OVERCALIBRATION
WIFI + CAMERA ML FUSION
TAP ANYWHERE TO BEGIN
TRAINING
INITIALIZING
PREPARING SENSORS
In development · not yet certified

Wearable 60 GHz Presence Sensing

A wrist-worn millimetre-wave radar that detects people in the space in front of you — where there is no fixed sensor and no infrastructure to rely on. Built on the LILYGO T-Watch Ultra reference design with one radar added.

0.2–5 m
Presence range, line of sight
1–3
Resolvable people
18.5 h
Radar running continuously

SENSOR COVERAGE, DRAWN TO SCALE · 100° × 5 m

Engineering review · August 2026

Earlier versions of this page advertised 100 m through-wall detection, RFID reading at 500 m, a custom ML-accelerated processor and a set of certifications the product does not hold. A full engineering and regulatory review found that none of those claims can be supported, and several are physically impossible at 60 GHz. They have been removed.

What remains below is what a wrist-worn millimetre-wave sensor can actually do, with the measurement or the statute behind each figure. The full workings — physics, bill of materials, manufacturing routes and certification path — are published in the engineering and cost analysis.

Regulatory status

What we hold, and what we do not

No conformity assessment is complete. Nothing on this page may be read as a certification claim.

Planned
CE · RED 2014/53/EU
ETSI EN 305 550 for the 57–64 GHz radar, EN 301 489-3, EN 62368-1, EN 62479. Test campaign not yet booked.
Planned
RED Article 3.3 cyber
EN 18031-1 and -2, mandatory for connected wearables since 1 August 2025.
Not held
ISO 9001 · ISO 27001 · MIL-STD-810H
Previously shown on this page in error. No audit has been carried out and no certificate exists.
Not claimable
"Swiss made"
The Swiss Made Ordinance (SR 232.119) Art. 2(1)(c) requires 50% of movement component value to be Swiss. Imported silicon cannot meet it. A Swiss-machined case under Art. 4 is achievable.
Deliberately outside
Medical device (EU MDR)
This is a wellness product. It makes no diagnostic, monitoring or alerting claim, and must not be used for any clinical purpose.

What it actually does

Every figure below is a datasheet number or a published measurement, derated for a sensor worn on a moving wrist rather than bolted to a ceiling.

It can
Detect that a person is present, 0.2–5 m ahead, in line of sight
Tell you which zone of its field they are in
Separate one to three people when they are not in line with each other
Sense breathing, and estimate heart rate, with the watch set down and still
Read gestures and micro-motion within about a metre
Run all day: 18.5 hours with the radar on continuously
It cannot
See through a wall — one plasterboard wall costs 24–64 dB round trip
Reach 100 m — that needs 3.2 kW of radiated power; it may legally emit 10 mW
Track 50 people — the best fixed sensors lose accuracy at five
Read RFID at range — the NFC coil works to a few centimetres
Measure a heartbeat while you are wearing and moving it
Detect concealed objects, or identify anyone
📡

Presence & zone occupancy

A 60 GHz pulsed coherent radar with the antenna inside the chip package. Reports whether someone is in its field, which zone they are in, and whether the motion is fast or slow. Reliable to 5 m from the wrist; the bench figure for the bare sensor is 7 m.

🫁

Respiration, in docked mode

Chest-wall movement is about 3 mm and readable. Set the watch on a surface, 0.3–1.5 m from a still subject, and it recovers breathing rate to roughly 0.5–2 breaths per minute. Worn and moving, the measurement does not survive — your wrist moves ten times further than the signal.

🔋

A radar that fits the battery

The sensor averages about 3 mW at a 10 Hz update rate and draws 43 mA from the cell at full rate. On the 1100 mAh battery that is 18.5 hours of continuous sensing, or a full 24-hour day at 73% duty cycle. Alternatives in the same class draw 6–12× more.

🧭

Works where nothing is installed

A ceiling sensor is better than this device in any room that has one. The case for wearing it is mobility: a lone worker moving through a plant, a responder in a building with no infrastructure, a carer walking rounds. That is the whole argument, and we do not stretch it further.

⚙️

Open, inspectable firmware

An Espressif ESP32-S3 running the Acconeer radar service alongside an LVGL interface. No custom silicon, no encrypted bytecode, no claims about firmware that cannot be extracted. The signal chain is documented so a buyer's own engineer can check it.

📋

Numbers you can audit

Every specification traces to a manufacturer datasheet, a federal ordinance, a distributor price list or arithmetic we show. Where a figure is our estimate, it says so. The full analysis, including the parts that are unflattering, is published.

How it works

Not Wi-Fi, and not through walls. A dedicated millimetre-wave sensor illuminating the space directly in front of you.

STEP 1

A 60 GHz pulse goes out

A wavelength of 4.96 mm, radiated at about 11 dBm from an antenna built into the 5.5 mm chip package. It travels through air and clothing, and stops at walls and bodies.

STEP 2

The echo comes back shifted

A person reflects the pulse. Because the sensor is coherent it measures the phase of that echo, so movements far smaller than the wavelength show up as a phase change.

STEP 3

The watch decides

On-device processing separates a living body from a static reflection by its motion — including the small, regular motion of breathing. No cloud, no upload, no image.

Because the returned signal is a range and a motion, not a picture, the device cannot identify a person, see an object's shape, or tell what someone is carrying. That is a property of the physics, not a limitation of this version.

