R1-2409455
discussion
Discussion on AI/ML for beam management
From InterDigital
Summary
InterDigital presents 29 proposals and 24 observations regarding AI/ML for beam management in NR, focusing on configuration frameworks, reporting overhead reduction, and lifecycle management. The document argues for Option 2 for UE-side model applicability to minimize signaling overhead and supports a unified CSI-ResourceConfigId for both Set A and Set B beams. It further proposes mechanisms for QCL parameter estimation for unmeasured beams, pattern-based reporting, and dynamic switching of inference location between UE and network.
Position
InterDigital supports Option 2 for UE-side model applicability identification, arguing that Option 1 requires excessive configuration overhead for candidate CSI report configurations. They prefer Alt 2 for beam configuration, utilizing a single CSI-ResourceConfigId for both Set A and Set B to minimize specification impacts while supporting scenarios with non-transmitted Set A beams. InterDigital proposes supporting QCL parameter estimation for unmeasured beams via neighboring beams, citing performance data showing Doppler shift errors below 10 Hz and RMS delay spread below 15 ns for 90% of the time. They support UE-assisted performance monitoring (Option 2) and dynamic switching of inference location between UE and network to balance computational load and reporting overhead. Additionally, they propose enhanced reporting for network-sided models, including up to 64 RSRP values and pattern-based reporting mechanisms to reduce overhead without sacrificing prediction accuracy.
Key proposals
- Proposal 4 (Applicability of inference for UE-side model): Support Option 2 for applicability identification, using gNB-configured candidate inference parameters rather than full CSI report configurations to reduce overhead.
- Proposal 5 (Configuration of Set A and Set B): Support Alt 2, where one CSI-ResourceConfigId is configured for both Set A and Set B, minimizing specification impacts.
- Proposal 8 (Configuration of Set A and Set B): Support CSI resource configurations with or without physical RS transmission parameters (e.g., resourceMapping) to allow for Set A beams that are not physically transmitted.
- Proposal 9 (QCL measurements for unmeasured Set A beams): Support configuration of RS resources to estimate QCL parameters for RS resources without transmission in Set A, utilizing neighboring beams.
- Proposal 10 (Pattern based beam indication): Support a beam indication mechanism using a beam pattern and corresponding TCI states for indicated beam patterns.
- Proposal 11 (Reporting beam measurements/predictions): Support a shared CPU counter between legacy and AIML-based CSI reporting to ensure efficient utilization of UE processing capabilities.
- Proposal 13 (Reporting beam measurements/predictions): Support reporting L1-RSRPs and beam information for up to M beams within an X dB gap to the largest measured L1-RSRP to reduce overhead.
- Proposal 16 (Reporting beam measurements/predictions): Support reporting prediction results of multiple future time instances in one report, aligned with CSI-Reporting periodicity.
- Proposal 20 (Reporting beam measurements/predictions): For network-sided models, support enhanced UE reporting of up to 64 RSRP values for the whole Set A over multiple time instances without explicit CRIs/SSBRIs.
- Proposal 22 (LCM and Consistency): Support reporting of UE-selected Set B based on a rule, such as a subset of the best measured beams, to improve prediction accuracy.
- Proposal 25 (LCM and Consistency): Support Option 2 (UE-assisted performance monitoring) to reduce reporting overhead and gNB computation load.
- Proposal 29 (Location of AIML Inference): Support a procedure to dynamically switch AIML inference location between UE and network based on workload or prediction performance.