Task Offloading and Resource Allocation in an RIS-Assisted NOMA-Based Vehicular Edge Computing

doi:10.1109/ACCESS.2024.3454810

2024-01-01

A Yakubu *

Ahmed H. Abd El‐Malek

Mohammed Abo‐Zahhad

Osamu Muta

Maha Elsabrouty

DOI：10.1109/ACCESS.2024.3454810

原文链接

原文求助

摘要

En 中文

With the rise of intelligent transportation (ITS), autonomous cars, and on-the-road entertainment and computation, vehicular edge computing (VEC) has become a primary research topic in 6G and beyond communications. On the other hand, reconfigurable intelligent surfaces (RIS) are a major enabling technology that can help in the task offloading domain. This study introduces a novel VEC architecture that incorporates non-orthogonal multiple access (NOMA) and reconfigurable intelligent surfaces (RIS), where vehicles perform binary or partial computation offloading to edge nodes (eNs) for task execution. We construct a vehicle-to-infrastructure (V2I) transmission model by considering vehicular interference and formulating a joint task offloading and resource allocation (JTORA) problem with the goal of reducing total service latency and energy usage. Next, we decompose this problem into task offloading (TO) problem on the vehicle side and resource allocation (RA) problem on the eN side. Specifically, we describe offloading decisions and offloading ratios as a decentralized partially observable Markov decision process (Dec-POMDP). Subsequently, a multi-agent distributed distributional deep deterministic policy gradient (MAD4PG) is proposed to solve the TO problem, where every vehicular agent learns the global optimal policy and obtains individual decisions. Furthermore, a whale optimization algorithm (WOA) is used to optimize the phase shift coefficient of the RIS. Upon receiving offloading ratios and offloading decisions from vehicles, edge nodes utilize the Lagrange multiplier method (LMM) and Karush-Kuhn-Tucker (KKT) conditions to address the RA problem. Finally, we design a simulation model based on real-world vehicular movements. The numerical results demonstrate that, compared to previous algorithms, our proposed approach reduces the overall delay and energy consumption more effectively.

Keyword:

Reconfigurable intelligent surfaces

Resource management

Optimization

NOMA

Edge computing

Cloud computing

Energy consumption

Deep reinforcement learning

Reconfigurable intelligent surface

non-orthogonal multiple access

real-time task offloading