Technical Report Summary: TR 26.823 - Study on Dynamically Changing Traffic Characteristics and Enhanced QoS Support in 5GS for Media Applications

Document Overview

This is an early-stage (v0.2.0) study item from SA4 examining dynamic traffic characteristics and enhanced QoS support for media applications and services in 5G systems. The document is structured to analyze various media service scenarios, conduct experimental evaluations, and provide recommendations for enhanced QoS mechanisms.

Main Technical Contributions

1. Media Applications and Service Scenarios (Clause 5)

The document identifies and characterizes eight distinct media service scenarios:

5.2 Real-Time Communication for Conversational XR

• Description: Multi-user XR experiences with shared scenes, supporting VR, AR, and hybrid configurations
• Key Features:
• Users represented by 2D/3D avatars
• Multi-modal immersive experience (XR glasses/headsets, immersive audio, haptics)
• XR Scene Manager maintains synchronized scene state
• Spatial computing services for proper virtual object placement
• Dynamic Traffic Characteristics:
• Uplink: Periodic XR traffic (pose, gestures, eye tracking, voice) plus pseudo-periodic/event-based environmental data (images, video, sensor data)
• Downlink: Periodic XR traffic (rendered frames or scene state) plus additional media flows (haptics, 2D/immersive video, audio) that can be paused/resumed based on user interaction
• QoE Criteria:
• Pose-to-render-to-photon delay: <20-50ms (with pose correction)
• Round-trip interaction delay: <50ms (ultra-low), <100ms (low), <200ms (moderate)
• Trackable Creation Delay (TCD) and Trackable-Detection-to-Render-to-Photon (TDRP) metrics
• Time synchronization between haptics and other media

5.3 Video on Demand Streaming

• Description: Traditional on-demand video consumption using HLS/DASH/CMAF
• Implementation: Based on TS 26.501 procedures with DRM support
• Key Challenges:
• Data wastage (content downloaded but not watched)
• Network switching scenarios (mobile to WiFi)
• Different dynamic patterns from player pre-fetching
• Traffic Characteristics: Flexible segment download patterns using byte-range requests based on CMAF sidx box
• QoE Criteria:
• Video start-up time
• Average stream quality
• Re-buffering/freezing time
• Number of stream switching events
• QoS Examples: 5QI 4 (GBR, 300ms PDB, 10^-6 PER) or 5QI 6 (non-GBR, 300ms PDB, 10^-6 PER)

5.4 Live Streaming

• Description: Content made gradually available over time (TV broadcasts, live sports)
• Key Differences from VoD:
• Segments become available over time
• Limited buffering for low e2e delay services
• Typical playback-to-live-edge delay: 5-120 seconds
• Challenges:
• Segment latency issues from radio transmission errors or CDN problems
• Start-up delay and channel switching delay critical for low-delay services
• QoE Criteria:
• Representation switch events
• Average throughput
• Initial playout delay
• Media start-up delay
• QoS Examples: 5QI 6 (non-GBR) or 5QI 4 (GBR), both with 300ms PDB and 10^-6 PER

5.5 Short Form Video Download

• Description: Social media short videos (typically ≤120 seconds) with swipe-based navigation
• Implementation: Progressive download using 3GPP file format (TS 26.244)
• Key Characteristics:
• Single request downloads movie atom then mdat atom
• Low start-up delay possible
• Trade-off between QoE and bandwidth wastage
• Dynamic Behavior: Pre-fetching algorithms based on user behavior patterns
• QoE Criteria:
• Start-up delay after swipe
• Average throughput/bit-rate
• Switch time between videos
• Consumer bandwidth wasted
• Battery consumption
• QoS Examples: 5QI 4 (GBR) or 5QI 6 (non-GBR), 300ms PDB, 10^-6 PER

5.6 Media Upstream Transmission for AI Inferencing in XR

• Description: Uplink media transmission for both discriminative AI (object recognition, scene understanding) and generative AI (LLM/MLM applications)
• Use Cases:
• XR spatial computing functions (segmentation, labeling, semantic perception)
• Network-assisted video-based AI inference for mobile embodied AI
• GenAI applications (visual assistant, text-to-image generation)
• Dynamic Traffic Characteristics:
• Pseudo-periodic: Continuous object recognition with rate modulated by user mobility
• Aperiodic/Event-based: User-triggered queries (GenAI tasks)
• Implementation: Combines periodic XR traffic with AI inference traffic (text, audio, images, video, or AI intermediate data)
• QoE Criteria:
• Task-specific metrics (mAP for object tracking, WER for translation)
• Start-up latency, inference latency, delivery latency
• End-to-end latency: <50ms (ultra-low), <100ms (low), <200ms (moderate)
• QoS: Uplink latency requirements FFS; end-to-end requirements similar to interactive XR

