Daily Dose of 6G

How do you test a receiver that learns?

The second half of TR 38.760-4 is the part of 6G that decides whether a product may be sold: radio resource management, demodulation, and over-the-air testability — now with machine learning inside the thing being measured. This edition closes the radio arc.

Transcribed from 3GPP’s published Release 21 timeline (© 3GPP 2026). Editions 10–12 sat on the Q4 2028 core row. This edition does not: demodulation, RRM and over-the-air testing are RAN4 performance work, one plenary behind core. The only performance row on the chart is the Rel-20 one highlighted here — the Rel-21 equivalent, the deadline this study’s output lands on, is absent.

Where the second half of RAN4 lives

Edition 12 counted RAN4’s 6G draft folders from the outside. The full listing shows the shape underneath: RAN4#120 ran 79 drafting folders, grouped into three named sessions — [120][100] Main Session, [120][200] RRM_Demod_OTA_Session and [120][300] BDaT_Session. The 19 that begin with 6G are spread across all three.

That split is the edition in one table. 6G RRM and 6G AI sit in the main session, where a group decides what a thing is. Demodulation and testability sit in the performance session, where it decides how well it must work before anyone may ship it. One study, two deadlines.

A week of 6G AI, in folder names

The drafting inbox for [120][107-A] 6G AI (part I) holds six sub-folders, numbered in the order a meeting week runs: 0_Pre-meeting Study, 1_Ad-Hoc_TuesdayCoffeeBreak, 2_AI-DPoD_result_collection, 3_Ad-Hoc_WednesdayAH, 4_Ad-Hoc_Thursday, 5_Way Forward. Three ad-hocs on one agenda item in one week, one of them booked into a coffee break. No folder name says what was decided — but a group that books three extra sessions did not agree on Monday.

Inside the result-collection folder are seventeen files, sixteen of them successive versions of one spreadsheet, “Simulation summary for AI based DPoD”, handed company to company: OPPO, Samsung, Ericsson, ZTE, Nokia and vivo all appear by name in the chain, plus the initials delegates use for several more. The seventeenth collects link-level simulation results for the same feature. This is what a 3GPP evaluation campaign physically looks like.

The middle row deserves a pause. In August 2026 the demodulation item was circulating a document called TR Demodulation Section Structure. Not results — headings. That fits a technical report sitting at version 0.0.1, and confirms Edition 12’s reading: RAN4’s document has not started because its inputs do not exist.

What DPoD is, and how sure we are

Digital pre-distortion is a transmitter correcting for its own power amplifier before the signal leaves. Andreas Oeldemann of Rohde & Schwarz and Dani Korpi of Nokia Bell Labs, in Microwave Journal, February 2026, describe the inversion: “Rather than the transmitter compensating for its own distortions, the receiver, which is typically a base station with more computational resources, performs the compensation.” Move the correction to the base station and a handset can run its amplifier closer to saturation and spend less battery per bit. Their subject is the application RAN4 is now collecting simulations on: energy-efficient 6G uplink.

The model 3GPP will not specify

Zahid Ghadialy, on free6gtraining in August 2026, restates the constraint RAN4’s AI work sits under: “3GPP has previously made clear that it does not currently intend to specify particular AI/ML models.” If the model is not in the specification, a conformance test cannot check that the right model was implemented. It can only check what the implementation does.

Nokia’s standardization team — Dimitri Gold, Fahad Syed Muhammad and István Z. Kovács — put the problem plainly two years ago: “Current testing methods are designed with deterministic analytical algorithms in focus, yet they are inadequate for addressing the adaptive and stochastic nature of AI.” Their slogan for it, “reliability starts with testability”, is the reason RAN4#120 has an agenda item called 6G_testability_OTA before it has a single 6G band.

RAN4 has been rehearsing this on 5G-Advanced. Item [120][214] NR_AIML_air_Ph2 — a Release 19 item, not 6G — carries a spreadsheet counting model parameters and FLOPs, a draft on “inter-vendor training collaboration and interoperability of CSF compression”, and a results sheet passed through fourteen versions and a dozen named companies. Two vendors’ neural networks must interoperate across an interface neither specified. That is the machinery 6G inherits.

Madrid, in eleven days — and one loose end tied

The September plenary round finally has a number, dates and a city: RAN#113, Madrid, 14–17 September 2026, from 3GPP’s plenary calendar. Two things were parked on it — RAN2’s deferred 6G dual-connectivity decision, and the operators’ request for preliminary multi-RAT spectrum sharing evaluation results by September. Both are RAN4-shaped questions arriving before RAN4’s document has a real version. The same calendar puts Rel-21 package approval at RAN#115, Rotterdam, 15–19 March 2027.

One smaller gap closed too. TR 38.960 — the fifth deliverable of FS_6G_Radio, the only one owned by the RAN plenary rather than a working group — is titled “Study on 6G Radio”, rapporteur Shinya Kumagai, NTT DOCOMO, the same name as the RAN1 part. Created 1 October 2025; eleven months later it still has no uploaded version. The umbrella meant to sit over all four parts is emptier than any of them.