盲信号分离模型的混叠矩阵估计算法  

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

作　　者：傅予力[1] 谢胜利[1] 何昭水[1] 

机构地区：[1]华南理工大学电子与信息学院,广东广州510640

出　　处：《华中科技大学学报（自然科学版）》2007年第9期94-97,共4页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(60674033);国家杰出青年自然科学基金资助项目(60325310);广东省自然科学团队研究项目(04205783);广东省自然科学基金资助项目(051035530;5006508);科技部重大基础前期研究专项(2005CCA04100)

摘　　要：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.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.

关 键 词：盲信号分离 自然梯度 欠定混叠 过定混叠 适定混叠 

分 类 号：TN911[电子电信—通信与信息系统]