Blind Source Separation for Noisy Time Series by Combining Non-Gaussianity and Time-Correlation

This paper addressed the separation of noisy time series (noisy signals with time structure). Based on the non-Gaussianity of innovations, we first present an objective function with negentropy forms about innovations of time series. Furthermore, this criterion is extended for the noisy time series separation through combing Gaussian moments into it. Maximizing this objective function, a simple blind source separation algorithm is presented. Validity and performance of the described approach are demonstrated by computer simulations.