강화학습 기반 Autonomous Framed Slotted-ALOHA 프로토콜

초록

We propose Adaptive Framed Slotted ALOHA (AFSA), an enhanced protocol designed to maximize the throughput of the Slotted ALOHA protocol by leveraging Q-ALOHA reinforcement learning. AFSA operates independently for all participating nodes, ensuring that no additional traffic is generated for the collection of training data required for Q-ALOHA reinforcement learning. each independently operating node incorporates STATIC, Exponential Moving Average (EMA), and Exponential Reward Scaling (ERS) into the reinforcement learning reward update function, evaluating both speed and stability. These mechanisms aim to evaluate and improve both speed and stability while minimizing looping situations caused by local minima problems. The performance evaluation reveals that applying the ERS function achieved a throughput of 84% after 200 simulation episodes—50% higher than the theoretical maximum throughput of 37% for traditional Slotted-ALOHA. Furthermore, applying EMA and ERS demonstrated more stable convergence compared to STATIC. The study confirms that AFSA can independently achieve near-optimal throughput without consuming additional bandwidth for training data collection.

키워드