Block methods for channel identification and source separation

On-line adaptive algorithms are not effective enough in tracking fast varying channels in blind and semi-blind modes. Block methods, long disregarded because of their numerical complexity, now hold appeal. In fact, the performance limitation now comes either from channel non-stationarity or from the absence of long training sequences, or both, but essentially no longer from the computational load. In addition, block methods are well adapted to burst-mode transmissions (TDMA systems). Some block algorithms are presented in this paper, aiming at identifying communication channels blindly. These algorithms provide analytical solutions (i.e. non-iterative) in order to be free of convergence problems. They exploit either the knowledge of the input alphabet, or the constant modulus property. The tools involved in the blind source separation (flat fading channels) and channel identification (frequency selective channels) problems are reviewed. The over-determined case (number of users exceeding the available diversity) is of particular interest. This assessment is not a survey of existing techniques, but a partial review of some original algorithms, often leading to closed-form solutions, and applicable in the presence of limited diversity