Optimal training for ML and LMMSE channel estimation in MIMO systems

In this paper, we present some optimal training designs for maximum likelihood (ML) and linear minimum mean square error (LMMSE) channel estimation for multiple-input multiple-output (MIMO) systems. As optimization criterion, the channel mean square error (MSE) is chosen. The key idea is not to restrict the channel estimation to a single transmitted symbol block, but to possibly exploit multiple symbol blocks, assuming the channel remains constant over these blocks. This leads to some new optimal training designs