基于最大负熵的复值信号独立成分分析的稳健技术研究  

Research on the Robust Technique of Complex-valued Signal Independent Component Analysis based on Maximum Negative Entropy

作　　者：冯平兴 FENG Pingxing(School of Network&Communication Engineering,Chengdu Technological University,Chengdu 611730,China)

机构地区：[1]成都工业学院网络与通信工程学院,成都611730

出　　处：《成都工业学院学报》2025年第2期43-47,共5页Journal of Chengdu Technological University

基　　金：成都工业学院教改项目(202403022C)。

摘　　要：稳健性是信号处理中的一个重要考虑因素,特别是对独立成分分析(ICA)的分离性能有显著影响。因此,提出一种基于最大负熵的复值信号稳健近似多项式,用于改善复值ICA的稳健性。在实际信号处理领域中,使用ICA分离复值信号是信号处理中常见的技术;相应地,利用负熵度量非高斯性是ICA中的一种常用方法。然而,负熵的估计性能极易受观测信号中的异常值影响。采用稳健估计理论和多项式展开方法,提出的函数可以在观测值受到异常值污染时,负熵仍然保持稳健的估计性能(Amari指数小于-10)。此外,结合所研究的近似方法,复值ICA算法在源信号与异常值混合的情况下仍能保持良好的分离性能(10 log I A<-10),证明了该方法的有效性。Robustness is an important consideration in signal processing,especially for the separation performance of independent component analysis(ICA),which has a significant impact.A robust approximation polynomial for complex-valued signals based on maximum negative entropy is proposed to improve the robustness of complex-valued ICA.In the field of practical signal processing,using ICA to separate complex-valued signals is a common technique in signal processing;correspondingly,using negative entropy to measure non-Gaussianity is a commonly used method in ICA.However,the estimation performance of negative entropy is highly susceptible to outliers in the observed signal.By using robust estimation theory and polynomial expansion method,the proposed function can maintain robust estimation performance of negative entropy even when the observed signal are contaminated by outliers(Amari index less than-10).In addition,combined with the approximate method studied,the complex-valued ICA algorithm can still maintain good separation performance(10 log I A<-10)even when the source signal is mixed with outliers,demonstrating the effectiveness of this method.

关 键 词：独立成分分析 稳健性 负熵近似 复值信号 

分 类 号：TN95[电子电信—信号与信息处理]