Learning Assisted Estimation for Time- Varying Channels

Channel estimation is a critical module to determine the performance of wireless receivers. For some communication systems, the channels are time-varying and without well-justified models, e.g., underwater acoustic channels, high mobility channels, and mm Wave channels. These channels are usually hard to use finite parameters to estimate and track. Channel estimation in these cases may significantly affect the symbol detection performance. In this paper, we develop learning assisted (LA) channel estimation algorithms. We use CNN and DNN based channel estimators to track channel variations. We demonstrate that the estimators can be dynamically updated using pilots through incremental learning. Different from the existing channel estimators, our algorithms combine learning techniques with preamble training symbols and pilots, and thus can track channel variations on-line and fit better for the current cellular systems, vehicular communications, and underwater acoustic systems. Simulation results validate the effectiveness of our algorithms.