Superimposed training-based noncoherent MIMO systems

The training-based scheme constitutes an efficient method for noncoherent multiantenna communications. In this paper, we propose a superimposed training-based linear dispersion (STLD) transmission scheme, where the transmitted symbols are superposition of the training pilots and the data symbols. The STLD system enjoys simple encoding and efficient suboptimal decoding. Since for noncoherent multiple-input multiple-output systems employing the STLD scheme, analytical expressions of block- or bit-error probabilities do not exist, deterministic optimization methods cannot be used to design the optimal STLD system. We therefore propose to design the optimal STLD systems using simulation-based optimization techniques, together with gradient estimation. Simulation results show that with either maximum-likelihood or suboptimal decoding, the STLD systems obtained by the proposed design algorithm generally outperform the conventional training-based scheme employing space-time codes designed for coherent channels.