6G found its band. Slicing found its customers.

Days before the Madrid plenary decides how 6G attaches to the world, the specs already tell a story: upper 6 GHz written into the radio tables, network slicing on store shelves in Britain, AI schedulers running on live commercial networks, phones dialed from space, and a plan to survive quantum computers. Here is what was actually written down.

One note on timing: 3GPP meets in Madrid on September 14 to 17 (RAN#113) to settle the open 6G questions, including the base-station split deferred from Singapore and the choice of 6G migration path. This issue covers what has already landed in published specifications through September 11. The decisions come next.

1. The 6G capacity band is no longer a proposal. It is a table.

The loudest spectrum fight of the year sharpened this month. The GSMA declared the device ecosystem ready for the upper 6 GHz range and urged regulators to open 6.425 to 7.125 GHz for mobile (Sep 7). Optus and Nokia trialed the band in Sydney at 3.5 Gbps with macro coverage comparable to 3.5 GHz. Spain's MásOrange began its own 6 GHz tests with Ericsson. And Britain's Ofcom countered that it wants only a sliver (7125 to 7250 MHz) for mobile at the WRC-27 conference, with the rest of Europe split.

Beneath the lobbying, the specification work is done. The first Release 20 edition of the base-station radio spec (TS 38.104 v20.0.0) extends the first frequency range to cover the new airwaves and defines the operating bands explicitly:

“Frequency range designation Corresponding frequency range FR1 410 MHz – 7125 MHz…” (TS 38.104 §5.1, Table 5.1-1)

“n102 5925 – 6425 5925 – 6425 TDD… n104 6425 – 7125 6425 – 7125 TDD” (TS 38.104 §5.2, Table 5.2-1)

This is not a placeholder row. Band n104 gets its own unwanted-emissions table, its own expected-radiated-power limits, and its own channel bandwidth rows up to 100 MHz. The band that regulators are still arguing about already has a full radio specification.

The companion detail matters just as much. A new Release 20 specification (TS 38.307 v20.0.0) defines:

“Requirements on User Equipments (UEs) supporting a release-independent frequency band” (TS 38.307 (title))

Translation: a phone can support a newly defined band without waiting for its release of the standard to catch up. Bands ship when regulators clear them, not when the release train arrives.

Why it matters: the upper 6 GHz debate is now about permission, not feasibility. The radios are specified. What is missing is agreement on who transmits.

2. Slicing went from slideware to store shelves

For years, network slicing was the feature operators demoed and nobody could buy. That changed in August. EE launched Fast Lane, a paid consumer tier on its standalone 5G network that gives buyers a dedicated slice when cells congest (Light Reading, Aug 20). VodafoneThree answered with tiered enterprise and critical-service slices, including a 15 Mbps performance guarantee, and declared open slicing war on BT. Meanwhile India's regulator moved to require per-slice tariffs and measured-speed disclosure, with a debate over capping any single slice at 80 percent of cell resources.

The management specifications behind this are advancing in step. The provisioning spec (TS 28.531 v20.1.0, whose keywords are literally “5G, Network Slicing, Slice, Provisioning”) describes the machinery that turns a commercial order into a running slice:

“The provisioning of network slicing includes the four phases which are preparation, commissioning, operation and decommissioning…” (TS 28.531 §4.1)

The same release cycle updated the Network Resource Model (TS 28.541 v20.3.0) that defines slice objects on management interfaces, and the performance measurements (TS 28.552 v20.3.0) that cover 5G networks including network slicing. In other words: the slice you sell, the object that represents it, and the counters that prove it performed are all specified together. That trio is what makes a guarantee like 15 Mbps billable instead of aspirational.

Why it matters: slicing revenue was always a 6G promise. Britain just made it a 5G product, and the management plane is keeping pace.

3. AI touched a live tower and the numbers held up

June's AI story was in the core network. August's is on the mast. SoftBank and Ericsson switched on what they describe as the first AI-native link-adaptation scheduler on a commercial 5G network in Japan, reporting up to 25 percent better spectral efficiency, 50 percent higher downlink throughput, and 10 percent average gains (Aug 20). T-Mobile and Ericsson followed with a large-scale 5G-Advanced trial of an AI-native scheduler showing 15 percent downlink and 10 percent efficiency gains. A sobering counterpoint came from a study of Nokia and Nvidia's AI-RAN pitch: the GPUs pencil out only at the right sites, with field trials expected late this year and commercial deployments late next.

The specification side of this is management, not magic. The AI/ML management specification (TS 28.105 v19.6.0) defines exactly what it sounds like:

“The present document specifies the Artificial Intelligence / Machine Learning (AI/ML) management capabilities and services for 5GS where AI/ML is used, including management and orchestration…and NG-RAN…” (TS 28.105 §1)

Its information model covers the full lifecycle operators need before trusting a model with a live cell: model loading, model updates, the inference function itself, and even inference emulation for testing. A scheduler that learns is only deployable if the network can version it, monitor it, and roll it back. That scaffolding is what this spec builds.

