BLIND SEPARATION OF N BINARY SOURCES FROM ONE OBSERVATION: A DETERMINISTIC APPROACH

Konstantinos I. Diamantaras, Efthymia Chassioti
kdiamant@it.teithe.gr

We show that it is possible to separate n ? 1 binary source signals from a single linear mixture, under very mild as­ sumptions, based on the clustering of the data. We develop the mathematical treatment of the problem and propose a recursive, finite algorithm for its solution. The applica­ tion of this method is only limited by the level of noise and by the combinatorial explosion as the number of sources n increases since the algorithm complexity is exponential w.r.t. n. Simulation results indicate that the method can successfully separate as many as 10 sources from a single mixture in a few seconds on a typical desktop PC.