Blind identification and separation of sources with sparse events

In this paper, a new tensor factorization method based on k-SCA [1] approach is developed to solve the under-determined blind identification (UBI) problem where k sources are active in each signal segment. Similar to k-SCA methods we assume our k is equal to the number of sensors minus one. This approach improves the general upper bound for maximum possible number of sources in a second order underdetermined blind identification method called SOBIUM. The method is applied to the mixtures of synthetic signals and the results are illustrated. Compared to the recently developed SOBIUM approach, the proposed method is able to identify the channels for more number of source signals. Using the estimated mixing channels the separation of sources is also easily possible.