How you locate someone where there is no GPS

The classic indoor alternative to GPS has been RSSI: estimating the distance to an access point from how strongly its signal is received. It works, but only just. Received power depends on the phone model, antenna orientation, whether the officer holds the phone or keeps it in a pocket, how crowded the platform is and what stands in between. Two phones at the same spot can report very different readings.

Wi-Fi RTT (Round-Trip Time) solves the problem another way. The FTM protocol (Fine Timing Measurement), standardised in IEEE 802.11mc in 2016, lets the terminal exchange a burst of frames with the access point and measure the round trip with nanosecond precision. Since the signal travels at the speed of light, that time is a distance.

What changes in practice

Android has exposed RTT natively since Android 9, with broader support from Android 10 onwards. That is what makes a smartphone-centric approach viable: no dedicated device is required.

This is where neat laboratory figures stop applying, and it is worth saying so before anyone discovers it during a trial.

Time of flight becomes a distance on the assumption that the signal travelled in a straight line. When a column, a stationary train or a tunnel curve sits between terminal and access point — a non-line-of-sight (NLOS) situation — the signal arrives reflected. It travelled further, so the time measurement is longer and the system believes the officer is further away than they are. The error is also biased: always in the same direction, never self-cancelling.

Multipath is the related phenomenon: the same signal arrives several times along different paths, and the receiver has to decide which arrival is the direct one. A platform, with its metal surfaces and parallel walls, is one of the worst possible settings.

What Metrosafe does about it

• NLOS detection: measurements whose profile betrays a reflection are flagged and lose weight in the estimate, instead of entering as if they were sound.
• Consensus checks: if a measurement contradicts the others and the recent trajectory, it is discarded.
• Filters with memory: the estimate carries the previous state, so an isolated spike does not teleport the officer thirty metres.
• Per-station calibration: each site’s systematic biases are measured during deployment and corrected. The same station behaves the same way every day; that is exploitable.

A generic location system treats space as a continuous surface: the person could be anywhere. A metro station is nothing like that. It is a narrow, heavily constrained graph: platform, concourse, interchange corridor, stairs, tunnel. People do not walk through walls or along the track.

Map matching puts that knowledge inside the estimator. Candidate positions falling outside walkable space are discarded; those inside are snapped to the axis of the space. The technique is well established in navigation and cartography, and its indoor effect is the same: it reduces estimator variance and corrects systematic bias.

And it is achieved at low computational cost, unlike alternatives such as particle filters, which solve a similar problem while consuming far more.

The literature reports 1-2 metres for Wi-Fi RTT under direct line-of-sight conditions. A platform is not that condition. Our target figure in station is 5 to 10 metres, and it is deliberately more conservative.

Accuracy in station

5-10 m

Publication latency

< 5 s

Refresh rate

configurable · 15-30 s recommended

Availability

SLA ≥ 99.95 %

Scale

≥ 70 concurrent terminals

Is 5-10 metres any use?

That is the right question, and the answer depends on what for. To guide a robot around a warehouse, no. To coordinate an incident in a station, yes, and with room to spare: in platform topology, five metres separate platform from concourse, and ten separate one direction of travel from the other. Nobody in a control room needs to know which tile their officer is standing on. They need to know where to send support and which entrance to use.

Promising sub-metre accuracy over Wi-Fi on a platform would be selling something the physics of the channel does not deliver. We would rather the figure survived the field trial.

IEEE 802.11az, published in 2022 and known as Next-Generation Positioning, is the evolution of Wi-Fi positioning: higher measurement rates, more bandwidth and better ambiguity handling, with sub-metre accuracy in favourable scenarios. Android exposes it on compatible devices from Android 15 onwards.

Metrosafe is designed to adopt it without a redesign: the fusion engine treats a ranging measurement as an input with an associated uncertainty, and does not care whether it came from 802.11mc or 802.11az. As the terminal fleet catches up, accuracy improves on its own and the investment already made is preserved.

• Dai, J., Wang, M., Wu, B., Shen, J. & Wang, X. (2023). A survey of latest Wi-Fi assisted indoor positioning on different principles. Sensors, 23(18), 7961.
• Cao, H. et al. (2024). LOS compensation and trusted NLOS recognition assisted WiFi RTT indoor positioning algorithm. Expert Systems with Applications, 243, 122867.
• Mansour, M. F. & Waters, D. W. (2013). Map-assisted Kalman filtering. IEEE ICASSP, 3208-3212.
• IEEE Standards Association (2023). IEEE 802.11az-2022.
• Android Open Source Project (2025). Wi-Fi RTT (IEEE 802.11mc, IEEE 802.11az).