一种非平稳卷积混合信号的自适应盲分离算法  

An Adaptive Blind Source Separation Algorithm of Non-stationary Convolutive Mixtures

作　　者：何文雪[1] 王林[1] 谢剑英[1] 

机构地区：[1]上海交通大学自动化所,上海200030

出　　处：《系统仿真学报》2005年第1期196-198,202,共4页Journal of System Simulation

基　　金：国家863基金项目(2001AA422420-02)

摘　　要：基于多时延解相关准则并利用自然梯度原则推导出一种非平稳信号在卷积混合情况下的自适应盲分离算法,仅通过对有限个观测时延样本的相关矩阵进行Frobenius范数最小化运算得到分离矩阵,算法计算量低,可实现信号连续处理,适用于在线运算。仿真结果表明算法仅需较低的分离滤波器长度和少量的时延样本即可获得良好的分离性能。A new adaptive blind source separation algorithm of non-stationary convolutive mixtures is presented by using multiple time-delay decorrelation and natural gradient rule. Only a small number of samples are used to minimize the Frobenius norm of their correlation matrices, so the algorithm has low computing cost and more adapt to on-line blind source separation. Simulations showed that the algorithm could get high separation performance even when low separation filter length and small number of samples are used.

关 键 词：盲分离 非平稳信号 自然梯度 代价函数 

分 类 号：TP274.2[自动化与计算机技术—检测技术与自动化装置]