基于零空间表示和最大似然的欠定盲源分离方法.docVIP

基于零空间表示和最大似然的欠定盲源分离方法 王荣杰1,3，詹宜巨2，周海峰3 (1. 中山大学 信息科学与技术学院，广东 广州 510006；2. 中山大学 工学院，广东 广州 510006；2. 集美大学 轮机工程学院，福建 厦门 361021) 摘 要：针对欠定情况下源数的估计、解混叠矩阵和源信号恢复关键技术，提出一种源数未知的欠定盲源分离算法，首先利用S变换和聚类技术相结合来估算源数和混叠矩阵，然后将源信号以零空间形式表示，再通过最大似然估计关于其后验概率以达到恢复源信号的目的。仿真实验结果表明了该方法不仅能同时分离服从超高斯分布和亚高斯分布的源信号，且它比其他传统的方法具有更优越的估计性能。 关键词：欠定盲源分离；S变换；模糊c均值聚类；零空间表示；最大似然 中图分类号：TN911.7 文献标识码：A 文章编号：1000-436X(2011)03- Underdetermined blind source separation based onnull-space representation and maximum likelihood WANG Rong-jie1,3, ZHAN Yi-ju2, ZHOU Hai-feng3 ( 1. School of Information Science and Technology, Sun Yat-sen University, Guangzhou 510006, China; 2. School of Engineering, Sun Yat-sen University, Guangzhou 510006, China;3. Marine Engineering Institute, Jimei University, Xiamen 361021, China. ) Abstract: Aiming to the estimation of source numbers, mixing matrix and source signals under underdetermined case, and a method of underdetermined blind source separation with an unknown number of sources was proposed. Firstly, an algorithm based on S transforms and clustering technology was introduced to estimate the number of sources and mixing mixtures. Then sources were represented as null space form, the recovery of source signals using a method based on maximum likelihood. The simulation results show that the proposed method can separate sources of super-Gaussian distrisutron and sub-Gaussian distribution, and compared to other conventional algorithms, estimated mixing matrix and separated sources with higher accuracy. Key words: underdetermined blind source separation； S transforms； fuzzy c-mean clustering; null-space representation; maximum likelihood 1 引言 近年来，鉴于盲源分离优越的假设前提，其模型被广泛用于数字通信、语音、图像处理、船舶工程、机器人导航和生物医学信号处理等领域[1~6]。所谓的盲源分离，就是在源信号和混合系统(或传输通道)等未知的情况下，仅根据源信号的统计特性，由观测到的混叠信号恢复出源信号。根据观测信号和源信号的个数，盲源分离可分为超(正)定盲源分离和欠定盲源分离。超正定盲源分离有很多成熟的方法，如ICA独立分量分析[7,8]、修正的Stone法[9]、阶统计代数法[10]等。但这些方法不适用于观测信号的个数少于源信号的个数的情况，欠定盲源分离(UBSS, underdetermined blind source separatio