Robust Learning-Based Data-Aided Channel Estimator in Time-Varying MIMO Channels

This letter introduces a robust learning-based data-aided channel estimator in time-varying channels. The core idea is to integrate data-aided and model-based channel estimates where the data-aided estimate improves the estimation accuracy and the model-based estimate captures the temporal channel variations. To optimally merge these two estimates, a Markov decision process (MDP) is formulated to minimize the mean-squared-error of the combined channel estimate. A reinforcement learning algorithm is then developed to efficiently solve the MDP. Simulations demonstrate that the proposed estimator significantly outperforms conventional estimators in both slow and fast fading channels.