Uncategorized
Wi-Fi positioning accuracy is quoted as a single number far more often than it deserves to be. A vendor says three metres, a buyer writes three metres into a requirements document, and nobody agrees on whether that was a median or a worst case, measured on which phone, in which building, with the shelves full or empty.
The honest answer is that Wi-Fi can put a person in the right room, on the right floor, most of the time, and that the method you choose moves the result by an order of magnitude. Trilateration from signal strength and time-of-flight ranging on a 160 MHz channel are both “Wi-Fi positioning”. They are not the same product.
This article covers what each Wi-Fi method physically measures, what caps its accuracy, what the IEEE and the FCC have actually specified, and how to write an accuracy clause that a supplier cannot pass by picking a friendly average.
What is Wi-Fi indoor positioning?
Wi-Fi indoor positioning estimates where a device is inside a building by comparing radio measurements taken between that device and known access points. Nothing is tracked by GPS. The position comes from signal strength, flight time, angle, or a stored map of past measurements.
Four families of technique sit under that one label, and the differences matter more than the brand on the access point. Received Signal Strength Indicator (RSSI) trilateration converts signal strength into an estimated distance. RSSI fingerprinting skips distance entirely and matches the live pattern of signal strengths against a survey recorded earlier. Fine Timing Measurement (FTM) measures how long a frame takes to fly there and back. Angle of arrival uses an antenna array to work out a direction rather than a distance.
| Method | What it measures | What caps its accuracy |
|---|---|---|
| RSSI trilateration | signal strength as a distance proxy | walls and bodies change strength, not distance |
| RSSI fingerprinting | a stored pattern of signal strengths | the map ages as the building changes |
| Fine Timing Measurement | round-trip flight time | channel width sets timing resolution |
| Angle of arrival | direction from an antenna array | reflections arrive from the wrong direction |
How accurate is Wi-Fi positioning, really?
Expect room-level or zone-level accuracy from signal-strength methods, and metre-level accuracy from time-of-flight methods on wide channels. Published, peer-reviewed evaluations are the only numbers worth planning against, because they state the building and the test set.
The most-cited public benchmark is the UJIIndoorLoc dataset, collected across three buildings at Universitat Jaume I covering close to 110,000 square metres, with 520 access points visible and 19,937 training plus 1,111 validation samples (UCI Machine Learning Repository, dataset published 2014). The baseline published with that dataset, a k-nearest-neighbour fingerprint match, reached a mean positioning error of 7.9 metres and identified the correct floor 93.86 per cent of the time (Torres-Sospedra et al., IPIN 2014).
Read those two figures together. An average error of a few metres, and a floor wrong roughly one time in sixteen. In a hospital or a shopping centre, the floor error is the one that produces a complaint, because a visitor sent to the right coordinates on the wrong level is simply lost.
Time-of-flight is bounded by physics instead of by map freshness, and the physics is pleasant arithmetic. The metre is defined from a speed of light fixed at exactly 299,792,458 metres per second (NIST, 2019 SI redefinition), so a radio signal covers roughly 30 centimetres in a nanosecond. Timing resolution scales with channel bandwidth. A 20 MHz channel gives you coarse timing and therefore metres of ranging uncertainty; a 160 MHz channel, available since Wi-Fi 5 and widespread in the 6 GHz band the FCC opened to 1,200 MHz of unlicensed use in April 2020 (FCC news release, 23 April 2020), narrows that by close to an order of magnitude.
That is the real reason modern Wi-Fi positioning is better than the 2015 version. Wider channels, not cleverer marketing.
Why does the same hardware give different accuracy in two buildings?
Because Wi-Fi positioning measures the building as much as it measures the device. Radio behaviour is a property of the geometry, the materials and the occupancy, so a system calibrated in an empty warehouse will not reproduce its numbers in a stocked one.
The variables are specific and they are all surveyable before you buy anything.
| Variable | Effect on error |
|---|---|
| Access point density and geometry | few or co-linear APs widen the error ellipse |
| Ceiling height and open atriums | signal carries between floors |
| Shelving, stock, water, crowds | attenuation changes hour to hour |
| Firmware and client chipset mix | phones do not all report the same measurements |
The last row is the one that surprises buyers. A fingerprint survey recorded on one handset does not transfer cleanly to another, because different radios report different signal strengths for the same physical conditions. The same is true of time-of-flight: FTM only works when both ends support it, and a building full of mixed-vintage devices will fall back to signal strength for some of them. Any accuracy figure quoted without a device list is incomplete.
Occupancy matters more than most site surveys admit. The 2.4 GHz band is absorbed by water, and a crowd is mostly water. A fingerprint map built on a quiet Sunday describes a different building from the one that exists at noon on a Saturday.
What does IEEE 802.11az change?
IEEE 802.11az, the next-generation positioning amendment, standardises higher-accuracy ranging and brings angle information and secure, scalable measurement exchanges into the Wi-Fi standard itself (IEEE Standards Association, IEEE 802.11az-2022).
It matters for three practical reasons. Measurement becomes a standard negotiated service rather than a vendor extension, so multi-vendor estates have a common language. The protocol supports many clients ranging concurrently, which is the difference between a demonstration and a stadium. And it carries protection against the spoofing and replay attacks that make distance measurement a security question as well as an engineering one.
What it does not do is repeal physics. Sub-metre ranging still needs wide channels, line of sight to enough access points, and infrastructure that supports the amendment on both sides of the link. A building wired for coverage, with access points placed to avoid dead spots rather than to form good geometry, will not reach the standard’s best case no matter which amendment the firmware claims.
