R1-2410019
discussion
Specification impacts for AI/ML positioning
From Lenovo
Summary
Lenovo's comprehensive technical document on AI/ML positioning for 3GPP RAN1, presenting 30 proposals and 4 observations covering specification impacts for enhanced positioning accuracy. The document addresses measurement definitions, model inputs/outputs, training data collection, and implementation consistency across different AI/ML positioning use cases.
Position
Lenovo advocates FOR a hybrid measurement approach supporting both sample-based and path-based timing representations depending on the use case, with strong emphasis on reusing legacy frameworks where possible (LOS/NLOS indicators, assistance data signaling). They push FOR comprehensive training data collection frameworks with flexible entity assignments and AGAINST mixed measurement approaches within single use cases (e.g., using sample-based in training but path-based in inference).
Key proposals
- Proposal 1 (Sec 4.1): The maximum sample measurement window length corresponds to the maximum number of Nt samples, which may be up to Nt =128
- Proposal 6 (Sec 4.3): Support hybrid approach - sample-based time domain representation for Cases 1, 2b, and 3b; legacy-based path timing for Cases 2a and 3a
- Proposal 8 (Sec 5.1): Support channel observation measurements in the form of DL CIR measurements for DL-based Direct AI/ML positioning based on DL-PRS
- Proposal 11 (Sec 5.2): The legacy LOS/NLOS indicator reporting framework can be re-used for the model output generated by AI/ML
- Proposal 15 (Sec 7.2): Ground truth label content should comprise PRU UE location information, location uncertainty, and timestamps for different positioning cases
- Proposal 18 (Sec 8.1): Introduce a new UE-based positioning method for Direct AI/ML Positioning-Case 1
- Proposal 21 (Sec 8.2): Consider three scenarios for training data transfer - LMF-side using 3GPP signaling, UE-side using proprietary signaling, gNB-side using proprietary signaling
- Proposal 25 (Sec 9.1): For target UE model monitoring in AI/ML positioning Case 1, support Options A-1, A-2 and A-3 for receiving ground truth label information
- Proposal 27 (Sec 9.3): The inference positioning configuration should be provided using the LPP ProvideAssistanceData message
- Proposal 28 (Sec 9.4): Network-side conditions should include DL-PRS configurations, TRP/ARP location info, synchronization info; UE conditions should include PRU measurements and requested training/inference areas