Joint Scheduling and Beamforming Design in Traffic-Aware RIS Aided MEC Network

This paper considers exploiting reconfigurable intelligent surface (RIS) in a multi-user multi-access mobile computing (MEC) system. To precisely model the traffic behaviors of the tasks, we consider queuing procedures at both mobile users and the cloud server. This consideration leads to concatenation of queues and makes our problem much more difficult than existing literature since analytic expression of the objective is impossible. Our goal is to jointly optimize the task scheduling, RIS configuration and computation resource allocation to improve the MEC network’s overall computation rate in the long run while maintaining the stability of the queues. To tackle this highly challenging stochastic problem, we develop a Lyapunov guided deep reinforcement learning (DRL) solution, which borrows the Lyapunov optimization theory to stabilize queues and at the same time can effectively optimize the intractable objective in an online manner. Numerical results will be presented to verify the effectiveness of our solution and the performance boost due to the deployment of RIS.