5.7 High Quality Real-Time Conversational Communication

• Description: Enhanced conversational services beyond current MTSI capabilities
• Evolution: Moving toward HD/Full HD + HDR formats (1280x720 or 1920x1080) using HEVC
• Implementation: Based on TS 26.506 (5G RTC) and WebRTC architecture
• Rationale: Improved device rendering and camera quality enabling higher quality immersive conversations
• QoE Criteria:
• Session setup delay
• Average video bitrate
• Bitrate stability/switches
• Frame freezes
• Re-buffering frequency
• Intra/inter frame video quality
• QoS Example: 5QI 7 (non-GBR, 100ms PDB, 10^-3 PER)

5.8 Media Upstream/Downstream AI Inferencing (Split Inference)

• Description: AI/ML model split between UE and network for computation offloading
• Architecture: Two scenarios:
• Network-to-UE: First part executed in network, intermediate data sent to UE for completion
• UE-to-Network: First part executed in UE, intermediate data sent to network, results returned
• Benefits:
• Privacy-sensitive and delay-sensitive parts remain at UE
• Computation/energy-intensive parts offloaded to network
• Favorable transmission properties (privacy conservation, lower bit-rate)
• Use Cases:
• Home robot distributed inference (TR 22.870 section 6.10)
• Image recognition with resource optimization (TR 22.874 section 5.1)
• QoE Criteria:
• Time to result (response time)
• Accuracy of result
• Use-case dependent (real-time vs non-real-time)
• QoS Examples (for real-time inferencing):
• 5QI 85: Delay critical GBR, 5ms PDB, 10^-5 PER, 255 bytes max burst
• 5QI 88: Delay critical GBR, 10ms PDB, 10^-3 PER, 1125 bytes max burst
• 5QI 89: Delay critical GBR, 15ms PDB, 10^-4 PER, 17000 bytes max burst
• 5QI 90: Delay critical GBR, 20ms PDB, 10^-4 PER, 63000 bytes max burst

2. Experimental Evaluation Framework (Clause 6)

The document establishes a framework for experimental evaluation covering all identified scenarios: - Section 6.0: General experimental approach and test setup (void - to be completed) - Sections 6.1-6.5: Individual evaluations for each scenario (void - to be completed)

3. Solutions Framework (Clause 7)

Three solution areas are identified for further study:

7.1 Dynamic Traffic Characteristics

• To document traffic characteristics for each use case

7.2 Enhanced QoS Usage

• To present suggested solutions for enhanced QoS usage per case

7.3 Enhanced QoE Metrics

• To present any new QoE metrics identified

4. Cross-Cutting Technical Aspects

Protocol and Format Support

• RTP-based: Real-time communication scenarios (conversational XR, high-quality RTC)
• HTTP-based: Streaming scenarios (DASH, HLS, CMAF)
• Progressive download: Short-form video (3GPP file format)
• AI data formats: Intermediate inference data (TR 26.927)

QoS Mechanisms Referenced

• 5QI values from TS 23.501
• GBR and non-GBR QoS flows
• Packet delay budget (PDB) and packet error rate (PER) parameters
• Maximum burst volume for delay-critical services
• Dynamic policy using media session handler (TS 26.501)

Synchronization Requirements

• Time synchronization between multiple media types (haptics, video, audio)
• Spatial synchronization for XR content placement
• Round-trip delay control for interactive experiences

Document Status and Next Steps

This v0.2.0 version represents early-stage work with: - Completed: Service scenario descriptions, typical implementations, QoE criteria, and example QoS usage - To be completed: - Experimental evaluations (Clause 6) - Dynamic traffic characterization (Clause 7.1) - Enhanced QoS solutions (Clause 7.2) - Enhanced QoE metrics (Clause 7.3) - Analysis and recommendations (Clause 8) - Conclusions (Clause 9)

The study integrates documents from SA4#134 and SA4#135 meetings, building upon existing 3GPP work in XR (TR 26.928, TR 26.819), AI/ML (TR 26.927, TR 26.847), and 5G media streaming (TS 26.501, TS 26.506).

