A source adaptive independent component analysis algorithm through solving the estimating equation

摘要

In this paper, a source adaptive algorithm for linear instantaneous independent component analysis is proposed. This new algorithm is based on solving the estimating equation through Newton’s method where no learning rate is needed which makes the proposed algorithm very easy to use. To achieve the source adaptivity, the density functions as well as their first and second derivatives are estimated by modified kernel density method. Empirical comparisons with several popular ICA algorithms confirm the efficiency and accuracy of the proposed algorithm.