S4-260561 - AI Summary

pCR [FS_6G_MED] Considerations on Work Topic 1 update on 6.2 for Network-Assisted Media Processing for Video Understanding and Enhancement

Back to Agenda Download Summary
AI-Generated Summary AI

pCR Summary: S4-260561 – Network-Assisted Media Processing in TR 26.870

Source: Huawei | Spec: TR 26.870 | Meeting: SA4#135-bis-e


What is Being Proposed

This pCR proposes additions to TR 26.870 (FS_6G_MED) to introduce network-assisted media processing as a study topic within the 6G media delivery architecture, motivated by the computational gap between device capabilities and AI-based video processing demands.

Technical Rationale

  • Video LLMs (e.g., Qwen3-VL-8B, InternVL3.5) require 270–1950 TFLOPS per image at 1080p; current mobile devices offer ~40 TFLOPS — making on-device video understanding infeasible
  • Video pre-processing tasks (deblurring via DeblurGAN/MPRNet, super-resolution via ESRGAN/SwinIR, low-light enhancement via Zero-DCE/RetinexNet) require 0.5–3.5 TFLOPS/frame — also challenging for lightweight devices (AR glasses, etc.)
  • Motivation drawn from SA1 TR 22.870 use cases: AI-based video analysis (6.19), mobile embodied AI offloading (6.28), AI for disability support (6.38), XR rendering offload (9.4)

Proposed Changes to TR 26.870

  1. Clause 6.2.1.1 – Adds "network-assisted media processing for compute-intensive tasks" to the list of architectural aspects to be studied
  2. New Clause 6.2.1.X – Introduces "Media Functional Entities for Network-Assisted Processing," covering: network-side video understanding (analysis, scene recognition, object detection), network-side video pre-processing (deblurring, super-resolution, low-light enhancement), and dynamic offloading decisions based on device/network conditions
  3. Clause 6.2.1.5 (Key Issues) – Adds a new key issue: how Media Functional Entities can be enhanced to support network-assisted media processing for resource-constrained devices
  4. New Annex A.2.x – Documents the SA1 use cases and observations/proposals as study background for multi-modal data processing

Proposals/Conclusions

  • Observation 1: Network side should undertake compute-intensive video understanding (videoLLMs) on behalf of devices
  • Observation 2: Network side should handle video pre-processing for lightweight devices
  • Proposal 1: Explore architectural impacts of introducing videoLLMs into the media delivery architecture
  • Proposal 2: Explore enhancements to Media Functional Entities to support video pre-processing capabilities
Document Information
Source:
Huawei Device Co., Ltd
Type:
pCR
For:
Agreement
Original Document:
View on 3GPP
Title: pCR [FS_6G_MED] Considerations on Work Topic 1 update on 6.2 for Network-Assisted Media Processing for Video Understanding and Enhancement
Agenda item: 11.1
Agenda item description: FS_6G_MED (Study on Media aspects for 6G System)
Doc type: pCR
For action: Agreement
Release: Rel-20
Specification: 26.87
Version: 0.2.2
Related WIs: FS_6G_MED
Spec: 26.87
Contact: Shuxin Ouyang
Uploaded: 2026-04-07T06:12:20.387000
Contact ID: 103366
TDoc Status: available
Reservation date: 07/04/2026 01:55:08
Agenda item sort order: 58