A Robust Blind Source Separation Algorithm without Whitening the Observed Signals
Wang Hang-jun, Luming Fang · 2006
A novel robust blind source separation algorithm that uses the geodesic method is proposed. Different from many methods that only can treat the whitened observed data, the proposed algorithm can separate the unwhitened observed data, i.e., the original observed data. More importantly, the algorithm is robust to the outliers due to the adoption of novel density models, which are different to the ones that used by many other algorithms. Simulations on artificial generated data and real-world ECG data reveal that the proposed algorithm has fast convergence, high separation performance and robustness to the outliers, compared with some famous algorithms