基于广义相关系数的后非线性盲信号分离算法

Blind source separation algorithm for post-nonlinear mixture based on generalized correlation coefficient

基于互信息最小化的独立性测度对各分离信号间的非线性相关度度量没有归一化的问题,提出一种基于广义相关系数的盲信号分离(BSS)算法.首先选取后非线性混叠模型(PNL)分析基于广义相关系数的独立性测度;然后采用Gram-Charlier扩展形式估计输出参数并获取评价几率函数,结合最陡下降法求得分离矩阵和参数化可逆非线性映射的算法迭代公式.仿真结果表明,采用所提出的算法能够定量分析各分离信号间的非线性相关程度,有效分离后非线性混叠信号.

According to the problem that the independence criterion based on the minimization of mutual information is not normalized, a blind source separation（BSS） algorithm for post-nonlinear mixture（PNL） based on general correlation coefficient is introduced in this paper. Firstly, the PNL is taken as an indraft point to summarize this algorithm, which is the more practicable approximation to realism rather than linear model, meanwhile the independence criterion based on the generalized correlation coefficient is discussed. Then score function based on a Gram-Charlier expansion of densities is proposed. Finally, combined with the steepest descent method, the computations of regular matrix and parametric nonlinear mapping are given. The simulation results show that the proposed method is effective in BSS for the PNL and for the ouantitative analysis of nonlinear correlation between variables.