Meta-transfer Learning for Massive MIMO Channel Estimation for Millimeter-Wave Outdoor Vehicular Environments

In vehicular communications environments, channels are characterized as dynamic and highly mobile. As uch, estimating the vehicular communication channel with a massive number of antennas installed at the transmitter and receiver is considered a daunting task for conventional estimators and deep-learning approaches. Classical estimators provide inaccurate estimation results, and the deep learning algorithms require a huge dataset for training the model. This paper proposes a transfer learning and meta-learning approach for channel estimation in outdoor vehicular environments with millimeter-wave transmission frequencies above 6 GHz. The proposed system learns a good initialization of the model weight parameters using a few samples and a small number of gradient steps to achieve model convergence. Simulation results show that the proposed algorithm outperforms the conventional least square estimator in the outdoor millimeter-wave vehicular environments.