Polyhedral concepts for deterministic blind separation of binary sources

We study a distinct nonstatistical technique to perform blind source separation in the context of wireless digital communications. The proposed approach verifies the so-called finite-sample property: it achieves exact separation of sources, given a finite set of noiseless data samples. Thus, it is specially attractive for fast changing multipath scenarios, where the receiver has only a small number of data samples corresponding to the same channel profile. In this approach, the linear equalizers appears as the sole vertices of a data-dependent polyhedron. This permits one to develop simple techniques for blind equalization of a single-user and blind separation of co-channel sources. The proposed methods are mainly feasible in moderate to high signal-to-noise ratio (SNR) scenarios. We develop analytical tools to assess the impact of the noise on the proposed approach. Some preliminary results are presented, and the developed theoretical analysis is validated through computer simulations.