Blind Extraction of Independent Signals from Their Linear Mixtures
Ju Liu · 2001
Observed signals are always the linear mixture of some independent components. Independent component analysis (ICA) is a novel technique for dealing with such a problem. Most of the existing algorithms separate individual independent sources simultaneously. In this paper, basing on the independence assumption of the original sources, we propose a new blind separating criterion, where the square of fourth-order cumulants of the sources are employed. We next develop an ICA approach which can sequentially extract independent components blindly one by one. A new deflation technique is used in this approach for removing the previously extracted signals from the mixture. Computer simulations show the validity of the proposed approach.