What is measured, and what is not.

The short version

Ten small radios sit in the FIIT library and measure what the room does to a Wi-Fi signal. They are not cameras, they are not microphones, and today they do not record anything that identifies a person or a phone.

If that ever changes, it changes here first. This page is versioned in the same repository as the research itself, so the date at the bottom is the date the practice changed, not the date somebody remembered to update the wording.

Who is responsible

The deployment is part of a doctoral research project at the Faculty of Informatics and Information Technologies, Slovak University of Technology in Bratislava. Questions, objections and requests go to the researcher directly — the contact address is on the About page, and a written question gets a written answer.

What is measured today

One node transmits a steady, deliberately boring stream of Wi-Fi frames. The others listen and record Channel State Information: a few thousand complex numbers per second describing how that signal arrived — which frequencies came through strong, which came late after bouncing off a wall.

That recording describes the room. Bodies change it, because a body is mostly water and water absorbs microwaves. It supports "roughly this many people are present". It does not support "this is who", and no amount of post-processing turns one into the other.

What is not collected

• No cameras and no microphones. There is no optical or acoustic sensor anywhere in the deployment.
• No content. Nothing anybody sends over the network is captured, read or stored. The measurement is of our own transmitter's frames arriving at our own receivers.
• No device identifiers. No MAC addresses of passers-by, no Bluetooth addresses, no association with anyone's phone. Your device is not a participant in this measurement and is not asked to be.
• No location trails. Nothing here follows a person from room to room.

Volunteer sessions

Some measurements need a known answer to compare against, so volunteers are occasionally asked to follow a short written script — walk this route, scan this card, stand still here for two minutes.

Those sessions are opt-in and briefed in advance. The script is written down before anyone starts, so a participant knows exactly what is being recorded. Participation can be stopped at any point without giving a reason and without consequence, and the session data can be discarded on request while the session is still identifiable as theirs.

The change we are considering, and have not made

Honest research notes what it is about to do, not only what it has done.

A planned extension would count how many radios are active nearby — the number of distinct devices transmitting in the band — and use that as a second, independent estimate of how busy a room is. It is attractive because it fails in different ways than the Wi-Fi measurement does: a room full of people sitting perfectly still is nearly invisible to the channel measurement and obvious to a device count.

It would also be the first thing this project does that touches device identifiers, so it does not start quietly. If and when it runs:

• Identifiers are hashed at the moment of capture with a rotating secret, never written to disk in the clear, and are used only to answer "how many distinct devices", never "which device".
• The hash secret rotates on a fixed schedule and old secrets are destroyed, so the same phone seen next month is not linkable to the same phone seen today.
• Only counts are retained beyond a short working window. The per-device rows are an intermediate step to a number, and the intermediate step is deleted.
• Modern phones already randomise the addresses they broadcast, so what is seen is in most cases a temporary pseudonym that the phone itself invented and will discard.
• This page is updated before the first such capture, with the retention window and rotation period stated as numbers.

One consequence deserves saying plainly rather than being buried: because this design deliberately makes it impossible to link a record to a person, it is also impossible to find and delete your record on request. That is not an evasion — it is the direct cost of building the thing so it cannot identify you in the first place. If you would rather not be counted at all, switching off Wi-Fi and Bluetooth removes you from the measurement entirely, and turning them back on does not restore anything, because nothing was kept.

Why this is done at all

Counting people with radio is a research problem with a specific unglamorous difficulty: the models drift. Move the furniture and yesterday's model is wrong without announcing it. The thesis argues that a second, independent opinion is what makes drift visible, and the whole point of the deployment is to test that argument on real hardware in a real room rather than in simulation.

The results are published. The vault, the code and this page are in the same public repository.

Contact

Questions, objections, or a request to be told more: jakub.dubec@stuba.sk.