R1-2508549
discussion
Discussion on AIML in 6GR interface
From NEC
Summary
NEC presents 13 proposals on AI/ML for the 6G radio interface, covering use case selection principles, extensions of 5G Advanced use cases, new use cases like RS overhead reduction and AI/ML receivers, joint processing of multiple functionalities, and a unified AI/ML framework emphasizing coordination with SA and dynamic processing capability management.
Position
NEC proposes a structured four-pronged approach to 6G AI/ML use case selection, beginning with 5GA extensions for fast TCI state activation/indication with predicted beams, MTRP, BFR, and Tx-Rx beam pair prediction. They propose studying AI/ML-based RS overhead reduction across initial access, BM/CSI, and channel estimation procedures, covering DMRS, SRS, CSI-RS, and SSB in both DL and UL. For receiver design, they present two distinct options including superimposed DMRS with data for joint channel estimation and data detection. They propose studying two-sided AI/ML modulation/demodulation schemes and joint processing of multiple functionalities through coupling independent AI/ML models. On framework, they require 3GPP coordination with SA from the beginning and propose a unified AI processing capability with dynamic management of processing resources accounting for non-3GPP application impact on UE capability.
Key proposals
- Proposal 1 (Sec 2.1): General consideration of 6G RAN1 use cases selection is based on identifying existing 5GA use cases and their extensions, studying new use cases on performance and cost, studying new use cases for traditional communication block enhancement/replacement, and studying joint AI/ML processing of multiple functionalities.
- Proposal 2 (Sec 2.2): Identify existing 5GA use cases and their extensions to be supported in 6G, such as AI/ML based BM with further enhancements on fast TCI state activation/indication with predicted beam(s), BFR, MTRP, inter-cell scenarios and Tx-Rx beam pair prediction, and CSI compression with time domain aspects like CSI prediction.
- Proposal 3 (Sec 2.3): Study AI/ML to reduce RS overhead in 6GR air interface across procedures (initial access, BM/CSI, channel estimation), domains (time, frequency, spatial), RS types (DMRS, SRS, CSI-RS, SSB), and scenarios (DL and UL).
- Proposal 5 (Sec 2.4): Study AI/ML based channel estimation together with channel equalization, including Option 1: AI/ML-based channel estimation with a traditional channel equalization block, and Option 2: A joint AI/ML model for combined channel estimation and equalization.
- Proposal 6 (Sec 2.4): Study AI/ML based receiver design with DMRS superimposed with data, including Option 1: AI/ML-based channel estimation at the receiver, and Option 2: AI/ML-based joint channel estimation and data detection at the receiver.
- Proposal 7 (Sec 2.5): Study AI/ML based modulation and demodulation, including Option 1: AI/ML-based demodulation scheme at the receiver assuming traditional modulation by transmitter, and Option 2: Two-sided AI/ML-based modulation and demodulation scheme.
- Proposal 8 (Sec 2.6): Study joint AI/ML processing of multiple functionalities in 6GR and identify use cases achieved through coupling independent functionalities, e.g., joint CSI compression and prediction, joint modulation and precoding, joint channel estimation and demodulation.
- Proposal 9 (Sec 2.7): Consider the agenda items for further study of representative (sub-) use cases as summarized in Table 1.
- Proposal 10 (Sec 3): Study 6G AI/ML framework by coordinating with SA from the beginning.
- Proposal 11 (Sec 3): Study unified framework for AI/ML for 6GR ensuring high performance, security, and adaptability to all use cases, including unified data collection framework and unified LCM framework.
- Proposal 12 (Sec 3): Study a unified AI processing capability among multiple AI/ML functionalities, including mechanisms to manage processing resources for multiple concurrently running 3GPP-defined AI/ML functionalities and methods to account for dynamic impact of non-3GPP applications on UE's available processing capability.
- Proposal 13 (Sec 3): Study LCM framework for online training/federated learning-based solutions for 6GR.