6G: What Is Actually Being Standardised, and What Is Still Research

Writing about 6G has a structural problem: the specification does not exist yet, so almost anything can be asserted about it without being wrong. The result is a literature where a research paper's speculative upper bound and a committee's agreed target read identically.

This section tries to keep those apart. Where something has been formally agreed, it says so and names the document. Where something is a study item, it says that. Where it is an open research question, it says that too — and those are often the more interesting parts.

What has actually been agreed

The framework document is ITU-R Recommendation M.2160, approved in November 2023, which defines IMT-2030. It is the closest thing to an authoritative statement of what 6G is for, and it is worth understanding what kind of document it is: a set of usage scenarios and capability targets, not a design.

It extends the familiar three 5G scenarios into six, and the additions are the interesting part — integrated sensing, integrated AI and communication, and ubiquitous connectivity join the enhanced versions of mobile broadband, massive connectivity and low-latency communication. Those additions are what make 6G a different system rather than a faster one.

What Is 6G? works through the targets, the timeline and which of the headline numbers survive contact with the framework.

The timeline

3GPP's study work on 6G sits in Release 20, with normative specification expected in Release 21. Commercial networks are generally projected for around 2030. The dates are driven by the ITU-R process, which is why they have been unusually stable across a decade of otherwise enthusiastic forecasting.

Spectrum: sub-THz and its physics

The bandwidth argument for going higher in frequency is simple and correct — there is more of it up there. Everything else about it is difficult.

Above roughly 100 GHz, molecular absorption becomes a real term in the link budget rather than a footnote: specific gas species absorb at specific frequencies, so the usable spectrum is a set of windows between absorption peaks rather than a continuous band. Free-space loss is worse, penetration through ordinary materials is close to nonexistent, and the analog-to-digital converters needed for gigahertz of bandwidth are power-hungry enough to shape the entire receiver architecture.

THz communication goes through the propagation numbers and the converter power problem, and lands on where these bands realistically fit: short, largely static links with clear line of sight.

Sensing: the radio as an instrument

The idea behind integrated sensing and communication is that a radio network already transmits known waveforms from known positions and measures what comes back. That is what a radar does. If the same infrastructure can communicate and sense, the network gains a capability at close to the marginal cost of the processing.

The trade-off is that the two uses want different things from the signal. Communication wants the waveform shaped for spectral efficiency and the resources given to users; sensing wants bandwidth for range resolution, and resources spent listening rather than talking. How much sensing you can extract before communication capacity suffers is the engineering question, and it does not have a single answer — it depends on how much resolution the application actually needs.

Intelligence as an architectural assumption

Two related articles cover the AI thread, and they are deliberately separate because the ideas are separate.

AI-native networks is about design posture: the difference between a system that accommodates learned components and one that assumes them. The distinction matters because most of what limits machine learning in 5G today is not model quality but access — to data, to the right interfaces, at the right timescale.

Digital twins in telecom covers the proving-ground problem. If the network is going to be operated partly by learned policies, those policies need somewhere to be evaluated that is not the live network. A twin is that somewhere, and its usefulness is bounded entirely by the fidelity of its models and the quality of the data feeding it.

Semantic communication: the genuinely speculative one

Semantic communications proposes something that sounds like a category error at first: transmit the meaning rather than the bits. The concrete version is less exotic — if the transmitter and receiver share a model of what the message is for, the transmitter can send only what the receiver cannot already infer.

It is worth taking seriously and worth being honest about. The theory is interesting, the demonstrations are real, and the interoperability problem is unsolved. A system where the meaning of a transmission depends on a model held at both ends does not obviously survive contact with multi-vendor networks and devices that persist for a decade.

How to read this section

Nothing here requires reading in order. If you want the framework first, start with what 6G is; if you want the physics, start with THz. The AI articles pair naturally, and semantic communication reads best after them.

The 5G material is the useful background for all of it. 6G is a successor system, and most of its problems are recognisable as 5G problems that ran out of room.