Iterative Mean Removal Superimposed Training for Frequency Selective Channel Estimation

The topic of channel estimation for single-input single-output frequency-selective time-invariant channels has been recently addressed using superimposed training (ST) techniques. Improvement of these approaches using iterative algorithms has shown to achieve the best performance, although with the drawback of increasing computation complexity. In this work we propose a new iterative ST algorithm that resemblances data dependent ST (DDST), but extracts the cycling mean of the data, at the receiver´s end. This mean removal ST (MRST) approach accomplishes similar performance in terms of channel estimation error and noise immunity, but it has less complexity when compared with the best iterative techniques found in literature. Simulation results of several iterative approaches and MRST in connection with both MLSE and MMSE equalizers are presented to corroborate the theoretical analysis.