Mutual information-based optimal training design for spatially correlated MIMO-OFDM

This paper deals with the optimal design of training sequences to estimate multiple-input multiple-output orthogonal frequency division multiplexing channels in the presence of spatial fading correlations. The optimal design criterion is the maximization of the mutual information between the unknown channel state information and the received signal in the training period. In contrast to previous methods based on the least square or minimum mean square error (MMSE) criterion in which the optimal training sequences can only be found for some extreme cases, here the optimal solution can be exactly obtained by convex programming for a general case. Furthermore, we propose tight bounds on the mutual information and solve the corresponding approximate design problems by the water-filling algorithm. Simulation results are provided to illustrate the superior performance of our methods.