Delay- and Energy-Efficient Task Offloading in Cell Free Massive MIMO-Enabled Vehicular Fog Computing

Abstract

Task offloading is a promising approach to efficiently realize delay-sensitive, computation-intensive applications in Internet of Vehicles (IoVs). However, task allocation and scheduling pose great challenges in Vehicular Fog Computing (VFC) environment due to resource heterogeneity, workload unpredictability, fixed Fog Access Points (F-APs), and the dynamic nature of fog environment. This paper investigates the delay- and energy-efficient task offloading strategy in Cell Free massive MIMO (CF-mMIMO)-enabled VFC network. CF-mMIMO system is integrated into the VFC network so that task transfer among F-APs is enabled. A Long Short Term Memory (LSTM)-based algorithm is designed to predict the workload of F-APs. Based on the result, the delay and energy consumption of a task if it is offloaded on a F-AP can be calculated. After that, a Multi-Agent Deep Deterministic Policy Gradient (MADDPG)-based algorithm is developed to explore the best combination of task offloading and resource allocation strategies to reduce the overhead of each vehicle, and to minimize the long-term system cost, eventually. Simulation results show that the proposed strategy not only exhibits good convergence performance in scenario which involves a mixture of continuous-discrete action spaces, but also achieves satisfying performance in terms of average cost under varied circumstances.