Blind Source Separation of a Mixture of Pseudoperiodic Chaotic Signals

Abstract

The straightforward blind source separation (BSS) algorithm based on nonlinear phase-space reconstruction and nonorthogonal joint approximate diagonalization (JAD) of several time-delayed covariance matrices, estimated with one data matrix of first embedding dimension and second data matrix of every another embedding dimension, is investigated by applying it to mixed pseudoperiodic chaotic Rossler signals and Mackey-Glass signals. Simulation results show that this algorithm gives a good performance in the separation of mixed pseudoperiodic chaotic or similar to pseudoperiodic signals when each source has non-zero autocorrelation function for a non-zero time lag, i. e. when analysis based on the second-order statistics (SOS) is applicable. Algorithm leads to better performance than many widely known BSS algorithms including the very efficient iterative FastICA algorithm. Ill 2, bibl. 15 (in English; summaries in English, Russian and Lithuanian).