基于复信息量的频域多通道盲解卷积算法  

Frequency Domain Multi-channel Blind Deconvolution by Complex Informax

作　　者：许海翔[1] 丛丰裕[1] 史习智[1] 

机构地区：[1]上海交通大学机械系统与振动国家重点实验室,上海200240

出　　处：《数据采集与处理》2008年第3期278-283,共6页Journal of Data Acquisition and Processing

基　　金：国家重点实验室开放式基金(VSN-2006-04)资助项目;上海市科委基础研究(05JC14026)项目

摘　　要：基于复信息量,发展了一种频域多通道盲解卷积算法。第一阶段,将频域内复变量的伪协方差矩阵融入到分离算法中,能完全实现非线性不相关,从而使得算法性能得到显著提高。第二阶段,通过Kullback-Leibler(KL)距离来解决次序不确定性问题。将复变量出现的频度与复变量幅值的均匀分布相结合的复变量概率密度函数能更加准确地计算KL距离,使得该方法的性能得到改善。空旷场地下的仿真试验证明了所提出的算法更加有效。Based on the complex informax, a two-stage frequency domain approach is explored to separate the convolutive mixtures. At the separation stage, the frequency domain pseudo-co-variance is added into the popular natural gradient separation algorithm to decorrelate two improper complex-valued vectors, and then the independent complex-valued components are generated more efficiently. The simulation on separating complex-valued mixtures validates the effectiveness of the method. At the permutation correction stage, the Kullback-Leibler （KL） distance is used to resolve the permutation. The probability density function （PDF） of the unmixed signal is estimated by two factors. A parameter is computed to measure how frequently complex-valued data appears among signals, and it is the ratio of absolute value of a data to the sum of all data absolute values. It is more accurate to compute the KL distance because the probability is entirely based on the data samples. The simulation in a broad scene proves that the method is effective.

关 键 词：多通道盲解卷积 伪协方差矩阵 复变量概率密度函数 KL距离 

分 类 号：TN911.72[电子电信—通信与信息系统]