Polyhedral methods for blind deterministic separation of binary co-channel signals

We present a novel approach for blind deterministic separation of a linear superposition of binary co-channel signals. The proposed approach does not rely on any input/output statistical model, and requires only a small number of data samples to perform channel equalization and/or source separation. Thus, it is specially attractive for fast changing multipath scenarios such as mobile wireless radio channels. We derive a framework which fully exploits the properties of the binary phase shift keying (BPSK) modulation format. Within this framework, linear equalizers emerge as the unique vertices of a special data dependent polyhedron. We discuss two techniques, which either capture a single vertex (single user contexts), or identifies them all (multiple user scenarios). Both techniques achieve exact equalization/source separation for noiseless finite data packets. Preliminary computer simulations illustrate the promising performance of the proposed approach.