Why it matters: the debate has moved from whether learning schedulers work to where they pay. The specs answer the next question, which is how to operate them.

Deep dives: the talk vs. what actually landed

What the community is saying, then what the specification text actually says.

Phones calling from space stopped being a demo

The buzz: direct-to-device had its busiest fortnight. AST SpaceMobile won FCC permission for 30-day direct-to-device tests in the 800 MHz band with a hundred off-the-shelf phones, then filed to extend the window. ST Engineering iDirect demonstrated a multi-waveform NR-NTN modem at ESA. The satellite association MSSA released a reference architecture for regenerative payloads aligned to current 3GPP releases, and the GSMA pushed an interoperability drive for the sector’s next phase. What landed: 6G’s first requirements study (TR 38.914 v20.0.0) treats space as a first-class deployment, with its own scenario clause and a design principle that terrestrial and satellite share one radio:

“Aim at a harmonized 6G Radio design for TN and NTN, including their integration” (TR 38.914 §5.5 (see also §4.10 Non-Terrestrial Network))

And the current generation is still being hardened. The first Release 20 edition of the radio-resource-management requirements (TS 38.133 v20.0.0) specifies exactly how a phone measures a satellite cell before reselecting to it:

“This clause defines the following conditions for NR intra-frequency measurements performed based on SSBs for cell re-selection: SSB_RP and SSB Ês/Iot, applicable for a corresponding operating band for satellite access.” (TS 38.133 Annex B.1.6 (see also B.1.7))

The takeaway: June’s satellite story was voice squeezed through sensor links. This month’s is broadband phones moving between terrestrial and satellite cells, with 6G designed around that handover from the start.

Security started planning for the quantum era

The buzz: the security group met in Prague in late August (SA3#129) with post-quantum cryptography on the agenda: hybrid key exchange in TLS 1.3, phasing out TLS 1.2, and certificate profiles, plus a liaison to the IETF pressing for the underlying standards. What landed: the study on the transition (TR 33.703 v20.0.1) frames the problem plainly:

“Studies the impact of using hybrid and standalone PQC algorithms in 3GPP procedures” and the “Impact to 3GPP procedures due to larger length of PQC key, signature, and message compared to the length of those in traditional cryptography.” (TR 33.703 §1 (see also §5))

The study runs to more than twenty candidate solutions for concealing the subscriber identity alone, including hybrid designs that pair today’s elliptic-curve encryption with the new lattice-based ML-KEM algorithm. Quantum-safe subscriber privacy is being worked as an engineering problem with named algorithms, not a research topic.

The takeaway: cryptography transitions take a decade, so they start early. 3GPP has started.

Also worth knowing

• Battery-less IoT grew an application layer. A new Phase 2 study (TR 23.700-26 v20.1.0) works out how the application plane handles ambient IoT: enriching device data exposed from the core network, serving value-added device information to consumers, and monitoring devices to minimize service interruption (built on SA1 requirements in TS 22.369 and the SA2 architecture in TS 23.369). Sensors that live off harvested energy still need software that notices when they go dark.
• Multicast streaming gets a second act. A first Release 20 report (TR 26.802 v20.0.0) evaluates multicast for 5G media streaming, naming the scenarios outright: “transparent multicast delivery, multicast linear IPTV delivery, hybrid unicast/multicast (e.g. MooD or service continuity), and multicast Adaptive Bit Rate (ABR) for Over the Top (OTT) live streaming” (§1). It also picks up the repair problem Release 18 left open: “only a post-session repair procedure is defined up to and including Release 18; in-session object repair procedures are declared as being for further study” (§5.9.1). Live sport at stadium scale is the use case that makes multicast math work.
• The network starts taking intent. A first Release 20 specification (TS 28.312 v20.0.1) “specifies concept, use cases, requirements and solutions for the intent driven management for service or network management” (§1). Operators state what they want; the management plane figures out how. Watch this one alongside the AI management story above.
• Public safety gets its receipts. A new study (TR 23.700-39 v20.0.0) “studies solutions to satisfy the requirements for logging and recording of mission critical services” across the common architecture (TS 23.280), push-to-talk (TS 23.379), video (TS 23.281), and data (TS 23.282) (§1). The mission-critical family itself moved too, with the common architecture and the video spec both advancing to v20.4.0. Recording every call is an accountability feature and a storage problem in equal measure.

How we make this: we track the 3GPP plenary and working-group outputs and the broader telco press, then cross-check the conversation against the actual specification text, so you get both what is being discussed and what concretely made it into the documents. This issue drew on the September 11 specification crawl and the published Release 20 texts. Madrid decisions land next month.

3GPP Scout is built by Carrot Labs. Not affiliated with or endorsed by 3GPP. Questions: [email protected]