AI/ML for NR air interface: beam management, CSI feedback

AI/ML Framework for NR

3GPP Rel-18 introduces a study-item-turned-work-item framework for applying AI and machine learning to the NR air interface. The framework defines a standardized lifecycle for ML models: data collection at the UE or gNB, model training (which may occur offline or online), model distribution and update over the air interface, and model inference during live operation. Three initial use cases were prioritized: beam management to reduce beam sweep overhead, CSI feedback compression to lower uplink reporting cost for massive MIMO, and positioning enhancement to improve accuracy in challenging…