What the standard has agreed

Of sixteen documents, five bind and two describe a product

The distinction that matters when reading standards is between a technical specification, which is a requirement that equipment has to meet, and a technical report, which is a study that may never turn into one. Sort the sensing documents of the five bodies that way and the result is long on study and short on specification. The short part is about drones (Exhibit 1).

Five rows bind, four of them cellular, and only two of those describe a product. TS 22.137 says what a 5G system shall deliver, and part 1 reads it band by band, but a service requirement is not a design. The design is in TS 23.137 and in the work item, and both confine themselves to a base station detecting and tracking aerial objects. Everything else in the table is study, including the channel model and the 6G requirements, however much work went into them. Sensing was first agreed in 3GPP in December 2023. The date that matters for anyone making a plan is March 2027, when RAN#115 is due to approve the first product specifications.

Release 20 specifies one product: a base station detecting drones with its own echo

The work item that turns the study into specifications is a short document, and what it leaves out defines the product as much as what it puts in.

RP-261566 was approved at RAN#112 in Singapore, which ran from 8 to 11 June 2026. The moderator was the RAN vice chair from T-Mobile USA, and the lead working group is RAN3, the one that writes interfaces, with a rapporteur from China Telecom. The objective is plain enough: “specify gNB-based mono-static sensing for UAV sensing target use cases … There will be no UE impacts”. It allows that intra-gNB bistatic sensing “may be considered”. Forty-six companies put their names to it, among them AT&T, T-Mobile USA, Verizon and FirstNet. At RAN#113 in Madrid three months later, 14 to 17 September, the revision RP-261901 named four new specifications for the new interface and suggested the numbers TS 38.480 to 38.483. They cover general aspects, layer 1, signalling transport and a new application protocol, and they go to RAN#114 for information and to RAN#115 for approval.

The study behind the work item, RP-253246, had already drawn four lines around it.

• Waveform. The study used the “existing DL NR waveform and DL NR reference signals”. There is no new sensing signal.
• Band. FR1 was given priority.
• Coordination. “No inter-gNB coordination will be studied.” Base stations do not cooperate with each other at the radio layer.
• Device. “There will be no UE impacts.” The phone is untouched.

Put those together and the first cellular sensing product is a software and interface feature running on radios that are already in the field. That is what makes it commercially near, and also what makes it technically modest.

The core network side is TS 23.137, approved at SA#113 in September 2026. Its scope is “only focusing on aerial object (e.g. drone) detection and tracking use case”, and “only gNB-based sensing is supported”. It adds a Sensing Function, SenF, split into a control part and a processing part with “no standardized interface between” them. The SenF talks to the base station, to the network exposure function and to a trusted application function (Exhibit 2). That application function is the only consumer there is. It asks for “Object Detection” or “Object Tracking” and gets back one-time, periodic or event-triggered reports, such as when the “number of Aerial objects exceeded a certain threshold”. Several editor’s notes are still open, so some of the detail may yet move.

Read as a product rather than as an architecture, the diagram names the customer. The output goes to an authorised application through an exposure interface, and never to a subscriber’s phone. So the first sensing product is an interface sold to an organisation that watches airspace: an airport, a venue, a security agency. Part 5 takes up who those buyers are and what they might do with the answer.

The open decision is whether the base station reports points or objects

TR 38.765 defined four levels at which a base station can hand its sensing measurements to the core. The choice between the last two is the only decision of substance still open, and it is less technical than it looks (Exhibit 3).

The study ruled out Level A “Due to its higher transport capacity requirement”, and it is easy to see why: a single raw report at the study’s largest assumptions is 116 Gb, which is more than a 100 Gb/s link moves in a second. It agreed Level C and made Level D conditional on “the resolution of potential privacy and security issues”. RAN put the question to SA3 in June. SA3’s reply, S3-263445 from its August meeting in Prague, said it had “considered the sensing measurements/data without any distinctions or dependencies on the level C/level D”. It pointed back to its own report, and the question went back to RAN unanswered.

• Level C. The base station extracts points. The sensing function in the core clusters them, fuses the sites, detects and tracks. The core vendor does the radar processing and can, in principle, buy points from any radio vendor.
• Level D. The base station detects, tracks and fuses for its own sites, and the core receives finished objects. The radio vendor does the processing and owns the answer.

