Live status · Read the latest
monad-knowledge Wi-Fi sensing lab · FIIT STU

MonadCount · Wi-Fi sensing research at FIIT STU

Can Wi-Fi tell how busy a room is?

We are studying how people change radio signals in the FIIT library. Explore a recorded day, see where the estimate is unsure of itself, and help us test the idea.

Recorded, not live: this is an estimate from a measured day, not a seat finder. Please do not plan your exam revision around it. Volunteer invitations are open, and the app itself is not public yet.

See a recorded day How it works

People, estimated, · Current · partial day · peak 11.6 at · Open this day

What the devices measure

The room, as one receiver hears it.

recorded One record from each window below. Scroll here with JavaScript on to watch it move.

lower subcarrierssignal strength across the Wi-Fi channel, one line per recordhigher

Thursday afternoonThursday evening One node transmits Wi-Fi frames. This receiver measures how each of 52 subcarriers arrives. People, doors and chairs bend that shape. Counting people from it is the research question.
Showing
monad05, recorded
How much the shape moves
1.1 dB vs 0.21 dB
Channel
ch 48

Recorded windows, same receiver and channel: Thursday afternoon () and Thursday evening (). The movement number is the typical spread of each subcarrier over the clip, in dB. It says the room changed, not how many people changed it.

Open reading roomSilent roomBack roommonad01 — Not reportingmonad01monad02 — Not reportingmonad02monad03 — Not reportingmonad03monad04 — Not reportingmonad04monad05 — Not reportingmonad05monad06 — Not reportingmonad06monad07 — Not reportingmonad07monad08 — Not reportingmonad08monad09 — Not reportingmonad09monad10 — Not reportingmonad10
The FIIT library, ground floor. The dots are the capture nodes.

The boxes in the FIIT library are small computers with Wi-Fi radios. One transmits, the others listen. What they record is Channel State Information (CSI): how strongly each slice of the Wi-Fi channel arrived. The plot beside this text is that record.

Walls do not move, so the shape holds still in an empty room. People do, and the shape starts to wobble. Turning the wobble into a head count, and knowing when not to trust it, is the research.

Bluetooth monitoring runs too. A phone is not a person, and plenty of people carry three, so the two measurements stay separate.

The boxes have no cameras or microphones. Please leave them and their cables where they are: moving one gives us an experiment we did not schedule.

Start with CSI from Zero · All devices, live · Find the devices · Your data and privacy

New here?

You do not need to know wireless sensing to start.

Explore and learn

Pick a recorded day and see where the estimate is sure of itself, and where it is guessing.

Help test the app

Join the volunteer beta list and try short measurement tasks at FIIT.

Student projects

Make this part of your thesis.

You can also do a bachelor’s or master’s thesis here. There are questions in signal processing and machine learning, mobile apps, embedded systems and simulation.

Tell us what you enjoy building or investigating and what you would like to learn. You do not need to arrive knowing wireless sensing. A reproducible bug report is already a very good introduction.

Beta testing and thesis work are separate ways to take part. Ask about either, or both.

Ask about a thesis topic

The research question

Can the estimate keep up with the room?

A room never holds still for long. A chair moves, a door opens, the sun heats a wall, and every one of them changes the radio signal. A useful counter has to notice these changes without mistaking each one for a person.

We are investigating whether Bluetooth measurements can tell us when a Wi-Fi model needs recalibrating. Volunteers walk known routes or report what they see, which gives the radio readings something to be checked against. Simulations let us rearrange the room and the crowd as often as we like before an idea meets real hardware.

This is work in progress. The experiment pages keep what we have measured apart from what we still need to test.

Student beta · coming soon

Help test the app.

The app is not public yet. We will first invite volunteers to a student beta, and app-store access comes later. There is no launch date yet, and we would rather not invent one.

You will try short measurement tasks at FIIT and tell us how they went, especially the confusing and broken parts. No wireless experience needed. A bug you find is a bug a stranger never has to.

Notify me about the beta

Opens the signup form. Leave your email and we will write when a place is free. Joining the list does not enrol you in a study or guarantee a place.

Research so far

Explore the experiments.

All experiments

answered EXP-LIB-01

Overnight CSI drift baseline, empty library

Finding The August overnight study measured median 0.0056 dB SD of one-minute ratio medians over ten hours and twenty node-frequency groups, with 90.2% delivery against the injector counter.

31 capture sessions

measured EXP-007

Packet-Rate Floor for Motion Sensing

Finding The August 22 ladder, analysed August 23, sent 555,193 frames with zero errors or skipped deadlines through 500 Hz.

172 capture sessions

Recorded walks · Explore the data and tools

The measurement archive

4,575capture sessions

10,617node-hours on air

11,880Mchannel estimates

7,502simulation runs

Hardware recordings and simulation runs are counted separately, because a simulated person has never once sat in the wrong chair. 257 quarantined sessions are left out of the hardware totals. Inspect the archive.

Go deeper: Occupancy · Experiments · Recorded walks · Methods and tools · Wi-Fi and Bluetooth explained

For developers: QGIS plugin · Source code and research vault