R1-2409480 discussion

Discussion on AI/ML-based positioning enhancement

From ZTE
Status: not treated
WI: NR_AIML_air
Agenda: 9.1.2
Release: Rel-19
Source: 3gpp.org ↗
ZTE's prior position on 9.1.2 at RAN1#118bis · AI-synthesized, paraphrased
verify sources →
Strongly supports maximizing reuse of existing 3GPP procedures and advocates for CIR with phase information over PDP for better positioning accuracy despite higher overhead, while also favoring sample-based measurements over path-based approaches.

Summary

ZTE presents a comprehensive contribution on AI/ML-based positioning enhancements for Rel-19, containing 30 proposals and 11 observations across model input, output, training, and monitoring. The document strongly favors sample-based measurements over path-based ones due to reduced implementation ambiguity and argues for the inclusion of phase information (CIR) in model inputs despite higher signaling overhead.

Position

ZTE proposes supporting sample-based measurements for Rel-19 AI/ML positioning, arguing that implementation ambiguities in path-based measurements cannot be removed, whereas sample-based ambiguities can be resolved via LMF configuration. They require the starting point of Nt samples to be configurable by the LMF (Option C/D) to ensure consistent positioning performance across different TRP implementations. ZTE supports using Channel Impulse Response (CIR) including phase information for model input, presenting technical evidence that CIR provides significantly better positioning accuracy than Power Delay Profile (PDP) with acceptable overhead increases. They oppose introducing a specific indicator to identify AI/ML-derived measurements, arguing that LMF awareness of the procedure and timestamp suffices. For model monitoring, ZTE proposes that label-free monitoring is implementation-specific and requires no specification discussion, while supporting LMF-centric metric calculation for Case 1. They also argue against introducing an associated ID for AI/ML positioning, stating that UE-side generalization can be handled via mixed dataset training.

Key proposals

Your notes

Private to your account