FLeXHO: A Federated Learning Based Inter-Cell Handover Optimisation for Open 6G RAN

High-speed mobility poses significant challenges to meeting Quality of Service (QoS) requirements of next-generation networks, as traditional handover mechanisms based on static thresholds often underperform in dynamic environments, resulting in higher handover failures, ping-pong effects, and persistent cell load imbalances. To this end, we propose a mobility-aware federated learning (FL)-based handover optimisation framework for open radio access network (O-RAN) compliant networks. The proposed architecture includes three xApps within the near-RT RAN Intelligent Controller (near-RT RIC): a metrics collector extended application (xApp) that obtains real-time radio and mobility measurements, an FL-based load predictor xApp that forecasts load imbalance using distributed LSTM models, and a handover handler xApp that performs proactive AI-driven mobility management. Implemented on an OpenAirInterface (OAI) based end-to-end testbed with 3GPP protocol stack support, the proposed system achieves notable performance gains: a 94.5% handover success rate, a 95% reduction in load imbalance, and a 73% decrease in packet loss. By maintaining a near-RT RIC decision pipeline latency on the order of 10 ms, this work demonstrates the practical feasibility of federated, xApp-driven mobility optimisation in O-RAN.