Deep Q-Learning for Intelligent Band Coordination in 5G Heterogeneous Network Supporting V2X Communication

Abstract

A heterogeneous or hybrid 5G network is required to support connected vehicles to implement the full range of cooperative ITS (intelligent transport system) services in diverse scenarios. In order to enhance data rate or reduce latency by increasing transmission bandwidth, 5G utilizes frequency bands below and above 6 GHz. The challenge is that multiple band coordination in 5G will be essential to mobile network operators. Even worse, traditional strategies could not meet the demand. Most current 5G research is focused in 5G network optimization. However, frequency coordination in 5G, as one of the most important requirements from operators, is left untouched. In this paper, a multi-agent deep Q-learning network (DQN) is developed as coordination solution. Transfer learning is introduced in DQN to decrease the deployment complexity of the proposed solution on 5G gNB (next-generation NodeB). By deploying the proposed solution in the system level simulation, the simulation shows an average 10% throughput enhancement, an about 24% accessed user number increasing, and around 70% training time saving, compared with normal Q-learning solution, and it enables the operators to optimally utilize all the valuable frequency resources to the best commercial value.