Reference design plus one sensor

Built on published parts

RadarAcconeer A121 · 57–64 GHz
Radar package5.5 × 5.2 × 0.88 mm, antenna in package
Radiated power11.4 dBm EIRP (limit: 100 mW)
Presence range0.2–5 m worn · 7 m bench
ProcessorEspressif ESP32-S3 · dual LX7 240 MHz
Memory16 MB flash · 8 MB PSRAM
Display2.06″ AMOLED 410 × 502, 600 nit
Battery1100 mAh / 4.07 Wh
Runtime18.5 h radar on · 5.3 days radar off
Case63.5 × 49 × 22 mm (reference design)
WeightNot yet measured
AssemblyUnder evaluation — see cost analysis

Dimensions are those of the LILYGO T-Watch Ultra reference design and will change once our board is laid out. No device weight is published by the manufacturer, so we do not quote one.

Not yet for sale

The device is pre-certification. We are placing a small number of units with pilot partners who need presence sensing where no fixed sensor exists.

Pilot unit
RADAR UHR · DEV
60 GHz presence sensing, wrist-worn
CHF 1,490 indicative

Indicative only · no unit may ship before CE conformity is complete

What a pilot includes

  • Pre-production unit, serial-numbered
  • Presence detection and zone occupancy
  • Respiration sensing in docked mode
  • Raw radar data access for your own engineers
  • Six-month evaluation with direct engineering support
  • Full disclosure of test results, including failures

No payment is taken at enquiry. Pilot pricing is confirmed in writing once the certification timetable is fixed.

Procurement Inquiries

For pilot enquiries, engineering questions and evaluation agreements.

Pilot programme:

pilot@radaruhr.ch

Technical Inquiries:

technical@radaruhr.ch

CAMERA POSE DEMO · SIMULATED RF
INITIALIZING
Requesting camera access...
RF HEATMAP — SIMULATED
--
TRAINING
0 FPS MODE: -- SAMPLES: 0 USERS: 1
pose pipeline — camera input, not radar
SESSION 1 | 0 embeddings | 0 rounds
This demo runs a pose model on your camera in your browser. The RF panel is a visualisation, not a radar measurement — no radar hardware is involved and nothing is sent anywhere.
© 2024-2026 RADAR UHR
ENQUIRE ABOUT A PILOT
Active Acoustic Sensing Enabled
FMCW CHIRP SONAR • 17 - 22 kHz

You may perceive faint clicking or ticking sounds. This is our Frequency-Modulated Continuous Wave (FMCW) acoustic sonar system transmitting near-ultrasonic chirp pulses through your device speakers.

Each pulse sweeps from 17,000 Hz to 22,000 Hz over a 25-millisecond window. Your device microphone captures the reflected wavefront, which is then decomposed via a 4096-point Fast Fourier Transform at 48 kHz sample rate (11.72 Hz bin resolution) to extract the room’s acoustic impulse response.

When a human body enters the acoustic field, it perturbs the baseline echo signature - the reflected energy distribution shifts across frequency bins proportional to the target’s radar cross-section and range. We compute the delta between the current spectral frame and the stored room baseline to derive a presence confidence score and distance estimate (round-trip propagation at 343 m/s yields ~37.9 ms for 6.5 m).

17 → 22 kHz
FM CHIRP BANDWIDTH
6.6 Hz
PULSE REPETITION FREQ
4096-pt FFT
SPECTRAL RESOLUTION
11.72 Hz/bin
FREQUENCY RESOLUTION

This technique is grounded in the same FMCW principle used in radar systems since 1940 and biologically observed in Chiroptera (bats), which echolocate using 20-200 kHz ultrasonic pulses. Our implementation adapts this for commodity hardware using the Web Audio API.

ⓘ Safety notice: The emission amplitude is held at −32 dBFS (0.025 gain), well below occupational exposure limits defined in IEC 61672-1. Frequencies above 17 kHz are beyond the audible threshold for most adults over 30 (ISO 7029:2017). The faint “click” you may hear is the 17 kHz onset - fully harmless and equivalent to ambient environmental noise levels.
PEER-REVIEWED REFERENCES
[1] Adib, F. & Katabi, D. “See Through Walls with WiFi!” - ACM SIGCOMM 2013
[2] Adib, F., Kabelac, Z. & Katabi, D. “3D Tracking via Body Radio Reflections” (WiTrack) - USENIX NSDI 2014
[3] Mao, W., He, J. & Qiu, L. “CAT: High-Precision Acoustic Motion Tracking” - ACM MobiCom 2016
[4] Nandakumar, R. et al. “Contactless Sleep Apnea Detection on Smartphones” (ApneaApp) - ACM MobiSys 2015
[5] Lian, J. et al. “EchoSpot: Spotting Your Locations via Acoustic Sensing” - ACM IMWUT 2021
[6] Zhou, B. et al. “BatMapper: Acoustic Sensing Based Indoor Floor Plan Construction” - ACM MobiSys 2017
[7] Shih, O. & Rowe, A. “Occupancy Estimation using Ultrasonic Chirps” - ACM/IEEE ICCPS 2015
[8] ISO 7029:2017 - Statistical distribution of hearing thresholds related to age and gender
AUTO-DISMISS IN 6s
Nathan Immis
Hyperlog Agency - Swiss AI Engineering
+41 78 643 85 73
nathan@pingwage.com
tap to dismiss