Research Of Independent Component Analysis Algorithm Of Higher Order Convergence

Independent Component Analysis (ICA, Independent Component Analysis) is an effective blind signal separation method developed in the recent years. Under the condition that we do not know the instantaneous aliasing parameters of the received signal, the technology recovers the source signal from the observation signal only on the basis of some basic statistical characteristics of input source signal. At present, it is widely used in the field of speech signal processing, image processing, wireless communication technology, biomedical engineering and others. Because of the wide application prospects of ICA, it attracts a lot of researchers to study thoroughly and gets further development and perfection.Firstly, this paper describes the basic theory of ICA algorithms, including the mathematical framework model of ICA, the data processing processes and so on. Secondly, several typical ICA algorithms and performance standards to measure algorithm are introduced. The typical ICA algorithm includes maximum entropy algorithm, minimum mutual information algorithm, fixed-point algorithm, gradient algorithm, the natural gradient algorithm and others.FastICA algorithm is carefully studied in this paper, especially the FastICA algorithm of third-order convergence and the fifth-order convergence. The advantages and disadvantages of the algorithm are analyzed. Aiming at the problem of sensitiveness of initial value, this paper improves effectively the algorithm and proposes two algorithms on the basis of the FastICA algorithm of third-order convergence and fifth-order convergence. One is the combination of steepest descent method and the FastICA algorithm. Another is higher order convergence FastICA on the basis of the relaxation factor. Simulation experimention results of image signal separation and voice signal separation show that the separation effect of the improved algorithm is superior to the effect of the traditional FastICA algorithm. Especially in the processing of initial value, it overcomes the problem of initial value sensitivity of the original algorithm and enhances balance and effectiveness of algorithm convergence.