一种功能增强的信号源盲分离新算法

An Enhanced New Algorithm for Blind Source Separation

提出了一种新的信号源盲分离算法。该算法不仅能够有效地求解源信号中同时存在超高斯信号和亚高斯信号的杂系混合 (hybrid mixture)的信号源盲分离问题 ,而且能够准确地估计未知信号源的数目 ,因而具有比一般盲分离算法更广得多的应用范围。对于杂系混合盲分离问题 ,一般的盲分离算法往往不能求解。现有的绝大多数盲分离算法总是假设信号源的数目是已知的 ,这在多数背景下是不适用的 ,从而大大限制了信号源盲分离这一信号处理方法的实际应用范围。通过利用概率密度函数估计的核函数法对信号源盲分离算法中的评价函数 (score func-tion)直接进行估计 ,并利用混合信号样本自相关矩阵的秩数与未知信号源数目的内在联系 ,使这两个关键性的问题在所提出的盲分离新算法中都得到了非常成功地解决。

A new blind source separation(BSS)algorithm is proposed.Conventional BSS algorithms cannot separate hybrid mixture of source signals that includes both super Gaussian and sub Gaussian sources.More importantly,most BSS algorithms always assume that the number of source signals is known a priori.This simple assumption may well be impractical in many cases,hence narrows the scope of the applications of BSS of the real world signal processing problems.The new algorithm proposed in the present paper successfully overcome these two main drawbacks by using kernel density estimation method to directly estimate the score functions in some BSS algorithms,and exploring the inherent relationship between the number of source signals and the rank of the averaged sample correlation matrix.The algorithm is not only capable of separating hybrid mixtures,but also capable of estimating the number of source signals accurately.This makes the algorithm more applicable than other existing BSS algorithms.Simulation results show the efficacy of the proposed methods.