Intelligent Computational Offloading Using Lyapunov Optimization and Deep Reinforcement Learning

In this work, we propose a novel intelligent computational offloading (ICO) algorithm that adopts a binary offloading policy such that each computational task of the wireless devices is either executed locally or fully offloaded to a MEC server. The ICO algorithm was tested for the maximum data rate and for the minimum power consumption, and it was found that it successfully optimizes the task offloading decisions and wireless resource allocations under the time-varying wireless channel conditions. It was demonstrated in the simulation results that ICO is able to achieve the offloading decisions four times faster than CNN and approximately two times faster than Lydroo. ICO was able to achieve long-term stability of the queue in both of these scenarios. In contrast, Lydroo failed to maintain it for the same number of users' wireless devices. In order to have a fair comparison, all three algorithms were tested for different learning rates by keeping a fixed number of wireless users, with the results showing that ICO was able to perform superiorly than CNN and Lydroo. These results show that the ICO can be used for making near real-time and optimal offloading decisions even in a fast-fading environment.