Proposed Updates to Terms of Reference for SA4 Sub Working Groups addressing AI data
This document proposes updates to the SA4 Working Group Terms of Reference (ToR) to address AI-related media services and data characteristics, particularly in the context of 6G development. The proposal has been under discussion since SA4#134 and SA#110, with multiple related papers submitted addressing the scope and responsibilities for AI traffic characteristics across different 3GPP working groups.
SA4#134 and SA#110 Discussions:
- Initial ToR update version discussed but no consensus reached at SA4#134
- SA#110 received multiple papers (SP-251499, SP-251540, SP-251576) discussing AI traffic characteristics responsibilities
- Key debate centered on whether SA4's mandate should be restricted to media-related AI or generalized to any AI/ML model
- Discussion included clarification of responsibilities between SA1, SA2, and SA4 regarding AI traffic characteristics
- SA#110 postponed the ToR update without providing clear guidance to SA4
Key Observations from Related Papers:
- AI traffic will be one of the most important data types in 6G networks with dedicated characteristics
- SA4 holds expertise in detailed traffic analysis (demonstrated in XR work in Releases 18-19)
- SA2 also scoped work items for AI traffic system impacts and QoS framework requirements
- Need for clear delineation of responsibilities to avoid duplication and ensure efficient collaboration
Triggering Event:
The approved FS_6G_MED study includes objectives to "collect and study AI representation formats and traffic characteristics used in AI-related media services" based on use cases including agents, multi-modal large language models, and diffusion models.
Expanded Scope for AI-Based Media Services:
- Added explicit mention of "AI-based media services" alongside existing services (XR, gaming)
- Included "use of artificial intelligence (AI) and machine learning models for multimedia"
- Added "[media-related] the relevant AI representation formats and data types for these services including but not limited to those making use of multi-modal large language and diffusion models"
- Note: The text shows "[media-related]" in brackets, suggesting optional qualifier that was debated
Current Responsibilities Acknowledgment:
Updated to reflect SA4's current work on XR-based services, Next Generation Video for 5G, Media Distribution, Media Cloud and Edge Processing, Glass-based AR, VR conferencing, and Immersive Voice and Audio Services.
Codec Specifications (First Bullet):
- Expanded from just codecs to include "transport and handling of such media including related session descriptions and storage formats"
Media Services and Architectures (Second Bullet):
- Added "media APIs" and "media profiles" to the list of responsibilities
- Clarified "media-centric cloud and edge computing architectures"
Guidance to Other 3GPP Groups (Third Bullet):
- Maintained responsibility for providing guidance on QoS parameters, traffic characteristics, and system implications
Quality Evaluation (Fourth Bullet):
- Expanded to include "multimedia and AI data representation quality evaluation"
- Added "new evaluation methods" to the scope
Third-Party Media Services Support (Fifth Bullet - Key Addition):
- Major expansion: Added comprehensive responsibility for supporting third-party media services and applications
- Explicitly includes "definition of traffic characteristics and QoE requirements for different services and data types"
- Critical addition: "including in particular also emerging including AI-based media applications and services"
- This bullet establishes SA4's role in providing traffic characteristics for AI-based applications to other WGs
End-to-End Performance (Sixth Bullet):
- Expanded to include "multimedia services" alongside speech, audio, and video
Interoperability (Seventh Bullet):
- Clarified scope as "from codec and media transport point of view"
Updated Services List:
The responsibilities now explicitly cover services including but not limited to:
- Multimedia telephony
- Mission critical services
- Multimedia unicast and multicast/broadcast streaming
- Content delivery
- Online gaming
- Extended realities (XR)
- New: Services based on cloud and edge computing architectures
- New: Services based on artificial intelligence (AI)/Machine Learning (ML) for multimedia
Multimedia Broadcast Streaming (MBS) SWG:
- Responsibilities include integration of codecs, system and delivery aspects
- Covers download, streaming, and messaging services
- Includes QoE, energy efficiency metrics, and traffic characteristics definition
Real-Time Communication (RTC) SWG:
- Focuses on real-time communication services including multimedia telephony and XR communications
- Covers services with real-time constraints
- Includes cloud/edge computing architectures updates and traffic characteristics
Video SWG:
- Specification of codecs for video, graphics, and visual media including immersive formats
- Key addition: "Artificial Intelligence (AI)/Machine Learning (ML) applied to multimedia scenarios, including neural network model formats and related optimization and compression aspects"
Audio SWG:
- Development and maintenance of speech/audio codecs and quality evaluation
- Covers immersive audio formats for multimedia telephony and XR communications
- Includes both capture and rendering aspects
SWG Operating Model:
- SA4 SWGs operate under SA4 WG
- All document decisions must be confirmed at SA4 level
SA4 maintains regular communication with:
- Other 3GPP WGs
- ISO/IEC JTC 1/SC 29
- DASH-IF
- IETF
- ITU-T Study Groups 12 and 16
The proposed updates establish SA4 as the primary 3GPP WG responsible for:
1. Identifying data types and traffic characteristics for AI/ML media services
2. Providing traffic models and characteristics to other WGs (particularly RAN1, SA2) that request such information
3. Defining AI representation formats relevant to media services
4. Evaluating quality aspects of AI-based media data
The use of bracketed "[media-related]" in the text suggests ongoing debate about whether SA4's AI responsibilities should be:
- Option 1: Limited to media-related AI only (more restrictive)
- Option 2: Extended to broader AI representation formats and data types (more expansive)
This ambiguity reflects the unresolved discussions from SA#110 regarding the division of responsibilities between SA2, SA4, and other WGs.
The document proposes agreement on these ToR updates and submission for approval to SA plenary in March 2026 (noting this is a revision of SP-241362).