Optimized Superimposed Training for MMSE Aware of Realistic Time-Variant Channel Models in OFDM

In this paper, a superimposed training (ST) scheme is optimally designed to estimate the coefficients of realistic time-variant channel models for orthogonal frequency division multiplexing (OFDM). The proposed analysis derives a closed-form solution for the averaging of received symbols that gives the lowest mean squared error (MSE) for minimum MSE (MMSE) estimator. In particular, this derivation is performed over an upper-bound of the MSE. Numerical realistic simulations of current New Radio (NR) standards prove that the proposed upper-bound is tight and the optimization can be effectively implemented. In addition, MMSE is compared with least squares (LS) estimator under different scenarios, and the performance is consistently enhanced particularly in low signal-to-noise ratio (SNR) conditions. Finally, the proposed averaging is compared with the state of the art (SOA) simulations which employ unre-alistic block-fading models, and results show that the optimum averaging can improve MSE up to one order of magnitude and reduce delay by 90%.