源信号数目未知的自然梯度盲信号分离算法

Natural Gradient Algorithm for Blind Source Separation with Unknown Source Number

总结了源信号数目未知的盲信号分离自然梯度算法,得到自然梯度算法发散的原因,分离矩阵的各行沿混合矩阵转置的零空间方向无效的冗余移动。借助投影自然梯度算法,从理论上证明,冗余分量的范数随迭代次数的增加呈指数分布。

A summary of the natural gradient algorithm for blind source separation with an unknown number ofsources is presented and the conclusion is drawn that it is the redundant movement of separating matrix in the zerospace of the transpose of mixing matrix that causes the divergence. By the projected natural gradient algorithm, wetheoretically prove that with the increase in the number of iterations, the norm of the redundant subjects to an expo-nential distribution. Keywortls blind source separation; natural gradient algorithm; the redundant; exponential distribution.