Computation Offloading in E-RAN via Deep Reinforcement Learning

Abstract

The current standards for fifth generation mobile communications pose significant performance challenges to current radio access networks. Elastic Radio Access Network (E-RAN) is currently considered an effective solution for future 5G access networks. Edge computing sinks computing and storage to edge nodes closer to users to meet the massive computing needs of large terminal devices. The integration of edge computing and wireless access network is an important trend in the field of Internet and communication. Therefore, task offloading and resource allocation in the E-RAN network architecture have become key issues that need to be addressed. We formalize the computational offload problem as a long-term optimization problem, with the goal of minimizing the total cost of system latency and energy consumption, and using an improved DQN network to determine offload decisions. Experimental results show that our proposed method can achieve better efficiency compared to other baseline methods.