Universal Vertical Applications Adaptation for Open RAN: A Deep Reinforcement Learning Approach

The most challenging issues for the fifth-generation (SG) radio access networks (RANs) are to support manifold verti-cal applications in any deployment environment and optimize the wireless service performance. The artificial intelligence (AI) and machine learning (ML) have been proposed in many studies as substitutions to traditional configuration frameworks. To execute AI/ML algorithms in a 5G RAN, the Open RAN (O-RAN) Alliance has created a new O-RAN architecture that includes an unprecedented computing platform called the RAN Intelligent Controller (RIC). With the establishment of the RIC platform, the RAN or third-party designers can design different AI/ML algorithms in the form of xAPPs to configure the transmission parameters and resources on the Uu interface to optimize the performance of different vertical applications. In order to obtain suitable transmission parameters and resources, this paper proposes an xAPP design based on the deep Q network (DQN) with a duel structure and double Q networks. The performance evaluation results demonstrate the effectiveness of the proposed design to maximize the throughput while satisfying the packet delay budget (PDB) and packet error rate (PER) requirements of the vertical application.