Skip to main content
Testbeds

AI Traffic Characterization

At a glance
  • What it is: the testbed SA4 uses to measure the traffic of AI media services for its 6G media study, TR 26.870 (FS_6G_MED, Release 20).
  • How: AI service scenarios run over emulated network conditions; the traffic is captured and turned into metrics.
  • Where in the TR: annex B describes the testbed, annex C the method and results, annex D the scenarios, network profiles, metrics and real-time video results, and annex E the network emulator (a temporary annex). See How the testbed maps to the annexes.
  • Status: TR 26.870 is a draft (V0.6.1, September 2026); its results are initial.

What it does​

The 6G media study records the traffic characteristics of AI media services from this testbed's measurements, as the basis for its statements on AI media services. The study obtains the characteristics with the framework of its annexes B, C and D (clause 6.3.3.1); the results are the record of the AI traffic characteristics it collects. What the study covers, and where AI traffic fits in it: Technical Analysis.

The Scenario Orchestrator drives the scenarios (chat, agents, image, video and audio) and MCP tool servers, through the LLM client layer, L7 capture with mitmproxy and the network emulator with tc/netem profiles, to the LLM and tool APIs on the Internet. Capture and analysis: L3/L4 capture with tcpdump and L7 capture with mitmproxy feed the TrafficLogger, which feeds the MetricsCalculator (TTFT, TTLT, UL/DL, loop factor) and the visualization.
Testbed architecture, from the 6G-Testbed repository.

Contributors & How to Contribute​

QualcommQualcommHuaweiHuaweiBecome a Contributor

What's Built​

Metrics collection
Metrics collection pipeline, with Reference ToolsBased on TR 26.870 V0.6.1, figure D.4.2-1

Annex B.2 of TR 26.870 describes the testbed architecture as a list and has no architecture figure; this pipeline (clause D.4.2) is its only figure of the software, and it names classes and functions of the repository's code. The diagram above shows the whole testbed. Click the figure to enlarge it.

Test design

Scenarios, network profiles and their call flows.

Metrics

Traffic, response time, service and task metrics.

Results

What the measurements show so far.

Repositories & Releases​

Early Access Early Access: available to members first

6G AI Traffic Characterization Testbed6G-Testbed
no release yet
View repositoryIssuesPull RequestsReleases
Network Emulator Dashboardnetemu-dashboard
no release yet
View repositoryIssuesPull RequestsReleases

Standards implemented​

3GPPTR 26.870

3GPP TR 26.870 (the 6G media study, FS_6G_MED, Release 20), a draft (V0.6.1, September 2026). Its annexes B to E describe the testbed, the method, the scenarios and the network emulator.

See 6G AI Traffic Characterization: Specification list for version history and 5G-MAG's own standards contributions.

Developer Updates​

Last Synced: 2026-10-26 01:55 UTC

Early Access Early Access: available to members first

6G-Testbed
4,505 views
1,259 clones
★ 15⎇ 12
Active
netemu-dashboard
7 views
53 clones
★ 0⎇ 0
Active

Open Pull Requests​

6G-Testbeddocs: apply the 5G-MAG README formatReady#11 · 4w ago
6G-Testbednetemu measurement package and agentic scenariosReady#10 · 9w ago