R1-2500060
discussion
AI/ML for Positioning Accuracy Enhancement
From Ericsson
Ericsson's prior position on
9.1.2
at
RAN1#119
· AI-synthesized, paraphrased
verify sources →
Presents a technical case against path-based measurements and CIR/phase inputs due to signaling overhead and alignment difficulties, initially favoring sample-based measurements with total-power PDP inputs. Later documents advocate supporting both sample-based and path-based measurements as a compromise to avoid blocking progress, while continuing to push against CIR support. Proposes using associated IDs or explicit assistance data from the LMF to ensure consistency between training and inference, and supports label-free self-monitoring by the model inference entity.
Summary
Ericsson presents a comprehensive contribution on AI/ML for NR positioning accuracy enhancement, focusing on integrating AI/ML methods with existing protocols, defining model inputs/outputs, and establishing training data collection and monitoring frameworks. The document contains 66 proposals and 38 observations across various sections, addressing protocol integration, signaling enhancements for model inputs, model output reporting, training data metadata, and lifecycle management.
Position
Ericsson proposes integrating AI/ML positioning into existing DL-TDOA and UL-TDOA/multi-RTT frameworks rather than defining new procedures, assigning procedural decisions to RAN2 and RAN3. They require the use of total-power PDP inputs summed over all receive antenna ports, explicitly opposing multi-port PDP due to doubled signaling size and marginal performance gains. Ericsson presents a technical case against supporting phase information or CIR inputs for model inference, citing random initial phase issues, high signaling overhead, and difficulty in aligning training/inference phases. They propose decoupling the LOS/NLOS indicator (reflecting physical LOS) from timing information (reflecting virtual LOS) in Rel-19 Case 3a reports. For training data, they require mandatory time stamps for unambiguous Part A/B pairing and support only UE location coordinates as ground truth labels. They establish self-monitoring as the baseline for model performance, with the inference entity responsible for metric calculation, while LMF handles functionality-level management. Finally, they propose using associated IDs to verify consistency of network-side conditions between training and inference.
Key proposals
- Proposal 1 (Sec 2.1): RAN1 assumes Case 1 (UE-based) is supported using UE-based DL-TDOA procedures as a foundation, leaving procedure definition to RAN2.
- Proposal 2 (Sec 2.2): RAN1 assumes Case 3a (gNB-based) is supported using UL-TDOA or multi-RTT procedures as a foundation, leaving procedure definition to RAN3.
- Proposal 3 (Sec 2.3): RAN1 assumes Case 3b (LMF-based) sample-based reporting can be differentiated from legacy UL RTOA reports by involved nodes.
- Proposal 4 (Sec 3.1.1): Support Nt values of {32, 64} for sample-based measurements, excluding 128 due to lack of significant gain.
- Proposal 12 (Sec 3.1.3): Update DL PRS-RSRPP and UL SRS-RSRPP definitions to use total-power PDP (summed over all receive antenna ports) for model input.
- Proposal 13 (Sec 3.2): Do not support phase information (including CIR and single phase value) for determining model input due to random initial phase issues and signaling overhead.
- Proposal 17 (Sec 3.2.4): Down-prioritize signaling approaches for CIR model input types for Case 3b and 2b due to large signaling sizes and negative network impact.
- Proposal 19 (Sec 4.1.1): For Case 3a, timing information in Rel-19 reports is for virtual LOS link, while legacy reports are for physical LOS link.
- Proposal 25 (Sec 4.1.1.3): LMF shall be able to distinguish whether timing information is from Rel-19 (virtual LOS) or legacy (physical LOS) reports.
- Proposal 29 (Sec 4.1.3): Measurement reports for Case 3a and 2a shall contain an indicator that AI/ML was used to produce the measurements.
- Proposal 31 (Sec 5.1.1): Time stamps are mandatorily provided by the training data generation entity for both Part A and Part B to allow unambiguous mapping.
- Proposal 37 (Sec 5.1.3): Support only ground truth labels in the format of UE location coordinates (x, y, z) for training data collection Part B.
- Proposal 42 (Sec 5.2): The training dataset is attached with metadata describing context information, including assistance data, signal measurement configuration, and environment conditions.
- Proposal 53 (Sec 6.2): Self-monitoring is the baseline for AI/ML positioning cases, where the model inference entity performs monitoring with or without external ground truth label information.
- Proposal 61 (Sec 7.2.3): Support associated ID for indicating NW-side additional conditions to ensure consistency between training and inference for UE-side models.