A k-subspace based tensor factorization approach for under-determined blind identi??cation

In the paper, a novel k-subspaces based tensor factorization method is developed to tackle the underdetermined blind source separation (UBSS) and specially underdetermined blind identification (UBI) problems where k sources are active in each signal segment. This approach improves the general upper bound for maximum possible number of sources both in UBI and UBSS problems. The method is applied to mixtures of synthetic and real signals and the results are illustrated. Compared with other well-established approaches, the proposed method is able to identify the channels and separate the sources for more number of source signals.