Independent component analysis based on gradient equation and kernel density estimation

In this paper, a new algorithm is proposed for linear instantaneous independent component analysis. This new algorithm is based on solving the gradient equation, and an iterative method is introduced to solve this equation efficiently. To make the proposed algorithm adaptive to source distributions, the density functions as well as their first and second derivatives are estimated by kernel density method. Empirical comparisons with several popular independent component analysis (ICA) algorithms confirm the efficiency and accuracy of the proposed algorithm.