Blind Source Separation Based On Generalized Autocorrelations

Over the past decades, Blind source separation (BSS) has received much research attention because of its potential applicability to many problems. Generally, classical BSS methods include Blind source parallel separation and Blind source extract. The so called Blind source parallel separation consists of simultaneous recovering all the independent components from the linear mixtures, but recovering all the source signals will take a long time. And Blind source extract recovers the independent components from the linear mixtures in turn.The BSS problem is systematically addressed at first. Based on the signal model, we analyze the indeterminacy inherent in BSS and the basic assumptions of the BSS problem. The contrast function theory and the optimal algorithm theory are investigated. The Blind source extract theory and algorithm are also investigated emphatically. Second, based on maximizing Kurtosis,We derive the gradient algorithm and a fast fixed-point algorithm.It is directed to the BSE problem when a desired source signal has temporal structures. Using the temporal characteristics of sources, we develop objective functions based on the generalized autocorrelations of primary sources. Maximizing this objective function, we design a BSE fixed-point algorithm and further give its stability an analysis in this paper. In order to overcome the defect of algorithm error heredity backward, based on the extract algorithm, a parallel separate algorithm is designed. Computer simulations show the effectiveness of this algorithm.