Semi-blind separation and detection of co-channel signals

Abstract
We propose a semi-blind algorithm for smart antennas that, in addition to some structural signal properties, utilises user identifiers that are available in existing cellular systems. Our algorithm estimates, first, the row span of the adequately composed data matrix and then projects the obtained basis vectors iteratively to the finite alphabet (FA) constellation. We call the projection part of our algorithm DILSF (decoupled iterative least-squares with subspace fitting). Assuming two co-channel users in a GSM-like TDMA system and a Rayleigh fading channel with finite angular spread, we obtained a BER<10/sup -3/ with only 3 antennas, spaced 10/spl lambda/ apart. We also demonstrate that co-channel signals are resolvable even if they are not separable in angle. We gain a factor of up to 70 in computational complexity by employing subspace tracking algorithms and demonstrate the robustness of the method against array imperfections.