and Nguyen-Jutten [5]. Since they ar

A GAUSS-NEWTON METHOD FOR BLIND SOURCESEPARATION OF CONVOLUTIVE MIXTURES

Sergio Cruces

TSC Group, ESI Telecomunicación, Univ. SevillaCamino de Descubrimientos, 41092-Sevilla, Spain

ABSTRACT

In this paper we present several Gauss-Newton algorithms forblind source separation of convolutive mixtures. The algorithmscan be interpreted as generalizations of two previous approachesdue to Gerven-Compernolle [7] and Nguyen-Jutten [5]. Sincethey are of the Gauss-Newton type, they exhibit a fast rate ofconvergence. Also, we present a stability analysis for twosources and instantaneous mixtures where we show that thealgorithms cannot converge to non-separating solutions.

1.INTRODUCTION

Blind Source Separation (BSS) is a fundamental problem in

signal processing with a large number of applications in speechprocessing, array signal processing, multiuser communications,etc. [2]. If we assume that several statistically independent non-Gaussian sources are transmitted through a Multiple InputMultiple Output (MIMO) Linear and Time Invariant (LTI)system, therefore arriving at an array of sensors in the form of aconvolutive mixture, the BSS problem consists in recovering theoriginal sources from the observations only, without knowing thetransmission channel and the sources.

Most of the existing approaches to BSS have been developed forinstantaneous mixtures. In practical situations, however, thishypothesis is not true and it is more common to find convolutivemixtures. Different algorithms [5][7] have been proposed toseparate convolutive mixtures of sources in a blind way. In thispaper we present a new approach based on the minimization of astatistical dependence measure by means of Gauss-Newtonalgorithms.

This paper is organized as follows. The signal model is presentedin Section 2. Section 3 develops the proposed method andderives two simplified versions. It also revisits existingalgorithms [5][7] and interprets them as particular cases of theproposed one. A convergence analysis for two sources andinstantaneous mixtures will be presented in Section 4. Section 5shows the results of computer simulations and Section 6 isdevoted to the conclusions.

2. SIGNAL MODEL

Let us consider N non-Gaussian and statistically independentsources s[n]=[s1[n], ,sN[n]]T that are mixed through a MIMOLTI system with memory. The elements of the vector of M≥Nobserved signals x[n]=[x1[n], ,xM[n]]T

are given by

Luis Castedo

Dpto. Electrónica y Sistemas, Univ. A CoruñaCampus de Elviña, 15071-A Coruña, Spain

N

La2

xi[n]=

∑∑a

ij

[k] sj[n k].

j=1k= La1

This mixture is characterized by the system’s transfer matrix

A(z)= La2 ∑

ak] z k

ij[ . k= La1 ij

In order to separate the sources, we introduce a MIMO LTI

z)= Lb2system with memory B(

bij[k]z k to produce the

k=∑

Lb1 ij

vector of separated signals y[n]=[y1[n], ,yN[n]]T where

MLb2

yi[n]=

∑∑bij

[k] x

j

[n k].

j=1k= Lb1

We will also make the following assumptions throughout this

paper:

AS1. The mixture is soft in the sense that each observed

signal has a dominant contribution from a different source.This hypothesis is valid when each sensor is closer to onesource than to the others.AS2. Sources are almost identically distributed in the sense

that cumulants of the same order do not exhibit largedifferences in magnitude.As pointed out by several authors [2][6], the BSS problem hasmultiple valid solutions since sources can be separated witharbitrary orders, delays and scale factors. To avoid the last twoindeterminacies we will assume diag(B(z))=diag(I) anddiag(A(z)P)=diag(I), where I and P are the identity and an

A Gauss-Newton method for blind source separation of convolutive mixtures

unknown permutation matrix, respectively.

Figure 1. Feed-forward separation model.

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email