How does FTM ranging work?
The initiator sends a request, the responder sends timed frames, and both record four timestamps: when each frame leaves and arrives. The round-trip time minus the turnaround at the responder gives the time of flight, and the speed of light turns it into distance. Distance to three or more access points gives a position by multilateration; Android's developer documentation says the result is "typically accurate within 1-2 meters".
The newer 802.11az amendment adds ranging modes for many devices at once and Secure LTF, which uses single-use pseudo-random sequences to resist man-in-the-middle and time advance attacks. IEEE's own article on its release speaks of sub-meter accuracy, without test conditions.
What spoils FTM accuracy
- Offsets per device. Berthold Horn's notes on FTM RTT report offsets from bandwidth and preamble choices that make distances "as much as 6 to 8 meter too small or too large" until they are calibrated. Espressif's example applies a zero-distance offset for the same reason.
- The building. Signals slow down in concrete and similar materials, so indoor distances come out longer than the straight line. The same notes suggest this can matter more than multipath.
- Signal and setup. Espressif recommends a line-of-sight path with RSSI no lower than -70 dBm and a 20 MHz soft access point for better results.
- The chip revision. Espressif's errata say ESP32-C6 chip revisions v0.0 and v0.1 cannot act as FTM initiator, with no workaround; it is fixed in revision v0.2. Check the revision before you buy boards.
- The phone. Android ranges only in the foreground or a foreground service, needs location or nearby Wi-Fi permission, and throttles apps that ask too often. Horn's notes add that iOS has no FTM RTT API for apps.
How does CSI sensing work?
Every Wi-Fi packet carries training fields that let the receiver estimate the channel. CSI exposes that estimate per subcarrier. When a person moves or breathes, the reflections change and so does the CSI. Espressif lists smart environment monitoring, human activity monitoring and wireless positioning as uses, from people walking to "subtle actions in a static environment, such as breathing and chewing".
On an ESP32 you enable it in the build configuration, register a callback, set the CSI configuration and switch it on; the callback receives the data with RSSI, noise floor, bandwidth and timing. Espressif describes three setups: one ESP32 with your router (cheapest, depends on where the router is), several ESP32s talking to each other, and a dedicated device that sends packets, which it calls the most accurate and reliable.
A CSI model learns one room with one set of devices. Move the router, swap the antenna or add people, and it needs new data. Espressif also notes that an external antenna works better than a PCB antenna, which is directional.
What IEEE 802.11bf changes
Until now each vendor exposed sensing its own way. IEEE 802.11bf-2025, "Enhancements for Wireless LAN Sensing", amends the 802.11 MAC and the PHY service interface for HE and EHT (Wi-Fi 6 and 7) in the 1 to 7.125 GHz bands and for DMG and EDMG above 45 GHz, so devices can request, run and report sensing measurements in a standard way. For a product it means sensing that can work with access points from other vendors, once the chipsets support it.
Write the acceptance plan first
Before choosing a chip, write down what the product must prove and how you will measure it:
- The quantity: distance, position, presence, activity or breathing, and the accuracy or detection rate the product needs.
- Ground truth: how you know the true distance or what really happened in the room.
- Conditions: rooms, materials, line of sight or not, number of people, router positions.
- Units: several boards and phones, because offsets differ per unit and per chip revision.
- The pass line, set by you before the test, and what happens if an option fails it.
Then test each option against that plan and compare the results, before anything goes into firmware.
How we work with this
This is how our R&D team works: we test Wi-Fi FTM/RTT, Wi-Fi CSI, UWB, BLE, LTE and mmWave radar in our own hardware lab, each against a written acceptance plan, and carry the option that passes into Rust firmware with per-device calibration. See signals and sensing, or start with a feasibility study that answers one question in writing: go, change or stop.
FAQ
Q01Can Wi-Fi sensing tell where a person is in a room?
Not on its own in most products today. Wi-Fi CSI reliably shows that something moves and can pick up subtle actions such as breathing, but turning that into a position needs several devices, training data from the same room and an acceptance test on site. Wi-Fi FTM gives the distance of a device that takes part in the exchange, not of a person.
Q02How accurate is Wi-Fi RTT positioning?
Android's developer documentation says that with distances to three or more access points the position is typically accurate within 1 to 2 meters. Device-specific offsets, the building materials and the access points you have change that, so measure it in your own building against a written acceptance plan.
Q03Do we need IEEE 802.11bf hardware to start?
No. FTM is part of IEEE 802.11-2016 and CSI is exposed by chips such as the ESP32 family today. IEEE 802.11bf-2025 standardises how sensing is requested and reported, which matters when you want sensing to work across vendors; check whether your chipset supports it before you plan on it.
Q04Is Wi-Fi sensing data personal data?
It can be. Output that shows when someone is home, moving or breathing is about people, so treat it as potentially personal data, design for minimal retention and check the legal basis with your data protection officer.
Sources
- IEEE 802.11bf-2025, Enhancements for Wireless LAN Sensing (ANSI webstore)
- IEEE: newly released IEEE 802.11az standard improving Wi-Fi location accuracy
- Android developers: Wi-Fi location, ranging with RTT
- Berthold K.P. Horn: FTM RTT issues, problems and limitations (MIT CSAIL)
- Espressif ESP-IDF: Wi-Fi FTM example
- Espressif errata: ESP32-C6 cannot be 802.11mc FTM initiator
- Espressif ESP-IDF: Wi-Fi channel state information
- Espressif ESP-CSI
- Halperin et al.: Linux 802.11n CSI Tool
This guide summarises public standards and vendor documentation for engineering teams. Figures are the sources' own, not our measurements. Check current chipset support with your vendor.
