Non Linear Blind Source Separation Using Cross Power Spectral Density and Cross Correlation Coefficient

Abstract

Blind source separation (BSS) is an emerging technique, which enables the extraction of target speech from observed mixed speeches. BSS algorithm sare based on restoring the statistical independence of the source signals. This paper is concerned with blind source separation of convolutive mixtures of acoustic signals, especially speech in which the source signals are instantaneously mixed in unknown nonlinear processes. A statistical an d computational technique, called independent component analysis (ICA), used for BSS, is examined by achieving maxim um entropy at the output end, also cross power spectral density and
cross correlation coefficient o f observed unmixed signal an d source signal is check to conform the received signal is the original signal. Effectiveness o f the proposed method is validated in BSS of male and female voice signals gets signal to interference ratio
8.75 with the proposed, algorithm , 7.03 an d 5.12 with  algorithm based on Mutual Information and non causal FIR filter. This paper synthesizes proposed algorithm on Field Program m able Gate Array (FPGA) - Xilinx VIRTEXV1OOOE that executes a t the maximum frequency o f 12.288 MHz. The performance com parisons between the proposed an d another two ICA-related FPGA implementations [4][5], show that the proposed FPGA implementation has potential in performing complicated algorithms on large volume datasets.