R1-2409660
discussion
Discussion on LS on applicable functionality reporting for beam management UE-sided model
From vivo
vivo's prior position on
5
at
RAN1#118bis
· AI-synthesized, paraphrased
verify sources →
Advocates for mandatory associated IDs as essential for AI/ML training-inference consistency and FG-specific granularity in functionality definition, while opposing providing inference configuration in early steps.
Summary
vivo analyzes three options for applicable functionality reporting for AI/ML beam management UE-sided models, arguing that Option 1 is inefficient due to signaling waste and CSI framework conflicts. The document prioritizes Option 2 and Option 3, which decouple associated ID interaction from inference configuration to ensure training-inference consistency. It contains two main proposals: prioritizing Options 2/3 and providing specific answers to RAN2's Question 4 regarding NW-side conditions.
Position
vivo presents a technical case against Option 1, arguing it leads to significant wastage of NW signaling, communication resources, and UE storage resources due to invalid inference configurations. They oppose Option 1 because it contradicts the existing CSI framework, where resource measurement cannot be executed immediately upon receiving configuration. vivo requires that Associated IDs be provided by the NW before the UE determines applicable functionalities, citing that consistency of NW-side additional conditions across training and inference is crucial to prevent performance degradation. They propose that Associated IDs in Step 3 are separate from inference configuration, and that information such as preferred Set B patterns can be included in Step 3 but does not constitute inference configuration. vivo prioritizes Option 2 and Option 3, which allow the NW to understand supported models before configuring CSI reports, thereby aligning with the current CSI framework and avoiding futile configurations.
Key proposals
- Proposal 1 (Conclusions): Prioritize Option 2 and Option 3 in the applicable functionality determination procedure.
- Proposal 2 (Conclusions): RAN1 should answer Q4-1 by stating it is not feasible for UE to decide applicable functionalities without NW-side additional condition (associated IDs).
- Proposal 2 (Conclusions): RAN1 should answer Q4-2 by stating that while other information can be transmitted in Step 3, it is not inference configuration.
- Proposal 2 (Conclusions): RAN1 should answer Q4-3 by stating that Associated IDs in Step 3 are separate from inference configuration.
- Proposal 2 (Conclusions): RAN1 should answer Q4-4 by stating that detailed information like Set B patterns can be configured but is not inference configuration.
- Observation (Option 1 Analysis): Option 1 causes significant wastage of NW signaling, communication resources, and UE storage resources due to invalid inference configurations.
- Observation (Option 1 Analysis): Option 1 contradicts the existing CSI framework because resource measurement and report feedback cannot be executed immediately after receiving the configuration.
- Observation (Option 2/3 Analysis): Options 2 and 3 allow the NW to have a full understanding of supported models before configuring CSI reports, avoiding useless configurations.
- Observation (Option 2/3 Analysis): In Options 2 and 3, if a periodic CSI report is configured in Step 5, the UE can directly perform inference, aligning with the current CSI framework.
- Observation (Technical Case): Ensuring consistency of NW-side additional conditions across training and inference is crucial to prevent unpredictable performance degradation.
- Observation (Technical Case): The applicability of functionality at the UE side is dependent at least on the provided associated ID.
- Observation (Step 3 Content): In Step 3, NW can indicate supported/preferred Set B patterns and other inference parameters from the NW perspective, but these are distinct from inference configuration.
- Observation (Signaling Flow): Associated IDs must undergo interaction prior to inference configuration to address consistency issues.
- Observation (Resource Efficiency): Option 1 leads to futile configurations if the UE receives a local associated ID for an unknown cell or one not trained during the training phase.
- Observation (Signaling Overhead): Providing candidate inference configuration parameters and associated ID lists in Step 3 (Option 2) may reduce signaling overhead for UE UAI reporting in Step 4.