Research And Application Of Blind Signal Separation Algorithm Based On Iterative Steffensen

In blind signal processing, researching on Bland Source Separation(BSS) is a new research rapidly developed in recent decades. In the case of the source signal and the transmission channel of signal that is the mix of the respective signals are unclear, separating the observed signal into the desired source signal is the main content, which based on some statistical properties of the source signal and knowledge of information theory.The main application of BSS are separation and identification in speech signal, data communication and signal processing, image processing and recognition, geo-spatial information processing, biomedical signal processing, text analysis and treatment, etc. In recent years, particularly BBS has applied to wireless communications, image processing, and biomedical engineering.Independent Component analysis(ICA) is a fast method of blind signal separation based on each source is independent of BSS method, it is one special algorithm of BSS. For now, the main methods of ICA are iterative estimation algorithm based on information theory and algebraic algorithm based on statistical principles, both of which independently and require the source signal is not Gaussian signal. Based on information theory in the research, the researchers from various countries given many estimation method in the mutual information criterion, maximum likelihood estimation and maximum entropy. More common are the maximum likelihood estimation, Infomax, the fast ICA algorithm independent on Component analysis(Fast ICA) and so on. Generally based on algorithm of the statistical theory, it has second- and fourth-order cumulant algorithm.FastICA is based on maximum non-Gaussian of the maximization signal, and using algorithm of fixed-point to solve the maximum of non-Gaussian. The algorithm uses Newton iterative method to calculate the value of multiple samples, so each iteration can separate a independent separation component. FastICA is to be widely adopted because of fast separation speed.In this dissertation, it describes the basic principle and method of blind source signal processing as well as ICA of BSS, and focus on research FastICA. Based on FastICA of largest negative entropyin a research, a improved FastICA based on a negative entropy is proposed, inserting attenuation factor and based on iterative theory of steffensen instead of Newton iteration, to solve the computational complexity, convergence accuracy and limited speed of Newton iteration. The improved algorithm by MATLAB simulation indicates the improvement in terms of separation speed and efficiency are obtained. At the same time, the improved algorithm in actual mixed speech signal blind separation, the experimental results also proved the superiority and feasibility of it.