Study on consistency issue for CSI prediction

Summary

Vivo presents a study on training-inference consistency issues for AI/ML-based CSI prediction in NR, analyzing how TXRU virtualization mapping mismatches cause significant performance degradation (up to 44.4% loss). The document contains 2 proposals and 5 observations addressing generalization performance impacts.

Position

Key proposals

• Proposal 1 (Discussion): Follow the solution based on associated ID as BM use case to address the consistency issue of training and inference for UE-side model
• Proposal 1 (Conclusions): At least TXRU mapping is identified as one NW-side additional condition which will have significant impact on generalization performance
• Proposal 2 (Conclusions): Follow the solution based on associated ID as BM use case to address the consistency issue of training and inference for UE-side model
• Observation 1: At least for 60km/h, or when the proportion of high-speed data collected is high, significant performance loss can be caused by the mismatch in terms of TXRU virtualization mapping model between training and the inference data
• Observation 2: When the user distribution ratio is 8:2, the SGCS loss of different TXRU virtualization mappings in the generalization evaluation case2 ranges from -5.6% to -9.4%
• Observation 3: When the user distribution ratio is 100% outdoor, the SGCS loss of different TXRU virtualization mappings in the generalization evaluation case2 ranges from -13.0% to -44.4%
• Observation 4: At least TXRU mapping is identified as one NW-side additional condition which will have significant impact on generalization performance
• Observation 5: The SGCS loss of different TXRU virtualization mappings in the generalization evaluation case3 ranges from -3.7% to -14.6%

Revision chain

1. R1-2409670 discussion revised
   Study on consistency issue for CSI prediction
2. R1-2410673 ← you are here discussion not treated Final

Your notes