Factorization based blind identification and separation of nonstationary seizure signals

In this paper, a new blind identification and source separation method, which explicitly uses nonstationarity of the sources in separation of their instantaneous mixtures, is developed and effectively used for separation of seizure signals. In this approach tensor factorization concept has been exploited for which the optimization steps require nonstationarity of the sources. Based on this method simultaneous blind separation and identification is achieved. The algorithm is applied to mixtures of synthetic nonstationary sources and for separation of seizure brain sources from natural EEG signals and the results are compared with those of some recently published methods.