“Flicker-free” appears on more datasheets than any other phrase in lighting, and none of them define it. The number that matters is percent flicker at the lowest dim level you intend to use, measured with a meter sampling above 2 kHz, and most fixtures that claim the label are only clean at 100% output. This page sets out the three metrics that carry weight in a specification — percent flicker, flicker index and SVM — the IEEE 1789 and EU limits behind them, the driver choices that decide the result, and the wording that puts a testable requirement into a purchase order. Flicker free lighting is a measurable property, and it can be bought as one.
1. Why “flicker-free” is not a specification
There is no standard definition of the phrase. A manufacturer can print it on a box for a driver that holds 4% modulation at full output and 35% at 5% dim, and be within the ordinary use of the words. Nothing in that claim is false; nothing in it is useful to a specifier either.
The gap is always the same: the claim describes the easiest operating point. At 100% the control loop is comfortable and modulation is low. At 1% the loop gain falls and a driver that switches into burst mode to hold low output flickers badly. That is the condition a guestroom sees at 2 a.m.
So ask for four things instead of an adjective. The percent flicker at 100%, 50%, 10% and 1% output. The modulation frequency. The meter and sampling rate used, because a meter sampling at 100 Hz cannot see a 400 Hz artefact. And the document the figures are measured against, whether that is IEEE 1789, the EU ecodesign rules or the project’s own schedule.
A workable project definition, used in hospitality work: percent flicker below 5% and flicker index below 0.1 at every dim level down to 1%, with SVM at or below 0.4. That is testable, it is comparable between suppliers, and it catches the fixture that looks good on a bench at full output.
2. Percent flicker and flicker index
Pourcentage de scintillement is the peak-to-trough depth: (max − min) / (max + min) × 100. If a fixture swings between 1,050 and 950 lux, the result is 100 / 2,000 = 5%. It is simple enough to compute from a waveform screenshot, which is why it survives.

Indice de scintillement is the area of the waveform above its mean divided by the total area under the curve, over one cycle. It runs from 0 to 1 and it captures shape. A symmetrical sine at 5% depth scores about 0.016; a narrow 1 ms spike every 10 ms at full depth scores far higher for the same peak-to-trough percentage. That difference is why both numbers belong in a specification.
The weakness of percent flicker is that it ignores duty cycle. A 1% duty pulse at 100% depth and a smooth 5% sine can report the same average light and behave completely differently. The eye integrates over 10–20 ms; a camera does not.
For reference, a well-filtered driver shows 1–4% at full output and 2–6% at 10%. A cheap driver with a small capacitor shows 20–35% at full output. At 1%, drivers without a stable low-end mode jump to 50–100% depth, which reads as a pulse rather than a shimmer.
| Indicateur | Formula | Target | Ce qu’il ne prend pas en compte |
|---|---|---|---|
| Pourcentage de scintillement | (max − min) / (max + min) × 100 | < 5% at every level | Waveform shape and duty cycle |
| Indice de scintillement | Area above mean / total area | < 0.1, ideally < 0.05 | Frequency; a slow wobble scores low |
| SVM | Weighted sum of frequency components | ≤ 0.4 from September 2024 | Depth at a single frequency in isolation |
| Frequency | Fundamental of the modulation | > 3 kHz if PWM is used | Whether the eye or a camera resolves it |
3. SVM and the IEEE 1789 recommendation
Both numbers above describe the light. Neither describes what a person perceives, because the eye is far more sensitive to modulation at 5 Hz than at 500 Hz. SVM — short-term flicker visibility measure — sums the frequency components of the waveform, weights each by the eye’s sensitivity curve, and returns a single number where 1.0 is roughly the threshold of visibility for a typical observer. An SVM of 0.4 means most people will not see it; 0.2 is comfortable in a dim room.
That weighting explains a limit that looks loose on paper. At 100–200 Hz the eye is several times less sensitive than at 10 Hz, so SVM ≤ 0.4 at 100 Hz permits a depth that would be unacceptable at 10 Hz. A driver with 30% ripple at 100 Hz can pass SVM while failing a 5% percent-flicker target, which is why a specification needs both limits.
IEEE 1789-2015 takes the same physics and gives recommended practice rather than a pass mark: below 90 Hz, modulation depth should stay under 8%; through 90–1,250 Hz the limit relaxes linearly as frequency rises. It is guidance, and it is quoted constantly in specifications because it is the clearest published statement of the trade-off.
European law is stricter in practice. The ecodesign rules for light sources, 2019/2020 as amended by 2021/341, require SVM ≤ 0.4 and Pst LM ≤ 1.0, with the tighter SVM value applying from September 2024. A fixture that met the earlier 0.9 threshold may now fail, and the declaration has to be supported by a test rather than a datasheet note. The compliance documentation for a project should carry that test report next to the CE declaration.
4. Driver design choices that reduce flicker
Flicker is decided in four places inside the driver, and each one has a price.
Output capacitance. The bulk capacitor decides how much 100 Hz ripple reaches the LED. Doubling it halves the ripple depth and costs board space plus a larger, more expensive part. This is why a 12 W retrofit driver and a 40 W architectural driver with similar output can differ by a factor of five in percent flicker.
