Sum-product decoding of layered space-time trellis coded signals over MIMO fading multipath channels

We consider an idealized synchronous single-carrier single-cell uplink communication where K users equipped with multiple antennas transmit their data to a base station also equipped with multiple antennas. Users have no (or very limited) channel state information. They employ space-time trellis codes to improve the reliability of their transmission link. Transmission occurs on fading multipath channels. Prior to transmission, users' spacetime codewords are interleaved in order to break channel memory. The multiuser joint decoding at the base station (assuming perfect channel state information) is reformulated as the computation of marginal probabilities on dense layered factor graphs with loops. This computation is done by means of the sum-product algorithm. We derive several instances of the sum-product algorithm by modifying the topology of the loopy graph and the definition of the corresponding exchanged messages and show through simulations that those variations can substantially impact the final delivered marginals.