A simple variable step algorithm for blind source separation (BSS)
Man-On Pun, Y. Hirai · 2002
Most of the existing online learning algorithms for blind source separation are confronted with the learning step-size problem. In this paper, we propose a way to decompose the demixing matrix into row vector space. From the row vector viewpoint, a variable step (VS) algorithm for the blind source separation problem is proposed. Instead of using a constant step-size, the VS learning algorithm adaptively changes the learning step-size of each row vector according to the convergence behaviors of its corresponding row vector. Our simulation results show that the new VS learning algorithm outperforms the conventional algorithms.