基于奇异值分解和均值聚类的单通道盲源分离算法研究  

Research on Single-channel Blind Source Separation Algorithm Based on Singular Value Decomposition and Mean Clustering

作　　者：黄小青 肖瑛 马艺伟 HUANG Xiao-qing;XIAO Ying;MA Yi-wei(School of Information and Communication Engineering, Dalian Minzu University, Dalian Liaoning 116605, China)

机构地区：[1]大连民族大学信息与通信工程学院,辽宁大连116605

出　　处：《大连民族大学学报》2021年第3期251-255,共5页Journal of Dalian Minzu University

摘　　要：提出了一种基于奇异值分解(Singular Value Decomposition,SVD)的均值聚类单通道盲源分离算法。首先将单通道信号利用SVD分解,依据中值准则进行滤波去除噪声分量,然后在去除噪声分量对应的特征值基础上,根据剩余SVD特征值重构对应分量信号作为盲源分离观测信号。将重构分量信号进行短时傅立叶变换(Short Time Fourier Transform,STFT)进行稀疏化处理,利用散点图判别源信号数目,最后采用均值聚类方法估计混合矩阵,以估计混合矩阵求逆作为分离矩阵实现单通道信号的盲源分离。利用计算机仿真结果证明了算法的有效性。A single-channel blind source separation algorithm based on Singular Value Decomposition(SVD)and mean clustering was proposed.The single-channel signal is decomposed by SVD and filtered according to the median criterion to remove noise components.Then,on the basis of removing the characteristic value corresponding to the noise components,the remained component signals are reconstructed according to the corresponding SVD characteristic values as the blind source separation observation signal.The reconstructed component signal is subjected to Short Time Fourier Transform(STFT)for sparse processing,and the number of source signals is determined by using the scatter diagram.Finally,the mean clustering method is used to estimate the mixing matrix,the inverse of which is used as the separation matrix to realize the blind source separation of single-channel signals.The result of simulation proves the effectiveness of the algorithm.

关 键 词：盲源分离 奇异值分解 聚类 短时傅立叶变换 

分 类 号：TN911.7[电子电信—通信与信息系统]