Neural Information Processing - Letters and Reviews
Vol. 10, No.11, November 2007
ICA Methods for Blind Source Separation of Instantaneous Mixtures:A Case Study
Pradipta Kishore Dash
The paper presents comparative assessment of Blind Source Separation methods for instantaneous mixtures. The study highlights the underlying principles of Principal Component Analysis (PCA) and Independent Component Analysis (ICA) in this context. These methods have been tested on instantaneous mixtures of synthetic periodic signals, monotonous noise from electromechanical systems and speech signals. In particular, methods based on Nonlinear PCA, Maximum Entropy, Mutual Information Minimization and Fast ICA, have been compared for their separation ability, processing time and accuracy. The quality of the output, the complexity of the algorithms and the simplicity (implementation) of the methods are some of the performance measures which are highlighted with respect to the above signals.
Keywords - ICA, blind source separation, nonlinear PCA