Effectively inserted training for channel state estimation of spatially correlated MIMO-OFDM

This work considers channel state estimation of multiple-input multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) with spatial correlation by inserting a very short-length training sequence into each OFDM symbol. The rate occupancy of the training sequence within one OFDM symbol is just 1/128. The objective for designing training sequences is to minimize the entropy of the error between channel state and its estimator. Unlike the case with full length training sequences, where the training sequence design is formulated as a convex optimization problem with the available computational solution, the problem of short-length training sequence design is highly nonconvex and thus is very computationally challenging. Our main result is to develop a low-complexity iterative procedure for its solution. Simulations show the effectiveness of our methods.