Signal separation by independent component analysis and fuzzy estimators

Abstract

Independent component analysis (ICA) is a developing field of interest for researchers in signal and artificial neural networks. ICA is an intelligent signal processing extension to the principal component analysis that becomes sensitive to non-Gaussian higher-order statistics. This paper presents the motivation of ICA and a treatment of the theory with a guiding example. The limitations of current ICA algorithms are discussed in general and the possible benefits of developing fuzzy engines as ICA estimators are discussed. In particular, a Mamdani-type fuzzy inference system for determining an optimal ICA rotation of whitened two-dimensional uniform noise is implemented as an example of the feasibility of this new direction in ICA.