Deep Reinforcement Learning for Optimizing Energy Efficiency in STAR-RIS Assisted Communication Networks

Synchronous Transmission and Reflection Reconfigurable Intelligent Surface (STAR-RIS) is emerging as a key enhancers in future communication networks due to its low cost, minimal energy consumption, programmability, and straightforward deployment. This surface notably improve wireless network performance and optimize the energy efficiency of communication systems. A critical feature of STAR-RIS is its ability to provide extensive signal coverage and intelligently manage energy consumption through adaptive on/off switching mechanisms, thereby minimizing network power use. This paper investigates a STAR-RIS assisted communication network aimed at maximizing system energy efficiency. Owing to the fractional nature of the objective function and the complexities introduced by the intelligent switching, traditional convex optimization techniques are inadequate. Consequently, this study introduces a novel approach using deep reinforcement learning, the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, and proposes an enhanced variant, TCD3. Simulation results indicate significant improvements in system energy efficiency with both algorithms, with TCD3 surpassing TD3 in terms of learning speed, average rewards, and reduced variance.