An algorithm for noisy Blind Source Separation with nonlinear autocorrelation and noise reduction using wavelet

Recently, Blind Source Separation (BSS) of received noisy signal has been great interest the in the field of signal processing. BSS can separate the original signals from their mixtures without any knowledge about the mixing process. The nonlinear autocorrelation function is used as an object function to separate the source signals from the noisy mixing signals that has to be maximized. Wavelet transform is a useful tool to maximize the nonlinear autocorrelation function and reduce the effects of noise. In this paper we will investigate the usage of wavelets to solve the BSS using LMS algorithm in the two scenarios. In the former scenario, the mixed signals were first separated and then the wavelet transform is used to eliminate noise effect. In the latter scenario, the wavelet packet transform of the mixed signal is obtained and then according to these packet signals, BSS is applied. To calculate the performance of the proposed algorithms, the parameter of Signal to Noise and Interference Ratio will be used. The source signals are selected from TIMIT database. Simulation shows wavelet in the latter scenario has better performance.

Keywords:

Blind Source Separations , Wavelet Packet Transform , Signal to Noise and Interference Ratio , LMS Algorithm , Nonlinear Autocorrelation