The BLAST architecture has been proposed for high-capacity and spectrally-efficient wireless communications in an indoor environment. The method relies on multi-transmit and receive antennas to send and receive information-bearing signals in parallel. The architecture assumes a rich independent-ray scattering mechanism to make the parallel information separable at the receiving ends. In practice, with the increased number of parallel sub-streams, the scattering may be less favorable so that signal decoding algorithms are needed. In this paper, we propose a statistical learning demodulating scheme for this task.