Improving the performance of the semi-blind cross-relation-based channel estimation method

As verified experimentally in literature, the performance of the semi-blind cross-relation-based channel estimation may suffer from significant degradation when the involved channels are ill-conditioned. In this paper, we investigate this degradation and we introduce simple conditions for the training part in order to mitigate the problem. The suggested training design is based on the minimization of the mean square error of the estimation in a high SNR regime. The attained performance is studied through analytical arguments and verified via extensive simulations.