An introduction to EIP-NLMS post-processing method for cross-talk cancellation in BSS outputs

In this paper, we propose a new post-processing method to suppress the residual cross-talk of Blind Source Separation (BSS) outputs. This method relies on the existence of the so called Exclusive Inactivity Periods (EIPs) in the source signals during which all sources are active except one. After detecting EIPs, we use the Normalized Least-Mean-Square (NLMS) algorithm to estimate the cross-talk in these regions. Simulation results show the ability of the proposed method in suppressing the residual cross-talk without distorting the target signal and improving the SIR about 3 dB compared to the NLMS post-processing.