# RTC SWG Report Post SA4#134 ## Executive Summary The RTC SWG held two adhoc calls post SA4#134 with 15 input contributions and 24 participants in the first adhoc, and 10 input contributions and 30 attendees in the second adhoc. Two tdocs were agreed. ### Key Outcomes by Work Item: - **AIML_IMS_MED**: Agreed to create one CR to TS 26.114 as output. Based on SA2 LS response, agreed to complete work with local/split/remote on DC AS (instead of MF) and document limitations. Agreed on device-inference call flow as basis for further work. - **FS_DCTC_eQOS_MED**: Discussion on test platform continued. - **FS_Q4RTC_MED**: Agreed on work plan and discussed TR skeleton and basic QUIC protocol. ## AIML_IMS_MED (Media Aspects for AI/ML in IMS Services) ### Scope Definition and Architecture **S4aR250216 - Scope in Light of SA2 LS** The group discussed the scope following the SA2 liaison response. Key decisions: - **Agreed** to complete work with local/split/remote inferencing on the DC AS (Data Channel Application Server) rather than the MF (Media Function) - Document limitations (e.g., DC AS not present for P2P case, so only UE inferencing may be possible for P2P) - Send LS to SA2 with more details on SA4's needs and scope - Take approach similar to Annex AC P2P and P2A2P: MF as relay, same approach for remote/split (P2A2P) and local (P2P) ### Call Flows **S4aR260004/S4aR260014 - Call Flow for Device Inferencing** Joint contribution from Samsung and InterDigital. After extensive discussion: - **Agreed as basis for further work** with modifications: - Steps 8-9 (model fetching by MF before application data channel establishment) removed - Editor's notes added regarding: - How to handle large models - Network capability considerations - Model selection mechanisms - Caching requirements in UE Key discussion points: - Model sizes can be significant (e.g., 80 MB), requiring consideration of download time and network impact - Application metadata would indicate supported tasks; JavaScript in application can check UE capabilities - MF acts as HTTP proxy and can cache - Need for both UE-side and network-side caching solutions - Partial model updates and model compression need to be addressed **S4aR250205 - Call Flow for Device Inferencing (Samsung)** Initial proposal that was merged into S4aR260004. Noted for further work. **S4aR260007 - Call Flow for Device Inferencing Update (InterDigital)** Discussion focused on: - Clarification needed on whether this replaces normal BDC bootstrap procedures - Terminology should be moderated to clarify high-level stage 2 description - Protocol details need refinement **S4aR250206 - Call Flow for Network Inferencing** Noted. Needs update to reflect inferencing on DC AS instead of MF. Questions raised about model download source (DCAR vs DC AS). **S4aR250207/S4aR260008 - Call Flow for Split Inferencing** Noted. Similar to network inferencing but for split scenarios. Needs update to have split inferencing on DC AS rather than MF. Parameters in request/response messages need synchronization with other proposals. **S4aR250218 - Update Call Flow for Local/Remote/Split AIML Inferencing (InterDigital)** Noted. Needs update due to MF/DC AS changes. Proposed two data channels: one for configuration and one for data exchange. Terminology modifications needed for steps 1 and 3. ### Task and Model Management **S4aR250208 - On AI Application Tasks and Subtasks** Noted. Discussion on whether to use subtask concept or stay with application/task/model terminology. Clarity needed on subtask definition before adoption. **S4aR250209/S4aR260006 - AI/ML Media Processing and Task Updating (Nokia)** Noted. Addresses task reselection and task update scenarios: - Task update vs. new task may require different procedures - Distinction between updating a model vs. adding new capabilities (e.g., translation language) - In P2P case, may need application data channel - Clarification needed on whether "network functions" refers to MF or other entities **S4aR250211 - Adaptive Model Delivery (Nokia)** Noted. Proposes two cases: 1. Standard model delivery 