Near optimal training sequences for low complexity symbol timing estimation in MIMO systems

Training sequences for data-aided timing estimation in multi-input multi-output systems are designed. It is observed that for low complexity implementation, the sequences must necessarily satisfy the zero cross-correlation zone property. By restricting our search to a more tractable subset of this class of sequences, we are able to minimize the modified Cramer-Rao bound in closed form and obtain sequences whose mean square error performance is close to that of the optimal orthogonal sequences. Two constant modulus sequences with even lower implementation complexity are also proposed.