Reliable and energy-efficient data collection from resource-limited sensors is essential to the success of future Internet of Things (IoT). In this article, we study the energy consumption minimization problem during the data collection from a generic wirelessly-powered sensor. Specifically, we determine the optimal data collection parameters, in terms of charging duration and charging power as well as sensor transmission rate, in real time according to the instantaneous channel condition while satisfying a certain latency constraint. For the scenario of ideal rate adaptive transmission with linear energy harvesting, we derive closed-form expressions for optimal transmission parameters. We also establish the condition on channel quality for successful data collection under a latency constraint. For the more practical case of finite block-length transmission with non-linear energy harvesting, we develop a deep reinforcement learning (DRL) solution for efficient online implementation. We also propose an online tuning scheme to cater for model inaccuracy and environment variation. The accuracy and effectiveness of our proposed approaches are verified by comparing with benchmark schemes. Our DRL-based approach has broad applicability and can solve other real-time optimal design problems in wireless communications.
Bibliographical noteKAUST Repository Item: Exported on 2022-09-16
Acknowledgements: This work is supported in part by a NSERC Discovery grant.