摘要
针对欠定瞬时混合模型,提出了一种基于时频域加权张量分解的欠定盲源分离方法。该算法利用短时傅立叶二次分布无交叉项及Wigner-Ville分布高分辨率的特性,在传统最小二乘代价函数基础上对WVD自由项时频点所构成张量进行加权调整,解决可能存在的数据丢失问题,同时施加Tikhonov准则处理由二次分布的边缘聚集性所引起的负值,采用LM算法最小化代价函数估计出混合矩阵,实现源信号的有效分离。
A new time-fi'equency approach to the tmderdetermined blind source separation was proposed based on the instanta- neous mixture model. The algorithm takes advantages of the characteristics of no cross-terms in STFT and high TF resolution in WVD. Based on the least square cost function, we use a modified weighted tensor constituted by auto-terms TF points to handle the possibility of missing values and deal with the negative values which may be caused by quadratic distribution marginal integration. Fi- nally, we minimized the cost function by LM algorithm to estimate the mixture matrix and separate the source signals effectively.
出处
《激光杂志》
CAS
CSCD
北大核心
2014年第4期20-22,共3页
Laser Journal
基金
国家自然科学基金(NO.61163066
NO.60902074)
关键词
欠定盲分离
时频分析
Tikhonov准则
加权张量分解
Underdetermined Blind Source Separation
Time-Frequency Analysis
Rule of Tikhonov
Weighted tensor decompo-sition
作者简介
张明君(1986-),女,2011级硕士研究生,主要研究方向为盲信号处理,信息安全.
通讯作者:郭文强(1975.),男,博士,教授,主要研究领域为人工智能,信息安全.