Deep Reinforcement Learning Based Energy-efficient Task Offloading for Secondary Mobile Edge Systems

Abstract

In order to support last-mile wireless connectivity of computation-intensive applications, edge systems can benefit from secondary (i.e., opportunistic) utilization of licensed spectrum. However, spectrum sensing for such secondary utilization can end up causing considerable energy consumption for already energy-constrained mobile devices. In this paper, we propose an energy-aware task offloading strategy for secondary edge systems that aims to find trade-offs between channel sensing and task offloading for mobile device energy optimization. The proposed strategy employs a Deep Reinforcement Learning based approach that rewards secondary mobile devices for taking part in cooperative spectrum sensing by allowing them to offload their compute-intensive tasks to edge servers in order to conserve energy. Using simulations, we demonstrate how effectively the proposed strategy can capture dynamic channel states and enforce intelligent offloading decisions. Results show our strategy’s benefits over optimization-based approaches and demonstrate its practicality for real-world use-cases where devices are controlled by different stakeholders.