Перейти к содержимому
Flickerfield

Все записи

Движок19 августа 2026 г. · 4 мин чтения

Why frame-locked timing matters

A phase accumulator on the display's own frame clock, what a 60 Hz panel can and cannot hit exactly, and why the app shows you its dropped frames.

If you only read one technical post here, read this one. Everything else the app does rests on a single claim: when it says 10 Hz, it means 10.000 Hz, and it will tell you when it cannot.

Why timing is the interesting variable

The 2023 PLOS ONE study compared rhythmic flicker with arrhythmic flicker matched for average frequency. Same rate on average, same amount of light, the intervals deliberately irregular. Reported geometric imagery peaked at 10 Hz rhythmic and was strongly reduced in the arrhythmic condition.

Take that seriously and an engineering consequence falls out. If regularity is part of what participants were responding to, then an app whose intervals wobble is not delivering a slightly worse version of the same stimulus. It is delivering a partially arrhythmic one. Jitter is not a polish issue. It is the independent variable.

What a phone display can actually do

A screen does not emit light continuously. It refreshes: 60 times a second on most phones, 120 on a fast panel, sometimes a variable rate in between. Every change you can make to what is on screen happens at one of those refresh instants. There is no in-between.

So the set of flicker frequencies a display can produce exactly is fixed by its refresh rate and the number of frames you spend on each half-cycle:

RefreshFrames per cycleExact frequency
60 Hz6 (3 on, 3 off)10 Hz
60 Hz8 (4 on, 4 off)7.5 Hz
60 Hz125 Hz
120 Hz1210 Hz
120 Hz1012 Hz

Ask a 60 Hz panel for 9 Hz and the arithmetic stops being clean: 60 ÷ 9 is 6.667 frames per cycle. The engine cannot show two-thirds of a frame. It has to choose between cycles of 6 and 7 frames and distribute the error, which means the light is averaging 9 Hz across a second rather than producing it.

That is a legitimate way to run, and it is what the app does for frequencies that do not divide, but you deserve to know which one you are getting. Hence the frame-lock badge: it is on when your chosen frequency divides your display's refresh rate exactly, and off when the engine is averaging.

The phase accumulator

The naive implementation of a flicker is a timer: "turn on, wait 50 ms, turn off, wait 50 ms". This is the wrong shape for the problem. Timers fire late, timers fire early, and every late fire adds a little error that never comes back. Ten minutes of that and the light has drifted somewhere unrelated to where it started.

The approach the app uses instead is a phase accumulator driven by the display's own frame callback: CADisplayLink on iOS, Choreographer on Android. Both hand you the timestamp of the frame that is about to be drawn. On each frame the engine advances a phase value by frequency × elapsed time, wraps it at 1.0, and decides from the phase, not from a counter and not from a timer, whether this frame should be lit.

Three properties follow, and they are the whole argument for doing it this way:

  1. Error does not accumulate. Phase is derived from the actual elapsed time, so a late frame produces one frame of error, not a permanent offset.
  2. Ramps are exact. When a program slides from 10 Hz to 6 Hz over two minutes, the instantaneous frequency is whatever the ramp says it is at that moment, and the phase keeps advancing continuously through the change. There is no discontinuity at the segment boundary.
  3. Duty cycle is honest. "On for 30% of the cycle" is a comparison against the phase value, so it means the same thing at 6 Hz and at 18 Hz.

None of this is exotic. It is the standard way to build an oscillator. It is simply not what you get from a JavaScript timer in a web view, which is why the engine is native on both platforms.

Dropped frames, and why they are on the summary screen

Phones do other things while your session runs. A notification arrives, the system decides to re-layout something, a background process grabs the GPU, the panel drops to a lower refresh rate to save power. Any of those can cost a frame. A dropped frame in a flicker engine is a cycle that was slightly the wrong length.

The app counts them and shows the count when the session ends, next to the frame-lock status and the frequency you actually ran. This is deliberately unflattering information. A session with a high dropped-frame count was not the stimulus you asked for, and if you are keeping a log of what you tried, you need to be able to see that rather than wonder why one session felt different.

If you want the cleanest run your phone can give: close other apps, turn off Do Not Disturb exceptions, unplug from charging if the phone is hot, and pick a frequency that shows the frame-lock badge on your display.

Precision is checkable, which is exactly why it is the only thing we advertise. What the light then does for you is a different question, and one we make no promises about.

Источники

  1. Amaya et al. (2023), effect of flicker frequency and rhythmicity, PLOS ONE 18(4): e0284271
  2. Apple Developer documentation: CADisplayLink
  3. Android Developers reference: Choreographer