R1-2409742
discussion
Specification support for positioning accuracy enhancement
From Intel
Summary
Intel presents 26 proposals and 2 observations regarding specification support for AI/ML-based positioning accuracy enhancements in Rel-19, focusing on data collection, model input/output characterization, and consistency between training and inference. The document argues that sample-based measurements are a specific implementation of path-based measurements and proposes reusing existing Rel-18 frameworks for time windows and area configurations to manage data collection. It further details mechanisms for ensuring consistency between training and inference via explicit assistance data or Associated IDs, and defines specific options for LOS/NLOS indicator interpretation and model monitoring.
Position
Intel argues that the sample-based measurement approach is a specific implementation of the path-based approach already supported in existing specifications, proposing to support path-based measurement with potential enhancements to the number of reported paths. They propose reusing Rel-18 frameworks for simultaneous DL/UL positioning measurements to configure time windows and validity areas for data collection, thereby managing staleness and relevance. For model input timing, Intel prefers defining reference time relative to a reference TRP or the start of the DL subframe from the UE Rx perspective, with propagation delay conveyed separately. Regarding model output, they propose clarifying the LOS/NLOS indicator interpretation, offering alternatives where the indicator either corresponds to the reported timing metric or the actual physical propagation path. To ensure consistency between training and inference, Intel proposes that measurement validity areas apply to both phases and that sensitive network information (e.g., TRP coordinates, beam info) can be optionally provided explicitly or implicitly via Associated IDs.
Key proposals
- Proposal 1 (Data Collection): Consider the entities for ground-truth label generation as presented in Table 2, distinguishing between UE, PRU, and LMF responsibilities.
- Proposal 3 (Data Collection): The LMF could provide a UE with configurations of measurement time windows for data collection, reusing the Rel-18 framework for simultaneous DL positioning measurements.
- Proposal 5 (Data Collection): The LMF could provide a UE with a configuration of a measurement validity area (reusing AreaID-CellList) within which to perform measurements for model training.
- Proposal 7 (Data Collection): RAN1 to investigate potential solutions to enable efficient compression of data collection, particularly for Case 2b where over-the-air transmission is involved.
- Proposal 9 (Model Input): Support path-based measurement for representation of time domain channel measurements, noting that sample-based approach is a specific implementation of path-based.
- Proposal 10 (Model Input): For timing information representation, consider Option DLRefTime-1 (reference TRP) or Option DLRefTime-2 (start of DL subframe from UE Rx perspective), with separate conveyance of propagation delay.
- Proposal 13 (Model Input): If phase information is supported, extend existing DL/UL RSCP definitions to include phase information for additional detected paths or samples.
- Proposal 14 (Model Output): Support reporting of RSRP/RSRPP for first and additional paths, along with timing information and optional LOS/NLOS indication, for AI/ML-assisted positioning Cases 2a and 3a.
- Proposal 15 (Model Output): For Case 2a, define LOS/NLOS indicator alternatives: Alt A (corresponds to timing metric in report) or Alt B (corresponds to actual radio path, may not match timing metric).
- Proposal 17 (Model Monitoring): Support both input-based and output-based functionality/model monitoring for AI/ML based positioning.
- Proposal 18 (Model Monitoring): For Case 1 label-based monitoring at LMF, consider Options A-1 (LMF determines labels from legacy metrics) and A-2 (UE determines labels from legacy assistance info).
- Proposal 20 (Model Selection): Model/functionality selection or switching could be realized via explicit indication or implicitly based on configurations related to data collection (time/area).
- Proposal 21 (Consistency): The measurement validity area configuration provided for model training is assumed to apply also for model inference to ensure consistency.
- Proposal 25 (Consistency): For Case 1, specific assistance data (e.g., TRP coordinates, beam info) can be optionally provided explicitly; if not, consistency is indicated via Associated ID(s).