An efficient and robust algorithm for BSS by maximizing reference-based negentropy

A family of contrast criteria referred to as “referenced-based” has been recently proposed for blind source separation (BSS), which are essentially the cross-statistics or cross-cumulants between estimated outputs and reference signals. These contrast functions have an appealing feature in common: the corresponding optimization algorithms are quadratic with respect to the searched parameters. Inspired by this reference-based scheme, a similar contrast function is constructed by introducing the reference signals to negentropy, based on which a novel fast fixed-point (FastICA) algorithm is proposed in this paper. This new method is similar in spirit to the classical FastICA algorithm based on negentropy but differs in the fact that it is much more efficient in terms of computational speed than the latter, which is significantly striking with large number of samples. What is more, this new algorithm is more robust against unexpected outliers than those cumulant-based algorithms such as the FastICA algorithm based on kurtosis. The performance of this new method is validated through computer simulations.