Reducing Latency in NOMA-aided MEC Networks: A Deep Reinforcement Learning Approach

This paper delves into the latency reduction in mobile edge computing (MEC) networks augmented by non-orthogonal multiple access (NOMA) under flexible timing conditions. We introduce an iterative mechanism for task offloading between two users in a NOMA setup, prioritizing the optimization of phase shifts in a Reconfigurable Intelligent Surface (RIS) to curtail uplink latency. Drawing inspiration from deep reinforcement learning's (DRL) adeptness in handling complex control challenges, we innovatively crafted a deep deterministic policy gradient (DDPG)-based time-delay minimization (DDPR) algorithm, to adeptly navigate the non-convex optimization inherent in latency minimization. Numerical results suggest that our proposed system markedly reduces latency in a brief period, outperforming current methods.