Summary of 3GPP Change Request S4-260229

Document Information

• Source: InterDigital Pennsylvania
• Title: Description of experimental approach and test setup for Conversational XR real-time communication
• Specification: 3GPP TR 26.823 v0.2.0 (FS_DCTC_eQoS_MED)
• Purpose: Agreement

Background and Motivation

This CR addresses the currently empty Clause 6.1.1 of TR 26.823, which focuses on the Study on dynamically changing traffic characteristics and usage of enhanced QoS support in 5GS for media applications and services. The study examines various use cases including real-time communication, streaming, short-form video download, and media upstream for AI inferencing.

The contribution specifically targets the real-time communication for Conversational XR scenario, providing a complete experimental framework based on InterDigital's XR platform for traffic measurement associated with QoE metrics.

Main Technical Contributions

Experimental Approach (Clause 6.1.1.1)

The proposal introduces a standalone custom client/server XR platform for simplified emulation of real-time communication for conversational XR with the following key features:

Platform Capabilities

• Control of transmission/reception of XR data at both client and server sides in uplink and downlink
• Measurement of QoE metrics including:
• Pose-to-render-to-photon latency
• Roundtrip-interaction delay
• Measurement of uplink/downlink traffic characteristics

Architecture Design

• Client side: Generation and transmission of user's XR data (pose information, actions, real environment data)
• Server side: Management of XR scene, generation and transmission of resulting XR scene data (scene state for local rendering, media)

Measurement Methodology

• Per-packet metadata insertion in both uplink and downlink flows for latency QoE metrics measurement
• Packet marking with timing information to enable end-to-end latency measurement
• Correlation between uplink packets (e.g., pose information) and corresponding downlink packets (e.g., scene state)

Testing Phases

1. Baseline measurements on wired network (ideal network conditions)
2. 5G network emulation using test channels or emulated 5G network with:
3. Deterministic, controllable, and reproducible conditions
4. Real-time typical radio and capacity conditions
5. Impact assessment of delays and packet losses

5G Network Conditions

The approach considers multiple network scenarios: - Nominal conditions - Cell-edge conditions - System load in multi-UE scenarios

Test Setup (Clause 6.1.1.2)

The proposal defines a comprehensive test setup with detailed observation points:

Client-Side Custom XR Application

Functions include: - Generation and transmission of user's data with metadata for QoE metrics measurement to remote Scene Manager - Reception of resulting XR scene data with metadata - Rendering and display of XR scene

Server-Side Custom XR Application

Functions include: - Reception of user's data with metadata - XR scene updates and QoE metrics collection - Transmission of resulting XR scene data with metadata

Observation Points Architecture

Five observation points are defined for comprehensive measurement:

Network-Level Observation Points: 1. Network_UE_OP (UE side): - 5G network emulator ingress for uplink XR data - 5G network emulator egress for downlink XR data

1. Network_UPF_OP (UPF side):
2. 5G network emulator egress for uplink XR data
3. 5G network emulator ingress for downlink XR data

Application-Level Observation Points: 3. Client_Application_OP_1 (client side): - Measurement of round-trip QoE metrics (pose-to-render-to-photon, roundtrip-interaction delay)

1. Client_Application_OP_2 (client side):

2. Measurement of inter-flow time synchronization QoE metrics in downlink

3. Server_Application_OP (server side):

4. Measurement of inter-flow time synchronization QoE metrics in uplink

Measurement Tools and Techniques

• Traffic characteristics measurement: IP packet analysis using open-source network protocol analyzer (e.g., Wireshark)
• Dual-point measurement: Both ingress and egress of 5G network emulator to:
• Highlight network performance impact on traffic characteristics
• Identify application/transport level impacts from network performance through variability analysis

Technical Significance

This contribution provides a complete experimental framework for evaluating conversational XR real-time communication, enabling: - Systematic measurement of dynamic traffic characteristics - Assessment of QoS feature benefits in 5G systems - Correlation between network conditions and QoE metrics - Reproducible testing methodology for standardization purposes