Optimal Estimation of Frequency-Selective Channels in OFDM-based Superimposed Training Schemes

Orthogonal frequency division multiplexing (OFDM) is the main transmission scheme for the fifth generation (5G) of wireless communications. To properly operate, an estimation of the channel coefficients is mandatory, for this reason, superimposed training (ST), a technique that transmits pilot and data symbols together, have become an attractive solution. In ST-based channel estimation, an averaging of the received signal is required to mitigate the data interference over the pilots. Unlike reported works that employed oversimplified block-selective channel models to determine the averaging length of ST, in this paper, a frequencyselective model is considered in the analysis. Thus, realistic and more accurate mean square error (MSE) expressions for the ST-based least squares (LS) and minimum MSE (MMSE) channel estimators are theoretically derived. Then, the optimum number of averaged subcarriers, which provides the lowest MSE, is analytically computed. Simulated results considering parameters from 5G New Radio (NR) standard show that the MSE and symbol error rate (SER) metrics can be improved by more than one order of magnitude in contrast to previous works, which either did not average any subcarrier or averaged a group of subcarriers given by the coherence bandwidth simplification. Finally, the robustness of the proposed method, which guarantees a high tolerance in performance for a wide range of imperfect input parameters, is addressed.