Flicker meter for lamps and screens

Because the measurement depends on how fast the sensor reads its rows, and nothing in a browser reports it. It cannot be worked out from the frame rate either: between frames the sensor idles for a blanking interval whose length is not published. A lamp running off the mains solves it, because it flickers at exactly twice the mains frequency and the grid holds that frequency to a few thousandths of a per cent. So the cheapest bulb in the house is the calibration standard. It is stored per device, so you do it once.

Because a camera row does not take a snapshot of the light — it collects light for the whole exposure time and reports the total. That is a smoothing filter, and it does not fade gently: at 1/500 s a 500 Hz flicker is canceled exactly. Measured against known ground truth, a lamp modulating 34% reads 34% at 1/4000 s, 30% at 1/500 s, and 5% at the 1/120 s a phone picks indoors. A meter that ignores this hands a strobing lamp a clean bill of health. Here the effect is divided back out harmonic by harmonic, and where the exposure has erased a harmonic outright the tool says so rather than pretending.

Probably both, and the reason is worth knowing. Screen PWM often runs in the kilohertz, where a single frame of the camera contains hundreds of cycles but the exposure smooths them away almost completely. Check the exposure line under the camera: if it is longer than about 1/2000 s, fast flicker is invisible no matter how deep it is. Turn the screen to full brightness first — most panels stop dimming by PWM there — then step the brightness down and measure at each level, which is how the effect actually shows itself.

They are the IEEE 1789-2015 limits, and they are a pair because neither frequency nor depth means anything alone. The lower curve is the no-observable-effect level: 1% of the frequency below 90 Hz, 3.33% above it. The upper is the low-risk line: 2.5% and 8% of the frequency respectively. A lamp at 100 Hz is in the clear below 3.3% depth and low-risk below 8%; the same 8% at 50 Hz is well over the line. Above about 1250 Hz the low-risk line passes 100% and the standard stops constraining anything.

Because we do not have the threshold curve those measures are built on. The stroboscopic visibility measure of IEC TR 63158 weighs every Fourier component against a sensitivity table published in a standard we have not bought, and a plausible-looking SVM computed from a guessed table would be worse than none at all — it is exactly the sort of number people quote in a complaint. IEEE 1789 is implemented because its limits are public, unambiguous and checkable.

Because the rows of one frame are the samples, so a short frame is a short recording. At 1080 rows a frame lasts about twenty milliseconds, which is two cycles of a 100 Hz lamp — barely enough to place the frequency, and in testing it wandered by up to 18%. At 1440 rows the same measurement came within 1.7%, and at 2160 within 0.4%. The tool asks the camera for the tallest frame it has; if yours is short, the photo route is the way round it, since a still is usually full sensor height.

Yes, with one caveat: the bands are measured across the lit part of the frame, so the light has to cover enough rows. A distant point source occupies a handful of pixels and gives nothing. Get the light large in the frame — or photograph the pool of light it casts on a wall or the road, which flickers identically and fills the view.