R1-2409781
discussion
Specification support for AI-enabled positioning
From NVIDIA
Summary
NVIDIA presents a comprehensive framework for AI/ML-enabled positioning in 5G-Advanced, focusing on specification support for measurements, model lifecycle management, and data consistency. The document contains 1 observation and 9 proposals covering channel measurement reporting, training data generation, and model monitoring procedures.
Position
NVIDIA proposes supporting both sample-based and path-based time domain channel measurements, alongside the inclusion of phase information, to enhance AI/ML model inputs for positioning. They require that assistance data for UE-based positioning (Case 1) ensures consistency between training and inference environments. NVIDIA proposes studying quality indicators for both channel measurements and ground truth labels to address noise in training data. They advocate for comprehensive specification support for the full AI/ML model lifecycle, including configuration, activation, monitoring, and update procedures. Furthermore, they propose defining UE capabilities for AI/ML tasks and specifying conditions for Feature/FG availability and model identification to ensure robust network control.
Key proposals
- Proposal 1 (Sec 3): Support reporting phase information in time domain channel measurements for AI/ML based positioning.
- Proposal 2 (Sec 3): Support both sample-based measurements (integer multiple of sampling periods) and path-based measurements (detected path timing) for time domain channel measurements.
- Proposal 3 (Sec 3): Ensure assistance data provided from LMF to UE for Case 1 maintains consistency between training and inference phases.
- Proposal 4 (Sec 3): Study the definition of quality indicators for both channel measurements and ground truth labels to introduce specification support.
- Proposal 5 (Sec 3): Introduce specification support for assistance signalling and procedures for model configuration, activation/deactivation, recovery/termination, and selection.
- Proposal 6 (Sec 3): Introduce specification support for assistance signalling and procedures for model performance monitoring and update/tuning, including metrics, triggers, and feedback reports.
- Proposal 7 (Sec 3): Introduce specification support for UE capability signalling covering model training, inference, and monitoring capabilities.
- Proposal 8 (Sec 3): Introduce specification support for defining conditions under which an AI/ML Feature/FG is available for functionality.
- Proposal 9 (Sec 3): Introduce specification support for including additional conditions (e.g., scenarios, sites, datasets) in model description information during model identification.
- Observation 1 (Sec 2): Digital twins can be utilized to generate physics-based synthetic data for AI/ML-enabled positioning training.