Qualitative performance analysis of blind source extraction

In blind source separation, several techniques have been proposed to illustrate the qualitative performance of separation algorithms. However, in general, we assume ideal condition, no noise and linear mixing model. In this paper, a number of important performance analyses are discussed. It is shown that estimation of the mixing/demixing matrix should not be the main goal, in the noisy and nonlinear case. Instead, it is proposed to compare outcome of ICA algorithms with different proposed performance techniques, derived for known mixing model and extended to real-world data. In this work, the delay variance vector is suggested as the meaningful performance criterion. A simulation study that compare a few well known ICA algorithms and performance techniques applied to noise data are included.