Power Allocation for the STAR-RIS-NOMA Short Packet Communication System Using Deep Reinforcement Learning

Optimizing power is crucial for modern intelligent communication systems. This paper explores power allocation in short-packet networks using Non-Orthogonal Multiple Access (NOMA) and Simultaneously Transmitting and Reflecting Reconfigurable Intelligent Surface (STAR-RIS) technology. STAR-RIS enhances coverage, saves energy, and improves signal transmission to users on both sides of the source. We propose using Deep Reinforcement Learning (DRL) with the Proximal Policy Optimization (PPO) algorithm to optimize transmission power and power allocation coefficients. PPO directly optimizes the agent's policy, using channel state information as input and producing transmission power values and power allocation coefficients as output. Numerical results show that DRL reduces computational complexity and requires minimal training data and time.