THE ELEVENTH IEEE SENSOR ARRAY AND MULTICHANNEL SIGNAL PROCESSING WORKSHOP (IEEE SAM2020)

297 / 2020-01-05 22:50:00

Adaptive Relative Newton Method for Blind Sparse Source Separation

IEEE SAM 2020: Full Regular Papers > Blind source separation and channel identification

This paper considers the problem of Blind Source Separation (BSS). Most of the proposed BSS techniques rely on the assumption that source signals are independent or at least uncorrelated. Unfortunately, these assumptions are not true in many applications where source signals usually show slight or strong dependence. In this paper, we propose to use the sparsity of signals which can be in the time domain or in a transformed domain, as a contrast tool to separate possibly dependent source signals. In particular, we investigate the adaptive context where the mixing matrix changes over time. The proposed algorithm which is based on the relative newton method, guarantees low computational complexity necessary in the adaptive case. Numerical simulations have shown the superiority of our algorithm as compared to the state of the art solutions.