R1-2508638
discussion
Evaluation Assumptions for 6GR Air Interface
From AT&T
Summary
This document from AT&T contains 6 proposals and 4 observations focused on establishing realistic evaluation assumptions for 6G Radio (6GR) air interface design. Key areas addressed include deployment modeling with specific ISD values, early integration of energy efficiency metrics, and traffic modeling for AI/ML services where Option-1a and Option-1c are rejected in favor of Option-1b.
Position
AT&T requires interference modeling, channel estimation, and traffic modelling to use realistic assumptions, and mandates that Energy Efficiency metrics be included as key performance indicators from 6GR day 1 rather than as add-ons. They specify a fixed ISD of 1732m for the SMa deployment scenario based on North American deployments. For traffic modeling, AT&T presents a technical case against Option-1a (token-based model) arguing inapplicability because client-side tokenization and over-the-air token transmission are not mainstream for commercial multimodal LLMs like ChatGPT or Gemini. They also oppose Option-1c (reusing FTP-3 or XR models) on the grounds that FTP-3 is designed for greedy file transfers and the XR model prescribes periodic high-rate flows, neither of which captures the irregular, multi-phased nature of LLM traffic. They propose using Option-1b, which models traffic at the PHY layer packet level with parameters such as packet size, delay requirement, and incorporates distinct LLM session phases including time-to-first-token, stream rate, and think-time.
Key proposals
- Proposal 1 (Deployment Scenarios and Antenna Modeling): For 6GR evaluations, realistic assumptions are used for interference modeling, channel estimation, and traffic modelling.
- Proposal 2 (Deployment Scenarios and Antenna Modeling): For 6GR evaluations, include Energy Efficiency metric(s) as key performance metrics from day 1.
- Proposal 3 (Deployment Scenarios and Antenna Modeling): For SLS evaluations, the ISD for SMa deployment scenario is 1732m.
- Proposal 4 (Traffic Models): For traffic modeling of AI/ML services for 6GR evaluations, do not pursue option-1a.
- Proposal 5 (Traffic Models): For traffic modeling of AI/ML services for 6GR evaluations, do not pursue option-1c.
- Proposal 6 (Traffic Models): For traffic modeling of AI/ML services for 6GR evaluations, use Option-1b.