D2D-assisted cooperative computation offloading and resource allocation in wireless-powered mobile edge computing networks

Abstract

With the increasing popularity of the internet of things (IoT) and 5th generation mobile communication technology (5G), mobile edge computing (MEC) has emerged as an innovative approach to support smart devices (SDs) in performing computational tasks. Nevertheless, the process of offloading can be energy-intensive. Traditional battery-powered SDs often encounter the challenge of battery depletion when offloading tasks. However, with the advancements in wireless power transfer technology, SDs can now achieve a sustainable power supply by harvesting ambient radio frequency energy. This paper studies the computation offloading in wireless-powered MEC networks with device-to-device (D2D) assistance. The SDs are categorized into near and far SDs based on their proximity to the MEC server. With the support of near SDs, far SDs can reduce transmission energy consumption and overall latency. In this paper, we comprehensively consider the allocation of energy harvesting time, transmission power, computation resources, and offloading decisions for SDs, establishing a mathematical model aimed at minimizing long-term average delay under energy constraints. To address the time-varying stochastic nature resulting from dynamic task arrivals and varying battery levels, we transform the long-term problem into a deterministic one for each time slot by introducing a queue and leveraging Lyapunov optimization theory. We then solve the transformed problem using deep reinforcement learning. Simulation results demonstrate that the proposed algorithm performs effectively in reducing delay and enhancing task completion rates.