Blind separation of multiple sequences from a single linear mixture using finite alphabet

In this paper, we propose an approach exploiting the relationship between the position of the clustering centers of the observed data and the mixing parameters to realize blind separation of multiple sequences, drawn from finite alphabet set, from a single linear mixture. In theory, we prove that the system estimation by the algorithm is optimum in the sense of least square (LS) by mathematical induction method. In the noise case, the simulation results show that the system estimation deteriorates smoothly with the increasing of noise variance.