Pre-filtering non-stationary signals to improve blind source separation
Allan Kardec Barros, Noboru Ohnishi · 2002
Much work has been carried out concerning blind source separation (BSS) that uses the principle of independent component analysis (ICA), which stands as an elegant, simple and powerful tool to BSS. However, the algorithms previously proposed are still sensitive to nonstationarity, which alters in a bad sense the separation. We propose here a simple pre-processing to the mixed signals based on low-pass filtering that ensures a much better performance for the ICA algorithms.