Multi-Agent DRL for task offloading with delay reliability assurances in ris-aided MEC in UDN

Abstract

Abstract This paper focuses on the scenario where Reconfigurable Intelligent Surfaces (RISs) are introduced in ultra-dense networks (UDN) to guarantee the transmission performance of task offloading of users in Mobile Edge Computing (MEC). Considering users’ task arrival and wireless channel stochasticity, we analyze the delay violation probability bound for each user’s two-hop tandem queueing system with the Martingale theory, i.e., the task offloading and edge computing queues. By taking the delay violation probability bounds as the service reliability constraint, we further propose a MADRL-based algorithm, which co-trains each user and BS as independent and heterogeneous agents and maximizes the total energy efficiency of the system by jointly optimizing the users’ offloading ratio, the RISs’ phase-shift, and the BSs’ resource allocation policy. Experimental results show that our algorithm can improve the EE by about 17.9% compared to other alternative methods under different reliability requirements.