Generalized Complexity Pursuit

In this paper, we study the blind source separation (BSS) problem of temporally correlated signals via exploring the nonlinear temporal structure and high-order statistics of source signals. A BSS method based on the nonlinear predictability of original sources is proposed, which extends linear coding complexity used by the original complexity pursuit to nonlinear coding complexity. Simulations by nonstationarity sources verify the efficient implementation of the proposed method, especially its robustness to the outliers.