R1-2603261 discussion

FL summary #1 for AI/ML in beam management

From Samsung
Status: noted
WI: NR_AIML_air
Agenda: 8.1
Release: Rel-19
Source: 3gpp.org ↗

Summary

This document summarizes the remaining issues and proposals for AI/ML-based beam management in NR Rel-19, specifically focusing on UE-side model performance monitoring, inference reporting, and processing timelines. It addresses critical specification gaps regarding PUCCH payload sizes for RS-PAI, PDSCH rate-matching around virtual CSI-RS resources, and CPU/APU occupation rules. The document records agreements reached in RAN1#124bis, including the adoption of zero-padding for small RS-PAI payloads and corrections to CPU occupation timelines for data collection.

Position

Samsung supports the majority view on zero-padding RS-PAI payloads to 3 bits for PUCCH transmission, arguing that padding after CSI multiplexing reduces overhead compared to per-report padding. They propose relaxing the configuration restriction for Set A resources to allow different time domain behaviors, enhancing flexibility for AI/ML inference configurations. Samsung requires clarification that CSI-RS resources configured only in Set A are not used for PDSCH rate-matching, thereby preventing unnecessary resource waste. They also support correcting the CPU occupation timeline for UE-side data collection to accurately include periodic CSI reports for beam management. Furthermore, Samsung argues for clarifying the determination of updated AI/ML CSI reports based on CPU and APU availability to ensure higher-priority reports are processed correctly.

Key proposals

Your notes

Private to your account