摘要： 当源信号各分量差异较大或全局矩阵为非行元素优势矩阵时,基于串音误差的变步长等变自适应分离(EASI)算法难以正确评价分离效果,导致步长选取错误。针对该问题,提出一种改进的变步长EASI算法。通过计算基于串音误差的步长平方根适当增大步长,同时定义信号分离度对步长做进一步调整,减小因串音误差分离结果评价不准确而导致步长错误变小的影响。仿真结果表明,该算法具有较好的步长调节能力,相对传统EASI算法和基于串音误差、基于串音误差步长平方根、基于分离度的变步长EASI算法分离效果更佳。

关键词: 盲源分离, 变步长, 等变自适应分离算法, 串音误差, 分离度

Abstract: When the source signal consists of components with large energy difference or the global matrix is a non-row element dominant matrix,the variable step-size Equivariant Adaptive Separation via Independence(EASI) algorithm based on crosstalk error cannot correctly evaluate the separation effect,resulting in incorrect step selection.Aiming at this problem,this paper proposes an improved variable step-size EASI algorithm.It calculates the square root of the step-size based on crosstalk error to increase step-size appropriately.Meanwhile,according to the defined separation degree,it adjusts the step-size so as to reduce the impact of small step caused by incorrect separation effect evaluation based on crosstalk error.Simulation results show that the proposed algorithm has good step-size adjustment ability.Compared with traditional EASI algorithm and variable step-size EASI algorithm based on crosstalk error,the step-size square root of crosstalk error or the separation degree,its separation effect is better.

Key words: Blind Source Separation(BSS), variable step-size, Equivariant Adaptive Separation via Independence(EASI) algorithm, crosstalk error, separation degree

中图分类号: 

• TP391