Service Delay Optimization in Mobile Edge Computing-enabled Wireless Networks

Abstract

As vehicular networks increasingly rely on real-time applications such as autonomous driving and intelligent transportation systems (ITS), Mobile Edge Computing (MEC) has emerged as a crucial technology for reducing latency by offloading computational tasks to nearby servers. However, optimizing service delays in MEC-enabled vehicular networks remains a significant challenge due to the limited computational capacity of vehicles and the dynamic nature of the environment. Existing research has primarily focused on task offloading strategies and single-core RSU resource allocation, but these approaches often fail to adequately address propagation and computational delays without requiring extensive RSU cooperation. To address these issues, studies introduced a dual-stage deep reinforcement learning (DRL)-based framework using deep deterministic policy gradient (DDPG) to optimize transmit power for minimizing the overall service delay and deep Q-network (DQN) for core allocation. In this paper, we extend the dualstage DRL approach by incorporating advanced DRL algorithms—prioritized experience replay DDPG (PER-DDPG), combined experience replay DDPG (CER-DDPG), twin delayed DDPG (TD3), and proximal policy optimization (PPO)—to enhance power optimization and reduce propagation delays. Additionally, we introduce a comparative analysis of core allocation algorithms, including dueling double DQN (DDDQN) and double DQN (DDQN), to evaluate computational delay with varying core counts. Our findings show that Doelling DDQN offers the most efficient core allocation, leading to lower computational delays. This research fills a critical gap in the literature, showcasing how advanced DRL techniques can significantly improve resource-constrained vehicular networks and guide the optimization of MEC systems for more responsive and scalable ITS solutions.