Robust Clipped Affine Policy for Online Power Control in Energy Harvesting Communications

This paper studies online power control for energy harvesting communications over fading channels. A fundamental power control policy, coined robust clipped affine policy, is derived by a linear-policy-based approximation to the relative-value function in the Bellman equation. By leveraging the relative-value approximation and derived policy, a domain-knowledge-enhanced reinforcement learning (RL) algorithm is proposed for online power control with or without one-step energy lookahead. Simulation results show that the proposed methods outperform all existing approaches, with less than 2% performance loss relative to their corresponding optimal policies.