A novel and fast blind source separation algorithm for convolutive environment

Abstract

In this paper a novel and fast algorithm for the blind source separation in convolutive media is introduced. This method estimates multiple independent source signals, using only their set of received convolutive mixtures. The number of sources and the delays in the arrival of their echoes are unknown. The channel is estimated by calculating the channel matrix which is not achieved in some other CBSS methods. In this algorithm the independent component analysis (ICA) is used as the first step to separate the signals, lags and noise components. In the second stage, a purely second-order statistic approach estimates the source signals, which is a novel CBSS algorithm. This unique structure results in an efficient and accurate estimation. Another new feature of our approach is the implementation of a fast estimator. The channel variations are usually slow compared to the sampling rate. Therefore, the fast estimator separates the source signals using only the received signals at the sampling instant, based on the estimated channel. The channel matrix and the separated source signals are updated by repeating CBSS process at regular intervals. The permutation ambiguity, which is a common problem in many separation methods, is resolved in this algorithm. This new approach is simpler, faster, more accurate and needs less memory compared to some methods recently introduced by others.