RL based transmission power control for energy harvesting sensor

Sensors can harvest energy from the surrounding environment, but the energy supply is always unstable. To achieve effective power control of sensors and enhance their performance metrics, such as data throughput, while ensuring long-term life, a reinforcement learning-based power control strategy was designed. Assume an end-to-end communication system, the sender harvests energy, stores it in a battery for data transmission, and continuously buffers data. In practical scenarios, the arrival of energy and data is random and unpredictable. In this study, the current state was only observed via the sender, which included harvested energy, battery level, collected data, data cache level, and channel gain. Decisions were made solely based on these limited observations. The soft actor-critic (SAC) algorithm was used to control transmission power, with an appropriate reward function and action clipping method. Experimental results demonstrate that the proposed algorithm outperformes baseline strategies and approaches the theoretical optimal in certain scenarios.