R1-2508732
discussion
AIML use cases for 6GR air interface
From OPPO
Summary
This document presents OPPO's views on AI/ML use cases for the 6G radio air interface (FS_6G_Radio), containing 40 proposals and 22 observations. The contribution advocates for an AI-Intrinsic 6GR design, proposing a wide range of use cases from DMRS and CSI-RS overhead reduction to modulation enhancement and token communication, along with principles for selection, a development timeline, and a unified Life Cycle Management (LCM) framework.
Position
OPPO proposes prioritizing AI/ML-Intrinsic design that significantly enhances the basic components of the 6GR transceiver chain, selecting use cases based on significant performance benefits and well-balanced tradeoff among performance, computation complexity, and power consumption. They propose studying the SuperImposed Pilot (SIP) scheme as a candidate solution for DMRS overhead reduction, where DMRS and data are superimposed over the same resource elements with a fixed power ratio (e.g., 5% for DMRS, 95% for data). For CSI enhancement, they propose studying Joint Source and Channel Coding (JSCC) and Joint Source and Channel Coding and Modulation (JSCCM) schemes, including lightweight UE-side designs using linear projection or simple DNN-based models. They argue for a unified LCM framework applicable to non-CSI cases, rejecting the legacy CSI-framework-based LCM as inapplicable to the diverse use cases expected in 6GR. For token communication, they propose that 6G core design shift from 'bit stream quality-centric' to 'service quality-centric', with inherent tolerance to token transmission errors and selective transmission of partial tokens with high importance, requiring specification impact on token error identification, scheduling, and HARQ.
Key proposals
- Proposal 1 (Sec 2.1): Consider the following principles to select AI/ML use cases for 6GR study.
- Proposal 5 (Sec 2.2): Suggested timeline to select AI/ML use cases for Rel-20 6GR, with Stage 1 in 25' Q4 and Stage 2 from 26' Q1 to 27' Q2.
- Proposal 10 (Sec 2.3): Strive for unified LCM framework for 6GR, including data collection, model monitoring, model paring, and LCM-related operation.
- Proposal 15 (Sec 3.1.1): To boost spectrum efficiency for 6GR, select the DMRS overhead reduction as an AI/ML use case.
- Proposal 17 (Sec 3.1.2): For extremely large antenna array of 6GR, select the CSI-RS overhead reduction as an AI/ML use case.
- Proposal 19 (Sec 3.2): To facilitate downlink MIMO transmission, select the CSI enhancement as an AI/ML use case.
- Proposal 24 (Sec 3.3): To optimize modulation over the legacy uniform constellations, select the AI/ML based modulation as an AI/ML use case.
- Proposal 29 (Sec 3.4): For 6G, NW side DPoD (Digital Post-Distortion) could be taken into account.
- Proposal 35 (Sec 3.5): Study the DCI decoding with prior information, which comes from last successful DCI.
- Proposal 36 (Sec 3.6): The new variants of AI/ML use cases studied in NR should be separately evaluated, if selected as use case for 6GR.
- Proposal 37 (Sec 3.7): To efficiently support AI services and token communications, studies are needed in RAN1, including service awareness, token error identification, and evaluation framework.
- Proposal 25 (Sec 3.3.1): Considering legacy uniform constellations as benchmark, study AI/ML-based modulation and/or demodulation to achieve better performance via one-sided model and downloadable constellations.
- Proposal 28 (Sec 3.3.2): To joint design modulation and MIMO for better performance, study the cross-layer modulation and precoding.