Multi-user Multi-channel Computation Offloading and Resource Allocation for Mobile Edge Computing

Abstract

We study the problem of computation offloading and resourceallocation in multi-user multi-channel mobile edge computing (MEC) systems. Each user equipment (UE) in the system has a computation-intensive and time-sensitive task that needs to be executed either locally or remotely in an MEC server. All UE tasks have individual deadline constraints that are treated as soft constraints. The MEC server has limited computational resources, which impose hard constraints on the overall offloading computation capacity. The objective is to minimize a cost function that is expressed as a weighted sum of energy consumption, delay, and deadline penalty of all UEs. The optimum design is performed with respect to three decision parameters: whether to offload a given task, which wireless channel to use during offloading, and how much MEC resources should be allocated for an offloaded task. We apply a Deep Reinforcement Learning approach known as Deep Deterministic Policy Gradient to solve the problem. Simulation results demonstrate that the proposed algorithm outperforms other existing schemes such as Deep Q-Network (DQN).