R1-2500518
discussion
Discussion on specification support for AI-ML based positioning accuracy enhancement
From Baicells
Baicells's prior position on
9.1.2
at
RAN1#119
· AI-synthesized, paraphrased
verify sources →
Strongly advocates for sample-based measurements over path-based measurements, citing superior positioning accuracy and avoidance of algorithm inconsistencies between vendors. Pushes for phase information support in Case 3b despite overhead concerns. Advocates for minimizing specification impact by reusing existing IEs and procedures wherever possible rather than defining new mechanisms.
Summary
Baicells presents 12 proposals and 9 observations regarding the specification support for AI/ML-based positioning accuracy enhancement in NR Rel-19, focusing on model inputs, training data collection, and quality indicators. The document argues for sample-based measurements over path-based ones for superior accuracy and proposes reusing existing legacy signaling mechanisms to minimize normative workload.
Position
Baicells supports sample-based measurements as model input, particularly for Case 3b, arguing they provide superior positioning accuracy compared to path-based measurements across various TRP numbers and bandwidths. They propose specific parameter ranges for Nt’ and k and support reusing legacy reference times and IEs, such as 'LoS/NLoS information' and 'Timing Measurement Quality', to minimize specification impact. For training data collection, Baicells suggests exploring gNB-side model training for Case 3a and redesigning procedures for Case 3b to allow LMF to indicate desired SRS configurations. They argue that new measurement requests and responses are beneficial for Case 3b to avoid bandwidth waste and ensure data alignment. Regarding monitoring, they support options that leverage existing mechanisms for Case 1 to achieve minimum specification effort.
Key proposals
- Proposal 1 (Model Input): Defines value ranges for Nt’ as {16, 32, 64, 128} and k as 0…5 for sample-based measurements.
- Proposal 2 (Model Input): Supports multiple options for the starting time of UE-side sample-based measurements, specifically Option B and Option D.
- Proposal 3 (Model Input): Supports reusing legacy reference time for UE in Case 2b, while leaving Case 1 implementation up to the device.
- Proposal 4 (Model Input): Supports phase information reporting for Case 3b to improve positioning accuracy.
- Proposal 5 (Model Output): Proposes reusing the existing 'LoS/NLoS information' IE in TS 38.455 for Case 3a, using special values (0 or 1) to indicate AI/ML output.
- Proposal 6 (Model Output): Supports the transfer of UE-side model information to the LMF via LPP for Case 1.
- Proposal 7 (Training Data Collection): Proposes exploring the feasibility of training models at the gNB side as the primary focus for Case 3a normative work.
- Proposal 8 (Quality Indicator): Proposes associating Timing Measurement Quality with timing information of Part A in NRPPa to reduce normative workload.
- Proposal 9 (Quality Indicator): Supports reusing the existing IE 'Timing Measurement Quality' for Case 3a and 3b inference to align with traditional methods.
- Proposal 10 (Quality Indicator): States there is no strong need for a power quality indicator as it is only for PDP and has not been well evaluated.
- Proposal 11 (Quality Indicator): Supports Option 1 (reusing an existing IE) for the quality indicator of Case 1 labels.
- Proposal 12 (Model Monitoring): Supports Options A-1, A-2, and B-1 for Case 1 monitoring to achieve minimum specification effort.