Blind Source Separation Of Noisy Signals Based On Wavelet Transform And Independent Component Analysis

Blind source separation techniqueis a hot issue inspeech signal processing,especially separating each source signal in noisy conditions, general blind separation methods can not solve the noise disturbance.Filter as traditional denoising methods, can only filter out part of the noise, and in order to remove noise needs to sacrifice a part of the signal. Compared with the wavelet threshold denoising method, the computation of wavelet modulus maximum denoising is too large, although the denoising effect is good, but not worth using this method. But wavelet threshold denoising that needing to be improving, hard threshold function is not discontinuity at the threshold. Soft threshold denoising method can overcome the problem of discontinuous, but the coefficient reconstruction with real coefficients increase with the function value does not decrease. This paper presents an improved algorithm based on threshold denoising, improved function based on soft threshold function, from the experimental data finding that the improved algorithm has some help for denoising Based on the basic conditions of blind signal processing, wavelet analysis combined with the respective advantages of independent component, Signal preprocessing before separation, reducing the influence of the noise on the blind separation algorithm to the minimum. The method is blind separation algorithm of wavelet and independent component analysis. First of all, before the introduction of the wavelet denoising method, focuses on the history from the development of Fu Liye transform and the denoising principle of wavelet transform, analysises the threshold selection and quantization rules how to affect the denoising effect,analysis of present two kinds of commonly used threshold function (hard threshold, soft threshold),a new threshold function is proposed, the improved function to avoid the disadvantages of above two kinds of function, through the experimental analysis, the improved method is better than the above two methods.This paper also analyses the FastICA algorithm, the algorithm to find the optimal solution by using the Newton iterative method, but this method also has inherited the Newton iteration, the convergence of the algorithm is determined by the first iteration, if the first value is the improper selection may result in blind source separation failure. Someone proposed a improved algorithm, the Newton iterative method with one-dimensional search, make it according to the one-dimensional search direction iteration, improved stability, but also reduces the convergence speed. This paper presents an improved algorithm, proposed the Newton iteration method with high order convergence,then introduce one-dimensional search to ensure the correct direction.This method overcomes the short comings of Newton iteration,and weaken the adverse effect of damped Newton method.