基于滤波器阶数估计的卷积盲分离算法  

Convolution blind separation algorithm based on filter order estimation

作　　者：付卫红[1,2] 李丹 Fu Weihong;Li Dan(School of Telecommunications Engineering, Xidian University, Xi＇an 710071, China;Collaborative Innovation Center of Information Sensing and Understanding, Xi＇an 710071, China)

机构地区：[1]西安电子科技大学通信工程学院,陕西西安710071 [2]西安中电科西电科大雷达技术协同创新研究院有限公司,陕西西安710071

出　　处：《华中科技大学学报（自然科学版）》2018年第6期116-121,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61201134)

摘　　要：为解决时域卷积盲分离中存在的初值选取以及方阵局限问题,提出了一种基于滤波器阶数估计的非正交联合块对角化算法.该算法通过对观测信号自相关矩阵的相邻特征值比值设阈值,实现了滤波器阶数的估计和对角块个数以及维数的自适应选取;引入预白化对非方阵的等效混合矩阵进行降维处理,消除了非正交联合块对角化算法中等效混合矩阵必须为方阵的局限性.仿真结果表明：提出的滤波器阶数估计方法准确率高,有效解决了非正交联合块对角化算法在卷积混合盲分离中的初值选取问题,并且分离效果要优于经典非正交联合块对角化算法和多目标优化非正交联合块对角化算法.In order to solve the determination of initial value and square-limited problem in time-domain blind source separation of convolutive mixtures,based on the filter order estimation,a new nonunitary joint block diagonalization algorithm was proposed.The algorithm set a threshold value of adjacent characteristics ratio to estimate filter order,which realized the adaptive selection of diagonal block number and dimension.The algorithm introduced pre-whitening operation to realize dimensionality reduction of the equivalent mixed matrix, and eliminated the limitation of mixed matrix which must be square in nonunitary joint block diagonalization algorithm.Experimental results show that the algorithm of filter estimation accuracy is higher than other algorithms,effectively solves the initial value selection problem of nonunitary joint block diagonalization algorithm,and separated effect is superior to classic nonunitary joint block diagonalization algorithm and multicriteria optimization for nonunitary joint block diagonalization algorithm.

关 键 词：盲源分离 滤波器阶数估计 自适应 联合块对角化 预白化 

分 类 号：TN911.7[电子电信—通信与信息系统]