Online Meta-Learning For Hybrid Model-Based Deep Receivers

Recent years have witnessed growing interest in the application of deep neural networks (DNNs) for receiver design, which can potentially be applied complex environments without relying on knowledge channel model. However, dynamic nature communication channels often leads to rapid distribution shifts, may require periodically retraining. This paper formulates a data-efficient two-stage training method that facilitates online adaptation. Our mechanism uses predictive meta-learning scheme train rapidly from data corresponding both current and past realizations. is applicable any network (DNN)-based receiver, does not transmission new pilot training. To illustrate proposed approach, we study DNN-aided receivers utilize an interpretable model-based architecture, introduce modular strategy based meta-learning. We demonstrate our techniques simulations synthetic linear channel, non-linear COST 2100 channel. results allows outperform previous self-supervision joint-learning by margin up 2.5 dB coded bit error rate rapidly-varying scenarios.