Reflections remain the hard limit. In a corridor with metal doors, the first signal to arrive may have bounced, and an algorithm that trusts it will confidently report a position on the other side of a wall.
Does the FCC’s 3 metre rule mean phones already know your floor?
No. The FCC’s rule is a statistical obligation on carriers for emergency calls, not a guarantee that any given phone knows its floor at any given moment.
The relevant benchmarks are worth knowing precisely, because they are the only widely enforced accuracy numbers in indoor location. The FCC requires wireless providers to deliver either a dispatchable location or horizontal coordinates within 50 metres for 80 per cent of indoor wireless 911 calls, and a vertical (z-axis) estimate within 3 metres above or below the handset for 80 per cent of such calls, phased in by market size with nationwide carriers reaching full coverage in April 2025 (FCC, 911 location accuracy rules).
Three metres vertically is about one storey. That is the regulator’s judgement of what is useful to a paramedic, and it is a reasonable floor-level target for a commercial system too. Note what the horizontal figure tells you: 50 metres, for 80 per cent of calls, is the bar for emergency response, which should temper any assumption that handsets already solve indoor positioning on their own.
The rule is also a useful rhetorical tool in a procurement. If a supplier cannot describe their accuracy the way the FCC does, as a percentile against a stated population of attempts, the number they are giving you is not measurable.
How do you write an accuracy spec you can actually test?
Write it as a percentile with a named test method, a device list and a building state. ISO/IEC 18305:2016 exists for exactly this purpose: it is the international standard for the test and evaluation of localisation and tracking systems (ISO/IEC 18305:2016).
| Clause to write | Why it matters |
|---|---|
| 95th percentile error, not the average | the tail is what users complain about |
| Floor identification rate, stated separately | 2 metres on the wrong floor is useless |
| Test routes, walking speed, device list | results do not transfer between handsets |
| Acceptance measured in the stocked building | an empty building is a different radio environment |
Two more clauses are worth adding. Specify how the system behaves when it is uncertain, because a position estimate that silently degrades is worse than one that says it does not know. And specify re-test conditions after any change to the access point estate, since moving six access points during an unrelated network refresh can quietly undo a fingerprint survey that took a week.
Ask for the cumulative distribution of errors, not a headline figure. It is one chart, every serious supplier already has it, and it answers in ten seconds what a procurement thread will otherwise argue about for a month.
When is Wi-Fi the wrong choice?
When your requirement is centimetre-level, safety-critical, or must work where there is no Wi-Fi infrastructure worth reusing. Wi-Fi positioning earns its place by riding on a network the building already has, and that advantage disappears the moment the accuracy requirement forces dedicated hardware anyway.
| Requirement | Wi-Fi positioning |
|---|---|
| Guide a visitor to a room, shop or gate | usually enough |
| Locate a wheelchair or pump on a floor | often enough with good AP geometry |
| Prove a tool was within 30 centimetres of a bolt | not the right tool |
Three situations where another technology is the honest answer:
- Centimetre-level industrial tasks, where ultra-wideband ranging is built for the job and Wi-Fi is not.
- Battery-powered tags that must last years on a coin cell, where Bluetooth Low Energy beaconing is the cheaper physics.
- Buildings with sparse or chaotic access point coverage, where the cost of fixing the network to serve positioning exceeds the cost of a purpose-built layer.
The reverse case is just as real and gets argued less often. If the requirement is “get a visitor from the car park to a clinic room without asking at a desk”, then room-level Wi-Fi positioning with reliable floor detection solves the whole problem, and buying centimetre accuracy is buying precision nobody will ever notice.
Every number in this article is a property of a building, not of a product, which is why the useful next step is a conversation about yours: your access point density, your floor plan, your device mix, and the accuracy you actually need at the point of use. Book a scoping call and bring your floor plans.
Frequently asked questions
Four questions buyers ask most often about Wi-Fi positioning accuracy, answered without marketing numbers.
Is Wi-Fi positioning accurate enough for wayfinding?
For most wayfinding, yes. Turn-by-turn guidance inside a building needs the system to know which corridor, room or floor you are on, not your exact coordinates. The requirement that usually bites is floor detection rather than horizontal error, since the published UJIIndoorLoc baseline identified the correct floor 93.86 per cent of the time. Specify floor accuracy separately from horizontal accuracy in any procurement.
Does adding more access points always improve accuracy?
No. Geometry matters more than count. Access points arranged in a line, or clustered where coverage was weakest, leave the position poorly constrained along one axis no matter how many you add. Accuracy improves when new access points are placed to surround the area being located and to give each point in it a distinct signal signature. A coverage design and a positioning design are different drawings.
What is the difference between RSSI and Fine Timing Measurement?
RSSI infers distance from how strong a signal is, which walls, furniture and people all corrupt. Fine Timing Measurement measures how long a frame takes to travel there and back, so it measures distance directly. Timing resolution depends on channel bandwidth, and because light covers roughly 30 centimetres per nanosecond, wider channels give tighter ranging. FTM needs support at both ends of the link.
Can Wi-Fi positioning tell which floor someone is on?
Usually, but not reliably enough to leave unspecified. Signals carry between floors through ceilings, stairwells and atriums, so floor classification is a separate problem from horizontal position and has its own error rate. The FCC treats vertical accuracy separately too, requiring a z-axis estimate within 3 metres for 80 per cent of indoor wireless 911 calls. Ask any supplier for their floor identification rate as a distinct number.