Topology. A two-stage design — power factor correction followed by an isolated DC-DC stage — rejects mains ripple far better than a single-stage design that combines both jobs in one switching block. Single-stage is cheaper and smaller; it is also where most 30% ripple figures come from.
Dimming method. PWM above 3 kHz, filtered PWM, or CCR. Each keeps the die at a stable operating point or removes the chopping entirely. A PWM driver at 200 Hz will fail any camera, whatever its percent flicker figure says, because the artefact is temporal rather than amplitude-based.
Low-end behaviour. Many drivers hold 1% output by skipping pulses or bursting, which produces deep, low-frequency modulation exactly where it is most visible. Ask whether the low end is continuous conduction or burst mode. If the answer is burst, that is the flicker you will be called back to fix.
| Design choice | Effect on flicker | Cost |
|---|---|---|
| Larger output capacitor | Halves ripple depth for each doubling | Board space, unit cost |
| Two-stage topology (PFC + DC-DC) | Cuts 100 Hz ripple to single digits | Efficiency loss, size |
| PWM > 3 kHz or filtered PWM | Removes camera banding | Filter components, slightly lower efficiency |
| CCR (amplitude) dimming | No chopping at all; smooth DC | Colour shift at low levels, range limited to about 10% |
| Continuous low-end control | Avoids burst-mode pulsing at 1% | Control loop cost, tighter tolerances |
| 105 °C capacitors, derated | Holds the figures over life, not only at day one | Part cost |
One point that rarely makes it onto a datasheet: the figures drift with age. An electrolytic capacitor loses capacitance as it dries, so a driver measured at 3% at installation can be at 15% in year five. Specifying 105 °C parts and derating ripple current extends that curve; it does not remove it, so ask for an end-of-life figure as well as a beginning-of-life one in the driver engineering review.
5. Testing with a flicker meter
Sampling rate is the first thing to check on any meter. A meter sampling at 1 kHz resolves a 100 Hz waveform and misses the 3 kHz component that ruins video. For specification work, use 2 kHz minimum and 10 kHz preferred, with a photodiode head rather than a lux meter cell, which filters the signal it is trying to measure.

Test at the operating points that matter, not only at full output: 100%, 50%, 10% and 1%, plus the off state. Let the fixture warm up for 30 minutes, because driver temperatures settle and the control loop shifts slightly. Measure in a dark room with no other source, at the surface the light lands on rather than at the lens, and hold the meter steady — movement shows up as modulation.
If no meter is available, a phone at 240 fps is better than nothing. Point it at a white wall, not at the lamp, and look for banding or a pulse. It will not give you a number, and it will not catch modulation above about 120 Hz, but it settles the “is it me or the light” question in ten seconds.
| Test condition | Why it matters |
|---|---|
| 1% output | Where burst mode and loop instability appear |
| 10% output | The usual hotel night scene and the most common complaint level |
| Low line (207 V) | Reveals drivers that only just hold regulation |
| 30-minute warm-up | Driver temperature changes the control loop and the capacitor’s behaviour |
| Sampling ≥ 2 kHz | Below that, the meter cannot see the artefacts that matter on camera |
6. What to put in the PO
A purchase order is the only document in the chain that a supplier has to answer. Put the numbers there, with the test method, and the argument about “flicker-free” disappears.
Useful clauses: percent flicker below 5% and flicker index below 0.1 at 100%, 50%, 10% and 1% of output, measured with a meter sampling at 2 kHz or above; modulation frequency above 3 kHz where PWM dimming is used; SVM at or below 0.4 with a report to EN 12464 or the ecodesign method; dimming range 1–100% defined as percentage of rated current, not of perceived brightness; and a flicker test report per fixture family, not per shipment.
Add the compatibility requirement while you are there: the driver model and its tested dimmer list, locked and named, with no substitution without a sample test. That single clause prevents most of the flicker complaints that arrive after handover, because it stops a purchasing team from swapping in a cheaper driver halfway through a project.
Finally, tie money to it. A 5% retention released against the flicker report and a sample approval is normal on hospitality work and costs nothing to write. It also concentrates the supplier’s attention on the low end of the dimming range, which is where the test will fail if it is going to.
| PO clause | Wording to use | What it prevents |
|---|---|---|
| Flicker limit | Percent flicker < 5% and index < 0.1 at 100 / 50 / 10 / 1% | Clean only at full output |
| Frequency | PWM above 3 kHz, or CCR with no PWM | Camera banding at events |
| Perceptual limit | SVM ≤ 0.4, report with the delivery | Failing the 2024 EU threshold |
| Dim range | 1–100% of rated current, continuous at the low end | Burst-mode pulsing at 1% |
| Locked driver | Named driver model, no substitution without sample test | Mid-project value engineering |
The cost of writing those five lines is an afternoon. The cost of skipping them is a snagging list that runs for a year, because flicker at 1% is the kind of fault that only shows up when the room is dark and the guest is trying to sleep.
Étape suivante
If a project needs flicker data rather than a datasheet claim, ask for the measurement. The team can supply per-family test reports at 100%, 50%, 10% and 1%, at nominal and low line voltage, with the meter and sampling rate stated — see the driver and test capability overview — and will review a draft PO clause against the fixtures actually specified. Sample fixtures for an on-site meter check are normally available before the order is placed.






