What a network can see

What a radar can separate comes down to three things

A sensing network has no connection to the object it is looking at. It has to work out three things from an echo: how far away the object is, how fast it is moving towards or away from the site, and in which direction it lies. For each there is a resolution, meaning the smallest difference the system can tell apart, and each resolution is set by a different property of the signal (Exhibit 1).

• Range. Distance comes from the delay of the echo. To separate two objects at slightly different distances, the signal has to be short enough in time that their two echoes do not merge, and a short pulse is the same thing as a wide channel. The formula is the speed of light divided by twice the bandwidth. For a 100 MHz carrier that comes to 1.5 m.
• Speed. Speed comes from the frequency shift in the echo. The longer the radar watches, the smaller the shift it can see, so velocity resolution is the wavelength divided by twice the observation time. Patience, in other words, buys speed resolution.
• Direction. Direction comes from the antenna. An array can resolve an angle of roughly one wavelength divided by its own width, and two objects side by side are only separated if the gap between them is more than that angle multiplied by how far away they are. A bigger panel sees finer.

Two words get used interchangeably here and should not be. Accuracy is how close a measured position is to the true one. Resolution is whether two objects can be told apart at all. TS 22.137 defines them separately, as “closeness … to its true position” and as “the minimum difference in the measured magnitude … to be allowed to detect objects in different magnitude”, and the distinction matters for reading any sensing claim. 3GPP’s own drone evaluations in TR 38.765 ran on 100 MHz carriers, whose finest range cell is 1.5 m. Horizontal accuracy at the 90th percentile ranged from 0.26 m to 11.1 m, and what moved the number was mostly whether one site or several were combined. A strong echo can be placed quite precisely inside its cell. Whether two drones flying close together show up as two is a question of resolution, and processing does not change it.

Below 7.125 GHz, every 5G carrier stops at a metre and a half

The maximum channel width is written into the 3GPP specifications band by band, and the number is the same everywhere in FR1. From n41 at 2.5 GHz to n104 in the upper 6 GHz band, a single carrier is at most 100 MHz wide, and so 1.5 m is where it stops (Exhibit 2).

There is a way to go finer within FR1, at least in principle, which is to take two or more adjacent carriers in the same band and process them as a single wide signal. Nothing in the Release 20 work item provides for it. Some operators do hold the spectrum, though. In Auction 107 Verizon came away with 160 MHz of contiguous C-band in the New York area, from 3.70 to 3.86 GHz; AT&T took 80 MHz there and T-Mobile 40 MHz. Processed as one channel, Verizon’s block would resolve a little under a metre, 0.94 m by the formula.

The trials so far have run on much less than that. When Samsung and Verizon measured crowd density in Dallas in September 2026, they used a CBRS radio on a 40 MHz carrier, which gives a 3.75 m range cell. China Mobile’s field data, reported to 3GPP in RP-261227, show a range resolution of 7.5 m, which implies an effective bandwidth of only about 20 MHz. The contribution does not say which band was used.

The step to 400 MHz comes in two places. In millimetre wave it is already there, as an optional carrier width in Release 19. Around 7 GHz it is what 6G is being designed for. TR 38.914 says the 6G requirement for aggregated system bandwidth, which may span more than one carrier, is “at least 400 MHz”, and in June RAN1 told RAN#112 that the maximum channel bandwidth around 7 GHz “has been concluded”. The slide does not record the number, but Qualcomm puts it at up to 400 MHz on the network side and Ericsson gives 400 MHz as the top of the 6G system bandwidth. In the United States the 7 GHz range is still federal spectrum, and it has to be identified and auctioned before any of this happens; we cover that in part 2 of the 6G series. The 400 MHz sensing channel arrives when that spectrum does, and not before.

The standard asks for more than a 5G carrier can give

TS 22.137 is the document that says what a sensing service shall deliver. 3GPP approved it for Release 19 in December 2023, and its Table 6.2-1 sets out seven categories of performance. Four of them call for a range resolution that no carrier in FR1 can provide (Exhibit 3).

Read across the table and the pattern is plain. Category 1, detection and tracking to 10 m, needs only 15 MHz, and any 5G carrier clears it on width alone. Categories 2 and 3 want a metre, which takes 150 MHz. Category 4 wants half a metre, which takes 300 MHz. Category 7, hand gestures at 0.375 m, is exactly the resolution of a 400 MHz channel, which is to say the width that 6G and millimetre wave carry and FR1 does not (Exhibit 4).

The requirements the industry is building to are a good deal looser than the service table. Three documents set them. 3GPP’s own Release 20 evaluation, in Table 4.2-1 of TR 38.765, targets 10 m of horizontal and vertical position accuracy and 5 m/s of velocity accuracy at the 90th percentile, and it tolerates 5% missed detections and 5% false alarms. China Mobile’s “Commercial Requirements of 5G-A”, in RP-261227, ask for a range resolution of 10 m or better and a position accuracy within 20 m. The 6G minimums in TR 38.914, Table 5.1.17-1, ask for 5 m horizontally, 8 m vertically and 4 m/s in an urban macro deployment, tightening to 2 m and 2 m/s for an indoor factory. The draft IMT-2030 requirements that ITU-R Working Party 5D finished in February 2026 carry the same values, according to Samsung Research, and go to Study Group 5 for approval in December 2026.

So there are two sets of numbers in circulation. The commercial requirement and 3GPP’s evaluation target sit around 10 m, which any 5G carrier meets on bandwidth. The finer categories in the service requirement, a metre and below, need 150 to 400 MHz in one channel, and nothing in FR1 offers that. It arrives with 6G around 7 GHz, or with millimetre wave.

