A Novel Approach of Estimating Mixing Matrix in Underdetermined Blind Source Separation Using Wavelet Coefficients

摘要：

In this paper, underdetermined blind source separation was probed. Firstly, our proposed searching-and-averaging-based method in time and frequency domain is expanded into the transform domain. This approach describes how to search some data points that are very close to the basis lines and use them to estimate the mixing matrix. Secondly, the observed data were sparsified through DWT, and the sparest subset wavelet coefficients were employed to estimate the mixing matrix by the novel approach. Finally, the mixing matrices, by which five speech signals were mixed into two or three observed signals, were successfully and exactly estimated.

作者: Ming Xiao Ronghua Li

作者单位: School of Electrics & Information Engineering South China University of Technology Guangzhou,China 5 School of Electrics & Information Engineering South China University of Technology Guangzhou,China 5

会议类型: 国际会议

会议名称: 第三届国际脉冲动力系统及应用学术会议

会议地点: 青岛

会议语种:英文

在线出版日期: 2006-07-21（万方平台首次上网日期，不代表论文的发表时间）