时变混合系统的在线FastICA算法  

On- Line FastICA Algorithm for Time- Varying Mixing System

作　　者：陈海平[1] 张杭[1] 路威[1] 张江[2] 

机构地区：[1]解放军理工大学通信工程学院,江苏南京210007 [2]总参第63研究所,江苏南京210007

出　　处：《通信技术》2014年第2期136-140,共5页Communications Technology

基　　金：国家自然科学基金:欠定条件下基于扰信分离的信干比增强方法研究(No.61001106)~~

摘　　要：现有的多数盲源分离(BSS,Blind Source Separation)算法都是假设混合系统是时不变的,然而在实际的通信系统中混合系统常常是时变的。传统的快速不动点(FastICA)算法具有快速收敛的优点,但是不能直接用于处理混合系统时变的盲源分离问题。为了提高盲源分离算法的收敛速度和对时变混合系统的跟踪性能,改进了传统FastICA算法,将混合信号分段,在各段样本中估计峭度并采用批处理的方法进行分离。仿真实验表明,改进后的FastICA算法能在时变环境中跟踪混合系统的时变,并能有效地抗多音干扰。Most existing BSS（ Blind Source Separation） algorithms are developed by assuming that the mix- ing matrix is invariable. However, the mixing matrix is commonly time-varying in practical communication system. The traditional fast fixed-point（FastICA） algorithm cannot directly separate the sources mixed in time-varying system, although is very fast in convergence. Thus in order to accelerate the convergence rate and improve the tracking performance for time-varying system, the FastICA algorithm is modified by divid- ing samples into several subsections, thus to estimate the kurtosis and separate the sources off-line in each subsection. Simulation results indicate that the modified FastICA algorithm could track the time-varying system effectively.

关 键 词：盲源分离 时变 快速不动点算法收敛速度在线算法 

分 类 号：TN911.7[电子电信—通信与信息系统]