Speed is paid for in time, and direction in antenna size

Bandwidth settles the range cell and nothing else. Velocity resolution depends on how long the radar watches, and angular resolution on how big the antenna is. Both run into limits that have nothing to do with how wide the channel is.

Start with speed. At 3.5 GHz the wavelength is 8.57 cm, so to tell apart two objects whose speeds differ by 1 m/s the radar has to watch for 42.8 ms. At 7 GHz the time halves to 21.4 ms, and at 28 GHz it falls to 5.4 ms. The gesture category is where this bites. Category 7 asks for 0.3 m/s, and at 3.5 GHz that takes 142.8 ms of observation, nearly three times the category’s own latency limit of 5 to 50 ms. At 28 GHz the same measurement takes 17.8 ms and at 60 GHz 8.3 ms. That is why we read the gesture category as a millimetre-wave case rather than a macro one (Exhibit 5).

Observation time has a second edge to it. The fastest speed a radar can measure without ambiguity is λ/4T_s, where T_s is the interval between sensing symbols. Send one symbol every 2.5 ms at 3.5 GHz and anything faster than 8.6 m/s folds back on itself; send one every 0.25 ms and you can measure up to 85.7 m/s. The drones in 3GPP’s evaluations fly at up to 180 km/h, which is 50 m/s. So how often the sensing signal goes out, which is also what decides how much airtime sensing consumes (the subject of part 3), decides how fast a thing it can track. A vivo prototype that reused a reference signal transmitted every 20 ms at 3.6 GHz added no overhead at all. By the same formula it could not have measured anything moving faster than about 1 m/s. Enough for breathing. Not enough for a drone.

Angle is where a macro site is at its coarsest. The baseline array in TR 38.765 has eight columns at half-wavelength spacing, which at 4 GHz makes a panel about 30 cm across. That resolves about 14.3 degrees. At 200 m two objects need to be 50 m apart to be separated; at 500 m, 125 m. The optional sixteen-column array halves that, to 62 m at 500 m. So the cell a 100 MHz macro site resolves at 500 m is about 1.5 m deep and 125 m wide: a thin arc rather than a box. It is the reason 3GPP’s evaluations lean so heavily on fusing several sites that see the same object from different angles: that is where most of the accuracy comes from.

Frequency helps here in a way it cannot help with range. A panel of a given physical size holds twice as many half-wavelength elements at 7 GHz as at 3.5 GHz, and so resolves half the angle. TR 38.914’s deployment scenarios assume base stations with up to 2,304 transmit and receive elements around 7 GHz. On an existing site footprint, the move to 7 GHz improves sensing at least as much through aperture as through bandwidth.

Two other radars already exist, and both see finer

The cellular network is not the only sensing technology with a standard behind it. Car radar and Wi-Fi sensing both have one, both are on sale, and each sets a benchmark the network has to be measured against (Exhibit 6).

Automotive radar has had five contiguous gigahertz at 76 to 81 GHz since FCC 17-94 in July 2017. The Commission’s order records that short-range radars “require an operating bandwidth of four gigahertz”, and Texas Instruments gives the result as a range resolution of 3.75 cm. That is forty times finer than a 100 MHz carrier in n77, and even a 400 MHz channel at 7 GHz or in millimetre wave is ten times coarser. The car industry’s own association, 5GAA, expects 3GPP-based sensing to complement the sensors on the vehicle rather than replace them, and on these numbers it is hard to see how it could do otherwise (part 8 of the 6G series has more).

Wi-Fi sensing has its own published standard, IEEE 802.11bf-2025. NIST’s overview describes a sensing procedure that “supports bistatic and multistatic” operation, and the Wi-Fi Alliance describes “320 MHz channels available in the 6 GHz band” for Wi-Fi 7. A 320 MHz channel resolves 0.47 m. That is finer than any single 5G carrier, and it is achieved inside the home and the office, on equipment the building owner was going to buy anyway.

Put the three side by side and the cellular case sorts itself out. The car’s radar is forty times finer and the building’s Wi-Fi about three times finer, so the network is not going to win on resolution. What the network has that the other two lack is not resolution but four other things.

• Spectrum. The network’s spectrum is licensed and exclusive, and there is an operator with the right to use it.
• Height. Its radios sit on rooftops and towers, above the clutter that defeats a sensor at ground level.
• Reach. It already covers a city, from radios that are already there.
• Accountability. There is a single operator responsible for the system.

Those four count for most outdoors, at height and at distance, which is where drones fly. It is no accident that the standard chose that use first, as part 2 explains.

The terms, briefly

• Range resolution. The smallest difference in distance at which two objects are seen as two. It is the speed of light divided by twice the bandwidth, and it is a property of the signal rather than of the processing.
• Velocity resolution. The smallest difference in radial speed the radar can separate: the wavelength divided by twice the observation time, which radar engineers call the coherent processing interval.
• FR1 and FR2. 3GPP’s two frequency ranges for 5G. FR1 runs up to 7.125 GHz, where the widest carrier is 100 MHz; FR2 starts at 24.25 GHz, where it is 400 MHz.
• Monostatic and bistatic. In monostatic sensing one site transmits and listens for its own echo. In bistatic sensing one site transmits and another listens.
• TS 22.137. 3GPP’s normative service requirements for sensing, with the seven-category performance table. It says what the system shall deliver and leaves the how to the design specifications.