On Exploration of SDR-based Wireless Power Transmission: Reinforcement Learning Perspective

Abstract

Radio frequency (RF) energy harvesting (EH) has emerged as a promising solution to power wireless devices, thereby facilitating a sustainable and continuous energy supply. This research leverages machine learning (ML), reinforcement learning (RL), and causal reinforcement learning (CRL) models to optimize EH in a practical Wireless Power Transfer (WPT) setup. The Orthogonal Frequency Division Multiplexing (OFDM) data generated under practical constraints using a hardware setup consisting of a Universal Software Radio Peripheral (USRP) device as a transmitter and an energy harvester module serves as the dataset for model training, thus translating the theoretical simulations into practical applications. Among the ML models evaluated, the Random Forest model demonstrated superior performance, with a MSE of 1.67% and a R2 score of 0.49. We implemented the Twin Delayed Deep Reinforcement Policy Gradient (TD3) algorithm for RL. For CRL, we combined TD3 with Instance-wise Feature Selection (INVASE) for causal feature selection. The CRL model outperformed the RL model in prediction capabilities, highlighting its potential to optimize EH. Our results validate the effectiveness of incorporating causal reasoning with RL for real-time decision making in dynamic environments. This study contributes to the practical application of intelligent EH systems, providing a solid foundation for future research and practical deployment. © 2025 IEEE.