Feature Extraction and Blind Separation of Convolutive Signals
In this paper, we propose a novel approach to blind separation of convolved signals assuming that source signals have temporal structures. First we introduce mathematical formulation of blind deconvolution problem. Blind separation problem is divided into two stages: recovering innovations and estimation of source signals. We employ natural gradient algorithm for blind deconvolution to recover innovation signals. By using the inverse filter of denuxing model, we cancel the mutual interference of different source signals. Furthermore, we employ the linear prediction method to extract the temporal structures of source signals and reconstruct source signals by using AR model. A computer simulation is given to demonstrate the effectiveness and validity of the proposed approach.
L.-Q. Zhang A. Cichocki
Laboratory for Open Information Systems RIKEN Brain Science Institute Hirosawa 2-1, Wako shi,Saitama 351-0198, JAPAN
国际会议
8th International Conference on Neural Information Processing(ICONIP 2001)(第八届国际神经信息处理大会)
上海
英文
827-832
2001-11-14（万方平台首次上网日期，不代表论文的发表时间）