A normalised kurtosis-based algorithm for blind source extraction from noisy measurements

In blind extraction of independent sources, the normalised Kurtosis is a normally used cost function for the cases without the initial prewhitening. The applications of this method are, however, limited to noise-free mixtures, which is not realistic. We therefore address this issue and propose a new cost function based on the normalised Kurtosis, which makes this class of algorithms suitable for noisy environments, a typical situation in practice. The proposed method is justified by a theoretical analysis and the performance of the derived algorithm is demonstrated by simulations.