A Fast Algorithm for Blind Separation of Complex Valued Signals with Nonlinear Autocorrelation

—Blind source separation of complex valued signals has been a hot issue especially in the field of multi-input/multi-output (MIMO) digital communications. Many contrast functions based on the nonlinear structure of the signals have been proposed to extract the unknown sources. However, these researches usually focused on the real-valued case, but ignoring the complex problem. This paper proposes a novel algorithm based on Newton iterations to solve the complex-valued case. The method has a potential capability of extracting complex sources with nonlinear autocorrelation. We also analyze the convergence conditions of the algorithm in theory. Numerical simulations for artificial complex signals corroborate the efficiency of the proposed method. Moreover, our algorithm performs more robust with lower computational cost than classical cumulant-based approach using the nonstationarity of variance (CANSV). Finally, experiments for the separation of single sideband signals illustrate that our method might have good prospects in real-world applications.