R1-2508697
discussion
Discussion on AI-based Smart Radio for 6G Air Interface
From ZTE
Summary
This ZTE contribution to 3GPP RAN1#123 proposes a three-phase discussion framework for native AI integration into 6G Radio (6GR) and presents detailed AI/ML use cases categorized under Efficient, Green, and Autonomous 6G air interface enhancements, containing over 60 observations and 30+ concrete proposals with supporting simulation results.
Position
ZTE proposes a three-phase approach (categorization-level down-selection, detailed analysis/simulation with framework study, and normative work in Release-21) for finalizing AI/ML in the 6GR interface, while allowing 5G-A AI/ML use cases such as AI CSI prediction and AI beam prediction to evolve directly into 6G without duplicated study. ZTE requires 6GR to be designed with flexibility to accommodate both AI-based and non-AI-based solutions, prioritizing use cases with compelling trade-off between performance and complexity, and categorizes all AI/ML use cases under three pillars: AI+ Efficient 6G (covering downloadable codebook, JSCC/JSCCM for CSI, CSI compression with SRS, low density CSI-RS with two-sided model achieving 132.6% SGCS gain, low density DMRS and SIP, constellation design with up to 1dB BLER gain, cross-layer modulation, AI-based UL precoding enhancement, and multi-TRP/cell beam management with 10% prediction accuracy gain), AI+ Green 6G (covering AI-based SSB prediction for RACH procedure with UE-sided model, APU management, model states management, and FLOPs-based power consumption modeling), and AI+ Autonomous 6G (covering AI-based traffic prediction and unified autonomous AI/ML framework). ZTE presents simulation results showing downloadable codebook achieves 4.9%~19.9% SGCS gain over eType II, JSCCM-based CSI feedback achieves at least 13.3% SGCS gain over separate source-channel coding at low SNR, and CSI compression with SRS achieves 75%~130% SGCS gain over pure SRS measurements depending on payload configuration.
Key proposals
- Proposal 1 (Sec 2.1): Based on the updated contents in section 5.2.2, the table can be updated according (Two-sided model for Sparse CSI-RS).
- Proposal 1 (Sec 2.2): Based on the updated contents in section 5.2.1, the table can be updated according (CSI compression + SRS).
- Proposal 1 (Sec 2.3): Based on the updated contents in section 5.2.5, the table can be updated according (AI enhanced UL precoding).
- Proposal 1 (Sec 2.4): Based on the updated contents in section 5.3.1, the table can be updated according (AI beam prediction for initial access).
- Observation 1 (Sec 3): 6G is envisioned as a Smart Radio capable of supporting native AI with design principles including flexibility to accommodate both AI-based and non-AI-based solutions.
- Proposal 1 (Sec 5.1): RAN1 studies at least the following AI/ML use cases for 6GR.
- Proposal 1 (Sec 5.2): RAN1 considers the following categorization of AI/ML use cases for 6GR.
- Proposal 1 (Sec 5.3.1.1): RAN1 studies AI-generated downloadable codebook or downloadable basis set for 6GR.
- Proposal 1 (Sec 5.3.1.2): RAN1 studies AI-based JSCC/JSCCM for 6GR.
- Proposal 1 (Sec 5.3.1.3): RAN1 studies integration of uplink SRS measurements and downlink AI/ML CSI compression for 6GR.
- Proposal 1 (Sec 5.3.2.1): RAN1 studies low density CSI-RS with AI prediction for 6GR for UE-side model, NW-side model and two-sided model.
- Proposal 1 (Sec 5.3.3.2): Study AI-based DMRS enhancement for 6GR, including at least low density DMRS and SIP.
- Proposal 1 (Sec 5.3.4.1): Study AI-based constellation design for 6GR.
- Proposal 1 (Sec 5.3.6): RAN1 studies AI-based multi-TRP/cell beam management enhancement for 6GR, including spatial domain beam prediction and temporal domain beam prediction.
- Proposal 1 (Sec 5.4.1): RAN1 studies AI for energy saving at least from two directions: reduced reference signal transmission/reception/measurement in time/frequency/spatial domain based on AI prediction, and more accurate scheduling/activation/deactivation/resource management based on AI prediction.