基于张量联合对角化的非平稳源卷积盲分离研究  

Convolutive Blind Separation of Nonstationary Sources Based on Joint Tensor Diagonalization

作　　者：王语凡 Wang Yufan(School of Automation, Guangdong University of Technology, Guangzhou 510006, China)

机构地区：[1]广东工业大学自动化学院,广州510006

出　　处：《黑龙江科学》2020年第16期1-5,10,共6页Heilongjiang Science

摘　　要：盲信号分离(Blind Source Separation)在很多领域有着广泛的应用。本研究提出一种类雅克比旋转的张量分解算法来提高源信号的分离性能。首先利用信号的自相关矩阵构造一系列张量,将盲信号分离问题转化为张量对角化问题,然后提出一种特殊参数结构代替Givens矩阵做类似雅可比连续旋转,估计对角矩阵,并求解了算法中参数的最优解。实验结果表明,该算法具有良好的分离效果。Blind Source Separation has been widely used in many fields.In this paper,a Jacobi-like Nonunitary Tensor Diagonalization Algorithm is proposed to improve the separation performance of the source signal.Firstly,a series of tensors are constructed through autocorrelation matrix of the signal.Blind signal separation problems are transformed into tensor diagonalization.A special parameter structure is proposed to replace the Givens matrix,estimate the diagonal matrix,and solve the optimal solution of the parameters in the algorithm.Experimental results illustrate the algorithm can get high separation performance.

关 键 词：张量对角化 卷积盲分离 非平稳信号 

分 类 号：TN911.7[电子电信—通信与信息系统]