Ask a Wi-Fi link is something moving and one radio answers well. Ask it where is the device and one radio cannot, no matter how clever the software. Position is a geometry problem, and geometry needs more than one viewpoint. That single fact shapes everything about indoor localization, so we start there.
01Why a single antenna cannot give position
A single Wi-Fi link measures how the channel changed between two points. From that you can estimate how far away a moving thing roughly is, because closer motion disturbs the signal more, and in principle you can estimate a distance to a transmitter. But a distance is a circle, not a point. If a device is three meters from your board, it could sit anywhere on a three-meter ring around it. One measurement can never collapse that ring to a dot.
This is not a limitation of CSI or of cheap hardware. It is basic geometry. To fix a point on a floor you need to intersect information from several known locations, and a lone antenna only ever offers one. We go into the deeper reasons — noisy phase, no spatial diversity — in the honest limits of single-antenna CSI. For positioning, the headline is simply: you will need anchors.
02How multiple anchors locate a device
An anchor is a Wi-Fi device at a known, fixed position. Put three or more around a room and you can combine what each one measures. There are three classic physics-based methods, and real systems often blend them.
| Method | What each anchor measures | How position comes out |
|---|---|---|
| Trilateration (ToF / range) | Distance to the device | Intersect distance circles from several anchors |
| Angle of arrival (AoA) | Direction the signal came from | Intersect bearing lines from several anchors |
| Time of flight (ToF) | How long the signal took to travel | Convert time to distance, then trilaterate |
Trilateration is the circles picture in the hero figure: each anchor's distance draws a ring, and the rings cross at the device. Angle of arrival instead measures the direction the signal came from, which needs an anchor with several antennas, and crosses bearing lines rather than circles. Time of flight turns the tiny travel time of the signal into a distance, which demands tight timing but can be very precise. CSI helps all three, because its per-subcarrier phase carries the timing and angle detail that a single strength number throws away.
03Fingerprinting: learning the room instead of computing it
There is a second, very different approach that leans into multipath rather than fighting it: fingerprinting. Instead of computing geometry, you teach the system what the radio looks like at each spot. In a survey stage you walk the space, stand at many known points, and record the CSI each anchor sees at each point. That map of location-to-signal is the fingerprint database.
Later, to locate a device, you compare its live CSI against the database and pick the closest match, or interpolate between the nearest recorded points. The reflections that ruin geometry become an asset here, because the messy, specific pattern at each spot is exactly what makes locations distinguishable. The catch is that the survey is laborious and the fingerprint goes stale when the room changes — move a metal shelf and part of your map is wrong.
04How CSI fingerprinting differs from RSSI fingerprinting
Wi-Fi positioning long predates CSI. The original trick was RSSI fingerprinting: record the received signal strength from each access point at each location and match against it later. It works, phones have used it for years, but it is coarse. RSSI is a single number per access point, it wobbles with crowds and interference, and two nearby spots often look almost identical. The classic difference between the two measurements is covered in CSI vs RSSI.
CSI fingerprinting swaps that one number for amplitude and phase across dozens of subcarriers, so each location gets a far richer and more distinctive fingerprint.
| RSSI fingerprinting | CSI fingerprinting | |
|---|---|---|
| Data per anchor | One strength number | Amplitude + phase, many subcarriers |
| Location detail | Coarse, nearby spots blur | Finer, spots stay distinct |
| Enables AoA / ToF | No | Yes, phase carries it |
| Complexity | Low | Higher survey and compute |
| Drift when room changes | High | High — richer map, still needs resurvey |
The short version: CSI gives you a sharper map and unlocks physics methods RSSI cannot touch, at the price of more setup. Both still need a survey and both still drift, so neither is magic.
05The honest accuracy
This is where expectations need managing. In careful laboratory conditions — a controlled room, several anchors, good calibration, sometimes multiple antennas per anchor — CSI localization can reach sub-meter accuracy. Those are the numbers that make headlines, and they are real for their setup.
In an ordinary home or office, the picture is more sober. Multipath, furniture, closing doors and people walking around all add error, and a realistic figure is better measured in a few meters, tightening as you add anchors and invest in calibration. The honest way to read any published figure is as a best case for its exact conditions:
- Lab, many anchors, calibrated: sub-meter is achievable.
- Typical home or office, a few anchors: expect meters, not centimeters.
- Moving people, changing layout: accuracy wanders and fingerprints go stale.
None of this makes Wi-Fi positioning useless — knowing which room or which corner a device is in is valuable, and that is very achievable. It just means you should size a project around room-level or few-meter accuracy and treat anything tighter as a goal to earn with more anchors and effort.
06Building it in your own space
For a hands-on experiment, place three or more CSI-capable boards at known, fixed points around a room you own. Decide early between the two philosophies: a geometry approach with trilateration or angle of arrival, which rewards good timing and multiple antennas, or a fingerprinting approach, which trades the survey effort for robustness to multipath. For a first project, fingerprinting is usually the kinder starting point because it needs no precise timing.
Keep the work inside a space you control. Indoor positioning is for locating your own devices and tags on your own floor plan — asset tracking, a home robot, a smart-home trigger — not for following people who have not asked to be followed. Build it for your own rooms, measure your own accuracy, and let the geometry, not a viral claim, set your expectations.
07FAQ
Can Wi-Fi find my indoor position?
It can, but not with one radio. A single link gives at most a rough distance, not a position. To place a device on a floor plan you need several anchors at known locations, and software that combines their measurements by trilateration, angle of arrival, time of flight or fingerprinting. With that in place, Wi-Fi positioning is real and useful.
How accurate is Wi-Fi CSI indoor localization?
In careful lab conditions, CSI methods can reach sub-meter accuracy. In an ordinary home or office, with multipath, furniture and moving people, a few meters is a more realistic figure, improving as you add anchors and calibration. Treat sub-meter claims as best-case lab results, not what you will see on day one.
How is CSI positioning different from RSSI fingerprinting?
RSSI fingerprinting uses one coarse number per access point, so its map is blurry and drifts over time. CSI gives amplitude and phase across many subcarriers, a far richer fingerprint for each location, and it also enables physics-based methods like angle of arrival and time of flight that plain RSSI cannot support. The trade is more complexity and calibration.
Do I need special hardware for Wi-Fi positioning?
You need several Wi-Fi devices at known, fixed positions to act as anchors, and the ability to read CSI from them. Low-cost ESP32-class boards can capture CSI and serve as anchors for experiments in your own space. The precise physics methods benefit from multiple antennas and tight timing, which is where cost and complexity climb.