A New FastICA Algorithm with Symmetric Orthogonalization

The separation of independent components from an array of mixtures is an interesting but difficult problem in signal processing. This paper re-examine the fastICA algorithm proposed by Hyvarinen and Oja for independent component analysis. The ways needed in fastICA algorithm for decorrelation the separating matrix can be deflationary or symmetric orthogonalization. Using the one-unit fastICA algorithm, this paper proposes a method to choose a suitable initial separating matrix of the fastICA algorithm with symmetric orthonormalization. The formed new algorithm can get faster speed without losing the intrinsic advantage of the original symmetric fastICA algorithm. Simulation results confirm the validity of the new algorithm