Joint Offloading Decision and Spectrum Assignment in D2D-Enhanced MEC Network: A Deep Reinforcement Learning Approach

Abstract

Due to the adaptability to dynamic environments and the capability of optimizing complex decision-making processes, deep reinforcement learning (DRL) has been recently applied to scheduling and resource allocation problems in wireless communications. In this paper, we propose a DRL-based joint task offloading and spectrum assignment approach in a device-to-device (D2D) enhanced mobile edge computing (MEC) scenario, where multiple resource devices (RDs) and an edge server collaboratively provide auxiliary task offloading services to task devices (TDs). Considering the time-varying wireless channel conditions, we formulate an optimization problem to jointly design the offloading decision and sequential spectrum assignment across multiple coherent time slots. The optimization objective is to minimize offloading latency while ensuring the offloading success rate for all TDs. To address this problem, we propose a two-step approach, comprising a game-based offloading matching (GBOM) to obtain the Nash equilibrium (NE) of offloading decisions, and a multi-agent DRL (MADRL) to solve the sequential spectrum assignment problem. Numerical results show that the proposed GBOM-MADRL algorithm mitigates the system offloading latency and improves the average probability of successful offloading.