AI-Native Handover Management for 6G Networks Leveraging O-RAN Architecture

Abstract

The evolution of advanced wireless communication has necessitated the transition from legacy networks toward more adaptable and intelligent sixth-generation (6G) architectures. Conventional radio access remains constrained in flexibility, intelligence, and scalability, limiting its ability to accommodate diverse and highly dynamic demands such as aerial mobility operations, where Unmanned Aerial Vehicles (UAVs) must sustain reliable connectivity across integrated terrestrial and satellite access. Open Radio Access Networks (O-RAN) address these limitations through open interfaces and the Radio Access Network (RAN) Intelligent Controller (RIC), which together make programmable intelligence available within the RAN. This paper places the terrestrial-satellite UAV handover (HO) decision inside the near-real-time RIC as an xApp, using a deep reinforcement learning (DRL) framework that combines a Deep Q-Network (DQN) with Prioritized Experience Replay (PER) and a reward shaped to weigh the long-term cost of switching against the reliability it secures. Assessed against the standardised HO baseline in its conventionally adopted configuration, the resulting policy sustains command-and-control (C2) link reliability under non-stationary aerial conditions at a bounded switching rate, and an ablation of the reward establishes which of its terms are responsible, showing that reliability-oriented mobility management for aerial users can be realised as a deployable near-real-time RIC function.