Why mmWave presence sensing changed the room

A bathroom light switches off while someone is still in the shower. A landing sensor insists a room is empty while someone sits reading in an armchair nearby. Both are the same failure, produced by the same low-cost part: a passive infrared sensor that has stopped seeing a body it is looking straight at. The millimetre-wave radar modules now built into ceiling and wall-mounted presence sensors close that gap, and the way they do it, and the ways they still get fooled, explains both why they cost more and where that cost is worth paying.

The two technologies solve "is someone here" by measuring completely different things. A PIR sensor watches for a change in infrared heat crossing between fixed zones in its lens; no crossing, no signal, regardless of how many warm bodies are sitting in the room. A millimetre-wave radar sensor transmits a radio signal and listens continuously for the frequency shift produced by movement as small as a breathing chest, so it can hold a room as occupied without anyone crossing anything. What follows sets out the mechanism behind each failure, what a radar chip is actually measuring, where the radar sensor generates its own false alarms, how zones get configured on a real device, and which rooms justify the price difference.

The pyroelectric sensor's blind spot

A PIR sensor's detecting element is a pyroelectric crystal sitting behind a segmented Fresnel lens that splits its field of view into a grid of narrow zones, each acting as its own small thermal camera. The output changes only when a heat source, typically a warm body against a cooler background, moves across the boundary from one zone into the next. A person walking through a doorway crosses several zone boundaries in a second and registers instantly. A person sitting still in a chair crosses none, and the output goes flat within moments. PacLights, in its guide to occupancy and motion sensors, puts the problem plainly: PIR technology is built to sense change, not sustained presence, so a device covering a desk or an armchair reads nothing at all once the person in it stops moving enough to cross a boundary. Extending the hold timer, the usual installer's fix, does not restore detection. It only delays the moment the light goes out, at the cost of running longer than it needs to.

What a radar chip is actually listening for

A millimetre-wave presence sensor works on frequency-modulated continuous-wave radar: the chip transmits a signal whose frequency ramps upward across a wide band, several gigahertz at a time, many times a second, and measures the frequency and phase of the reflection that comes back. Infineon's BGT60UTR11AIP, a chip built for exactly this job, radiates across a 5.6 GHz sweep within the 57.4 to 63.0 GHz band with a ±60° field of view, and the manufacturer lists its target uses as presence detection with zone segmentation for smart home and doorbell products, alongside vital-sign tracking for sleep trackers and baby monitors, reading breathing rate and heart rate off the same reflected signal. A resting chest wall moves by a fraction of a millimetre with each breath. At 60 GHz that motion still produces a measurable Doppler signature, so the chip can keep reporting a person present without that person crossing any boundary at all.

Regulators had to formally catch up with this distinction. In the rulemaking that opened up the 60 GHz band for this kind of sensing, the US Federal Communications Commission rewrote its own definition of a field disturbance sensor after chipmakers argued the existing wording did not fit what radar could now do.

That is not the same claim as "radar detects total stillness." No sensor does that. What changes is the size of movement needed: a PIR needs a body to cross a zone boundary, a radar chip needs only the sub-millimetre motion a living body produces on its own.

Radar has false alarms of its own

None of this makes a mmWave sensor immune to false triggers; it simply moves the problem. Aqara's installation guide for its FP2 presence sensor warns installers directly that curtains, fans, wind-stirred plants and hanging clothes are common causes of a false alarm if the sensor faces them, and it requires each one to be marked as an interference source during setup, with the room cleared of people and pets while that step runs. A separate step in the same guide handles the sensor's own reflections: an edge boundary marked along each wall, because reflections off walls and other surfaces create phantom targets that the algorithm has to be told to ignore. The exit side of detection is not instant either. Aqara's documentation states that the transition from presence to absence typically takes six to thirty seconds of continuous non-detection. That debounce is not a fault; it is deliberate. It is still worth knowing before wiring the sensor to a light switch that people expect to go dark the moment they leave.

Configuring zones instead of one blob of coverage

What separates a mmWave sensor from a PIR in practice is that its coverage is not a single cone; it is a map. The FP2's floorplan is divided into a 320-cell grid, sixteen columns by twenty rows, each cell 0.5 by 0.5 metres, and the installer can draw up to thirty named zones on top of it, each given a type so the algorithm tunes itself differently for a reading chair than for open floor. Wall-mounted, the sensor covers a 120° horizontal angle out to an eight-metre radial distance and a six-metre width; ceiling-mounted at 2.8 metres it covers a two-metre-radius fall-detection zone instead. Sensitivity is a deliberate trade set per zone: run it high in a static room such as a bedroom, where a slower response would miss someone settling in, and low in a dynamic area such as a hallway, where anything more sensitive fires on every passer-by. Proximity thresholds for approach and departure events follow the same logic, set in three fixed steps rather than a single factory number.

The power ceiling that keeps these sensors short-range

A presence sensor covers a room and not a floor because of a transmit-power ceiling written into radio regulation, not an engineering preference. The US and the EU both segment the 57 to 64 GHz band and cap output differently depending on which slice a device uses.

The atmosphere does much of the rest of the containing on its own. Oxygen absorption near 60 GHz reaches roughly 16 dB per kilometre at ground level, according to ETSI's own technical annex on atmospheric attenuation, which is one reason this band was opened to short-range sensing rather than reserved for anything needing real distance. A sensor built to these limits was never going to cover a warehouse. It was built to cover a room, reliably, which is a different design target from covering ground.

Where the extra cost earns its place

The rooms worth the premium are the ones where a person sits still for a meaningful stretch without producing the kind of gross movement a Fresnel lens needs: a bathroom or toilet, a desk where someone reads a screen without gesturing, a nursery where a light cutting out mid-feed is exactly the failure a sensor is there to prevent. Because the underlying signal for presence and the underlying signal for breathing rate are the same chest-wall motion, a radar sensor bought for lighting can, on hardware such as Infineon's own reference chip, double as a coarse vital-sign check with no extra hardware fitted.

A corridor, a stairwell, a garage and a closet need none of this sophistication. Nobody sits still in a corridor. A plain PIR at a fraction of the price does the job a mmWave sensor was never built to improve on.