Parametric Model and Estimator Classifier for Optimal Averaging in Mobile OFDM Systems with Superimposed Training

Abstract

Superimposed training (ST) is an attractive technique for channel estimation in orthogonal frequency division multiplexing (OFDM) modulation. However, its main challenge is the intrinsic interference due to the joint transmission of pilot and data symbols, which can be mitigated by averaging the received signal. Previous works analyzed the mean square error (MSE) of the channel estimation, for both least squares (LS) and minimum MSE (MMSE) estimators, and showed that, under realistic channel models, the optimum number of averaged symbols could be computed by solving a transcendental equation. In this paper, as a practical implementation proposal, these optimum averaging values are parametrically approximated with a multilinear regression model. Also, it is proposed an accurate classifier that, under delay and performance tolerances, is able to select the most suitable estimator between LS and MMSE.