Linear Strength-based Algorithm to Sparse Dependent Sources Separation

摘要

In this paper, we present a novel linear strength-based method for the blind separation of correlated sources, which do not fulfill the independence assumption. In contrast to standard independent component analysis methods, we consider sparse of sources in the time-frequency space, also known as Sparse Dependent Component Analysis (SDCA). First, the frequency sub-bands of the observed data are obtained by wavelet packet transform. Next, a Linear Strength (LS) method is employed to detect the sparest cluster of frequency sub-bands. Finally, the mixing matrix can be estimated based on principle component analysis. Through simulation, we demonstrate consistent performance in terms of robustness as well as the effectiveness of the proposed LS-SDCA algorithm.