The companies have split along exactly that line (Exhibit 4). China Mobile, with twelve co-signatories, and Ericsson in a paper of its own, want Level D; Huawei asked for Level C alone. Neither the SA3 reply nor RAN#113 settled it. The next decision point is RAN#114 in Boston, 7 to 10 December 2026, when the four specifications arrive for information.

What the Level C or Level D decision allocates is the radar itself, between vendors. Level C puts detection and tracking in the core’s sensing function, which makes a multi-vendor network possible on paper. Level D keeps them in the base station, which is how the operators that have deployed sensing run it today and is the arrangement Ericsson and Nokia support. Privacy is the stated reason for the dispute, and SA3 has declined to settle it.

SA3 decided not to specify a privacy solution

SA3’s study, TR 33.777, was approved at SA#113 in September 2026. On privacy, it records a decision not to decide.

The service requirements had already pointed the question outward. TS 22.137 requires sensing KPIs “for both situations where consent can be obtained, and where it cannot”, and allows that “Subject to regulation and user consent” the network “may be able to link sensing results with 3GPP subscriber identity”. Put the two documents together and the privacy of the people in the beam is left to regulators and to each operator’s own policy. The first regulator to give an answer was Germany’s data-protection conference in June 2026, and we come to it in part 5.

The 6G requirements are set, and Europe has opened the first regulatory file

For 6G, 3GPP has agreed the requirements and the starting points, while the waveform, the reference signals and the sensing modes remain open. Outside 3GPP, CEPT has become the first regulator to treat a network’s sensing transmissions as something that needs conditions of its own.

TR 38.914, approved at RAN#112, sets the 6G requirement. The radio and architecture “shall at least support use cases of detection and/or tracking of passive objects, at least including UAVs, human, vehicles and AGVs [and] communication assistance”. The minimum performance figures are in part 1. RAN1 has agreed to start from the 6G communication frame structure, cyclic prefix and subcarrier spacing for sensing, and lists the sensing frame structure and numerologies as open. TS 22.137 gained a 6G clause at SA#113 under which the 5G functional requirements “apply analogously to the 6G system”, with no 6G performance table. Companies are already staking out positions. T-Mobile USA and Nokia asked for bistatic and multistatic sensing and for service layers that “expose APIs for third-party applications” (RP-260887, noted). Ericsson, Jio, T-Mobile USA, China Telecom and others wrote that “5G ISAC functionality can be considered as a subset of 6G functionality” (RP-261416).

Outside 3GPP the paper is thinner. Rec. ITU-R M.2160 names integrated sensing as one of six IMT-2030 usage scenarios and says sensing “could be measured in terms of accuracy, resolution, detection rate, false alarm rate, etc.”, without putting a number on any of them. The draft requirements go to Study Group 5 in December 2026, and candidate technologies are accepted from February 2027. ETSI’s industry group has published four informative reports, of which the first lists 18 use cases and the fourth 19 key issues. The only sensing standard in force anywhere is IEEE 802.11bf, and it is for Wi-Fi.

The first regulator to act is European. At its 70th plenary in June 2026, CEPT’s Electronic Communications Committee agreed to create a work item on harmonised technical and operational conditions for sensing in 3.4 to 3.8 GHz. It is answering a European Commission mandate for “the sensing use case of detecting and tracking flying objects (in particular aerial drones or balloons)”, limited to base stations and “with the objective to facilitate the identification of such unlawfully flying objects”. Its liaison of 11 September 2026, logged at RAN#113 as RP-262217, asks 3GPP for four parameters of a sensing base station, for both monostatic and base-station-to-base-station bistatic modes. They are the power spectral density of the sensing signal, the channel bandwidth, the activity factor and a block-edge mask. The dates that decide what happens next run from December 2026 to December 2028 (Exhibit 5).

3GPP and the European regulator have arrived at the same first product from different directions: base stations detecting drones in mid-band spectrum. Everything wider, which means people, vehicles, phones as sensors and bistatic networks, waits for Release 21 and its freeze in December 2028.

The terms, briefly

• Normative and study. A technical specification (TS) or an approved work item binds products; a technical report (TR) binds nothing.
• gNB. The 5G base station. “gNB-based mono-static sensing” means the base station transmits and receives its own echo.
• SenF. The Sensing Function in the 5G core that TS 23.137 adds; it controls sensing, processes reports and exposes results.
• Measurement level. How much processing the base station does before it reports: raw samples (A), profiles (B), points (C) or objects (D).
• AF and NEF. The application function is the authorised outside consumer; the network exposure function is the gateway it reaches the core through.