An ICA Algorithm Based on Generalized Gaussian Model for Evoked Potentials Estimation

Abstract

Independent Component Analysis (ICA) is a recently developed Blind Source Separation (BSS) algorithm based on single observation sample. The success of the algorithm depends on its probability density model can better fit the signal inherent statistical distribution. For the problem that existing algorithms can not well fit the probability density model of source signals, this paper proposes an ICA algorithm based on the Generalized Gaussian Model (GGM). This new algorithm, combining with the Maximum Likelihood of ICA, utilizes GGM to fit the signal probability density model, and uses it to estimate Auditory Evoked Potential (AEP). Experiments show that the algorithm can fit the signal inherent statistical distribution very well and estimate purer Evoked Potential (EP) signals more effectively.