2. Adaptive delivery with MF as proxy Discussion points: - Need clarification on model selection process (UE, user, or other entity) - Low precision and incremental model updates need more description - Make model update optional in second case ### Application Manifest and Negotiation **S4aR250213/S4aR260011 - On Application Manifest for AIML Applications (Nokia)** Noted. Proposes JSON format for application manifest with metadata. Needs alignment with other contributions. Email discussion to be triggered on this topic. **S4aR260009/S4aR260012 - Negotiation Messages (InterDigital)** Noted. Proposes metadata for negotiation messages. Questions raised: - Where will application info format be used? - Are we defining a new DC protocol? - Motivation is to identify what metadata needs to be considered **S4aR260010 - Negotiation Messages for Split Inferencing** Noted. Needs revision to sync with basic call flows. ### Codec and Compression **S4aR260005 - On NNC Decoding in Web Environment (Fraunhofer HHI, Nokia)** Noted. Discussion on Neural Network Coding (NNC) decoder in web environment. No alternative decoder currently available. Comparison document to be shared on differences. ### Specification Strategy **S4aR250215 - Discussion on TS** **Agreed** to add an annex to TS 26.114 and prepare CR accordingly. Consensus for smaller scope with annex in 26.114 rather than separate specification. **S4aR250217/S4aR250220 - Draft CR on AIML Processing in IMS Calls** Noted. Will be developed as CR to TS 26.114 based on agreed scope. ## FS_DCTC_eQOS_MED (Study on Usage of Dynamically Changing Traffic Characteristics and Enhanced QoS Support) ### Technical Report **S4aR250212 - Draft TR 26.823** Noted. Presented for information. Issues identified: - No revision marks and incorrect version number - Missing agreed document numbers in change history - Style conformance issues - Skeleton was only agreed in closing plenary Expected revision for next adhoc with: - Change marks - ETSI style compliance - Agreed documents listed in annex - Correct version number ### Evaluation Strategy **S4aR250219 - Evaluation Testing Strategy Proposal (InterDigital)** Noted. Discussion points: - Network impairment testing vs. in-the-field testing needs more work - No need to limit to one evaluation platform - RTT measurement possible also for QUIC - Need details on "specific" platform and parameter sources - Which specific application for AI testing - Hard to set up OPs for OTT; need reference platform with control - Joint meeting with other evaluation/studies to develop common evaluation framework - Reference to 26.955 for joint discussion - QoE metrics handling needs resolution ## FS_Q4RTC_MED (Study on QUIC for RTC Media) ### Work Plan **S4aR250210/S4aR250221 - Draft Work Plan for FS_Q4RTC_MED** **Agreed** (S4aR250221). Discussion points: - Note on normative work to be triggered before end of study - Evaluation work needs alignment with other studies - Need clear item in timeplan on common evaluation framework - Coordination with QStream starting from Goa meeting - Careful on "common" aspects as different tools may be needed ### Protocol Overview **S4aR260001/S4aR260013 - Overview of QUIC-based Media Delivery Protocols (Nokia)** Noted (S4aR260013). Good base document. Comments to be provided by email on requirements and other aspects. ### Technical Report Skeleton **S4aR260003 - Draft TR 26.836 v0.0.1 (Skeleton) (NTT, Nokia)** Noted. Issues identified: - Clause 4.2.3 number missing - Some existing content can be moved to annex - Pros and cons need clarification (for RTC or general?) - Introduction/title of 5.3: F→Functional - Consider including evaluation under each application scenario - Version number needs offline check with Andrijana ## Meeting Details **First Adhoc Call:** - Date: December 17, 2025, 14:00-16:21 CET - Host: Nokia - 15 input contributions, 24 participants **Second Adhoc Call:** - Date: January 28, 2026, 15:00-17:15 CET - Host: Nokia - 10 input contributions, 30 attendees Minutes taken by Gaëlle Martin-Cocher (first adhoc) and Bo Burman/Shane He (second adhoc).