A reference suite design for blind signal separation
Markus Borschbach · 2005
Abstract:- According to the common underlying mathematical model for Independent Component Analysis (ICA), the fulfillment of a BSS for either a linear and scalar type of composition, a convolutive and linear type or a nonlinear type has different conditions. So far, several approaches have been developed in the last decades for stationary and non-stationary data. To identify key research priorities, the different origins of neural network approaches for BSS are briefly reviewed and divided by classes of specific theoretical and application features. A principal guideline for the design of reference data sets for the comparison of all the existing ICA methods by its individual strengths and weaknesses for performing BSS is developed.