Combined blind/nonblind source separation based on the natural gradient

It is a known fact that blind algorithms have convergence times of an order of magnitude longer than their nonblind counterparts. However, as shown in this letter, the knowledge of a subset of signals can greatly accelerate the convergence of blind source separation. The convergence behavior of the proposed algorithm is compared with the blind-only case.