← Archive · 6G Tracker
 

Daily Dose of 6G

Edition 03 20 August 2026 Series: Requirements

What 6G is actually being asked to do

TR 22.870 is 590 pages long, took eighteen months, and is the root of every other 6G study in Release 20. Six use case families, four cross-cutting dimensions — and one of them is not what you would have predicted five years ago.

Transcribed from 3GPP’s published Release 21 timeline (© 3GPP 2026). Rel-20 is the 6G study phase; Rel-21 is the first 6G normative release.

The document that starts everything

TR 22.870, "Study on 6G Use Cases and Service Requirements", is the first completed 6G study in 3GPP and the root of the dependency tree. Work item FS_6G_REQ, WID SP-241391, running 4 September 2024 to 3 March 2026, rapporteur Xiaonan Shi (China Mobile), under SA1 chair Vasil Aleksiev (Deutsche Telekom) and vice-chair Feifei Lou (Nokia). It was approved at TSG#111 in Fukuoka in March 2026 and runs past 590 pages.

Everything downstream inherits from it. SA2's architecture study, the RAN scenarios work, the security study — each opens by referencing what SA1 said 6G is for. Its conclusions are now flowing into normative Stage 1 as TS 22.270.

The six use case families

FamilyWhat it covers
AI-Native 6GNetwork AI agents, third-party AI agents, model training and inferencing as network functions, AI data and traffic management
ISACIntegrated sensing and communication — the network as an instrument as well as a pipe
Ubiquitous connectivityTerrestrial and non-terrestrial as one system; the "anywhere becomes everywhere" framing
Immersive communicationXR, holographic and multi-sensory services with hard latency and synchronisation constraints
Massive communicationIoT at a scale and energy profile 5G does not reach
Industry & vertical applicationsDeterministic industrial automation, and the sector-specific requirements that follow

Four dimensions that cut across all six

  • Resilience — service continuity under adverse conditions, including disaster scenarios
  • Energy efficiency — treated as a first-class requirement, not a deployment optimisation
  • Data management — how the network collects, stores and exposes data, with privacy constraints attached
  • Security and privacy — stated as design principles at requirements stage rather than bolted on by SA3 later

The striking thing is not any single family — most were visible in 6G research years ago. It is that AI appears as a use case family in its own right, with the network itself hosting agents, rather than as a technique applied to the other five.

Why "AI-native" is a requirements statement, not a slogan

SA1 wrote requirements for the network to support AI agents as first-class entities — network-operated agents and third-party agents both — plus model training and inferencing, and the data and traffic management those imply. That is a very different claim from "we will use ML for scheduling".

It creates obligations three groups down. SA2 must decide what an agent is architecturally (four of its eight work tasks sit under "beyond connectivity", including AI and computing coordination across UE, core and application server). CT3 must decide what it speaks — TR 29.832, rapporteur Apostolos Papageorgiou (Nokia Germany), is weighing IETF agentic protocols against 3GPP-specific ones. SA3 must decide how to secure it. All three obligations trace back to text in TR 22.870.

Reading it critically

A 590-page requirements study is not a design, and it is worth holding lightly. Use cases in a 3GPP Stage 1 TR are contributed by member companies and survive by consensus, which means the document reflects what its contributors wanted recorded as much as what the market will bear. Some families — immersive communication in particular — carried similar promises into Rel-15 and Rel-17 without the deployments following.

The test is which requirements survive into TS 22.270, the normative Stage 1 specification opened in March 2026. A use case that does not produce a normative requirement is a use case the industry declined to commit to. That transition is the subject of Edition 05.

One figure we are not asserting

The ATIS Release 20 webinar characterised the use case distribution as roughly 28% AI and 28% system/operational aspects. We have not verified that against the TR itself, and vendor-deck figures on this release have already proven unreliable once — the Rel-21 Stage-2 dates in Edition 02 came from the same source and needed correcting. Treat it as indicative.