盲信号分离模型的混叠矩阵估计算法

针对传统盲信号分离方法通过估计分离矩阵实现盲信号分离难以同时适应适定、欠定和过定模型的问题,给出了一种新的方法,直接估计混叠矩阵实现盲分离.首先给出估计混叠矩阵的梯度学习公式,并分析了该梯度算法对适定模型的有效性,然后将它推广到过定混叠和欠定混叠模型,从而得到了一种适用于各种盲分离模型的混叠矩阵估计算法.仿真例子检验了所提出的算法在适定情形下与原有算法有类似的特性,而又可以同时适应过定和欠定模型.

An estimate algorithm of the mixing matrix in blind source separation models

Typically,blind source separation(BSS) is achieved by estimating the demixing matrix of a model.This method cannot suit the cases of well-posed,underdetermined and over-determined models.To avoid the disadvantage of typical method,a new method via estimating the mixing matrix is proposed.At first,a uniform gradient based learning algorithm is given.The performance of the algorithm is discussed for the well-posed models.Then,the algorithm is extended to over-determined and underdetermined models.Finally,we obtained an algorithm that can be used to the three kinds of BSS models.Simulations illustrate that the proposed algorithm has similar performance as the typical algorithm for well-posed models,and suits over-determined/ underdetermined models as well.