R1-2410467
discussion
Specification support for AI-ML-based positioning accuracy enhancement
From Qualcomm
Summary
Qualcomm presents a comprehensive technical document for AI/ML-based positioning accuracy enhancement in 5G NR, containing 31 proposals and 34 observations spanning positioning integration, model training/inference consistency, data collection, model input/output aspects, and lifecycle management. The document advocates for treating AI/ML positioning as enhancements to existing positioning methods rather than entirely new approaches.
Position
Qualcomm strongly advocates FOR treating AI/ML positioning as enhancements to existing positioning methods (reusing mature procedures and IEs) and FOR explicit provision of assistance data rather than implicit indication approaches. They push AGAINST network-side training for Case1, AGAINST implicit indication (associated ID) for training/inference consistency, AGAINST sample-based measurements (Alt-A) in favor of path-based measurements (Alt-B), and AGAINST reporting phase information due to overhead without clear accuracy benefits. They champion UE-vendor-controlled model development and explicit assistance data provision.
Key proposals
- Proposal 1 (Sec 2): For AI/ML positioning cases (Case1 to Case3b), proceed with AI/ML positioning specifications as part of existing positioning methods - Case1 as part of UE-based DL-TdoA and DL-AoD positioning, Case3a and Case3b as part of NG-RAN-assisted UL-TdoA, NG-RAN-assisted UL-AoA, and multi-RTT positioning
- Proposal 2 (Sec 3.1): In AI/ML positioning Case1, for training and inference consistency, UE is responsible to ensure consistency and LMF provides UE with necessary information
- Proposal 4 (Sec 3.3): In AI/ML positioning Case1, for training and inference consistency, deprioritize the implicit indication (e.g., associated ID) when providing existing assistance data of UE-based DL-TdoA and DL-AoD
- Proposal 7 (Sec 4.3.1): In AI/ML positioning Case3a, for model output, support enhancing the meaning when reporting multiple LOS/NLOS indicators and their associated measurement results to indicate multiple hypothesis for the reported path/TRP
- Proposal 8 (Sec 4.3.2): In AI/ML positioning Case3a, for model output, support enhancements in reporting of earliest path LOS and timing info measurements in which gNB reports multiple-hypotheses soft info of LOS indicator and timing information
- Proposal 9 (Sec 5.1): In AI/ML positioning Case1, for model training and development, deprioritize NW-side training, and deprioritize model transfer Case z4
- Proposal 11 (Sec 5.3): In AI/ML positioning Case1, for data collection, support two labelling assistance options from LMF to UE: Option 1: Ground truth label generated by LMF and provided to UE; Option 2: Position calculation assistance data provided from LMF to UE, and UE calculates ground truth label
- Proposal 17 (Sec 5.4.3): In AI/ML positioning data collection Case1, for Part B label and quality indicator of direct AI/ML positioning, when generated by LMF, support location coordinate types and uncertainty of IE LocationCoordinateTypes that can be requested by device
- Proposal 21 (Sec 6.1): In AI/ML positioning Case3b, for model input measurements, support Alt-B path-based measurements as the default and baseline measurement reporting for LMF-side model
- Proposal 22 (Sec 6.3): In AI/ML positioning Case3b, for model input measurements, no support for reporting phase information (e.g., CIR) for model input running at LMF side
- Proposal 23 (Sec 7): In AI/ML positioning Case1, for inference related request, support at least the following information from LMF to UE: Signaling on whether location information needs to be obtained using AI/ML, legacy, or both AI/ML and legacy method
- Proposal 29 (Sec 8.1): In AI/ML positioning Case1, for monitoring, support assistance options for label-based monitoring metric calculation: Option A-1, Option A-2, Option B-1
- Proposal 3 (Sec 3.3): In AI/ML positioning Case1, for training and inference consistency, support both Case1 as part of UE-based DL-TdoA including all existing ADs and Case1 as part of UE-based DL-AoD including all existing ADs
- Proposal 12 (Sec 5.3): In AI/ML positioning Case1, for data collection, support at least information in existing assistance data of UE-based DL-TdoA and DL-AoD with explicit indication from LMF to UE for enabling model development and ensuring training and inference consistency
- Proposal 30 (Sec 8.1): In AI/ML positioning Case1, for monitoring, support UE-initiated monitoring with assistance from LMF to UE including RS resources configurations/activations and monitoring assistance information