LED PWM Dimming and Flicker: What Percent Flicker, Flicker Index and SVM Measure

LED PWM flicker is the visible price of a dimming method, not a defect: pulse-width modulation never makes a lamp dimmer in the optical sense, it switches the emitter fully on and fully off and leaves the observer’s visual system to average the two states. How much of each cycle is dark, how often the dark periods repeat, and how the waveform distributes energy across frequencies decide whether a person sees flicker, sees a stroboscopic effect, sees nothing but still measures something, or sees nothing until a camera produces bands that were never visible to the eye.

The numbers belong on the table before the mechanism. A 100 Hz square wave at 50% duty cycle and 100% modulation measures 100% percent flicker and a stroboscopic visibility measure (SVM) of about 4.9 against a threshold of 1. The same waveform at 1 kHz still measures 100% percent flicker and falls to about 0.88 SVM. Nothing changed except the switching rate. That is why a datasheet line reading “flicker: low”, or even “flicker: 5%”, carries almost no information without a frequency and a waveform attached to it.

Quick answer

PWM dimming chops the current through an LED into pulses. Average current, and therefore average light output, scales with duty cycle; peak current during each pulse does not change. The light is never actually reduced, only interrupted, and the interruption rate is the dominant variable in whether anyone notices.

Three conclusions follow. First, percent flicker and flicker index describe the shape of one cycle and say nothing about frequency, so a 100 Hz square wave and a 10 kHz square wave with the same duty cycle both measure 100% percent flicker. Second, SVM was built to fix that: it decomposes the waveform into harmonics, weights each by a measured visibility threshold, and returns a number where 1 means “just visible to an average observer, 50% of the time”. Third, none of these metrics establishes safety, comfort or legality. They establish visibility, which is a narrower claim than the marketing language around “flicker-free” implies.

The rules that survive contact with real products are narrower. A lamp whose LED PWM flicker appears only when a phone camera is pointed at it is not necessarily faulty; it is modulating faster than the eye integrates and slower than the camera’s readout. A lamp with 100–200 Hz dimming and 100% modulation measures 100% percent flicker and an SVM well above 1, whether or not the person holding it can see anything.

Question	Metric that answers it	Metric that does not
How deep is the modulation within one cycle?	Percent flicker, modulation depth	SVM, PstLM
How much of the cycle is spent near the peak?	Flicker index	Percent flicker
Will an average observer see a stroboscopic effect?	SVM	Percent flicker, flicker index
Will an average observer see flicker directly?	PstLM	SVM
Will a camera band?	None of them; needs readout time and exposure	All of them

What pulse-width modulation actually does to an LED

An LED is a current-driven device: light output tracks forward current rather than forward voltage, which is why every practical lamp contains a driver that regulates current. The topologies that do this are covered separately in Automotive LED Drivers Explained; this article assumes the regulated current exists and asks what happens next.

Once current is regulated, dimming has two honest implementations. Reduce the current continuously, often called analog or linear dimming, or leave the current at its set point and interrupt it. Pulse-width modulation is the second. The switch is on for t_on out of a period T, and the ratio is the duty cycle D = t_on / T. Average current follows arithmetically: average current = D × peak current, and average luminous flux is approximately D × peak flux.

The approximation in that second expression is where casual explanations break. LED flux is close to linear in current but not exactly so, because junction temperature and current density both change efficacy. For flicker metrics the approximation is harmless, since the metrics come from the measured optical waveform rather than the current; for comparing two lamps it is not, which is why “50% duty cycle therefore 50% of the lumens” is an estimate rather than a measurement.

The optical waveform follows the current closely. The U.S. Department of Energy’s solid-state lighting programme explains why: LEDs and OLEDs have very little thermal or phosphor persistence, so they respond almost instantaneously to the current the driver delivers, and a driver can deliver current in an unlimited variety of shapes, modulation depths, duty cycles and frequencies (DOE Flicker Research). An incandescent filament cannot do this. Its thermal mass smooths the ripple, which is why the same 100 Hz mains modulation that is invisible on a halogen lamp is a design problem on an LED.

What the eye does with the result is the other half. DIAL’s laboratory describes the process precisely: with PWM the LED is switched off completely and then fully on again, and because the human eye and a standard lux meter both respond more slowly than the switching, the process registers as a reduction in luminous flux rather than as a sequence of flashes (DIAL). The lamp is dim only in the sense that an average is dim, which is why an instrument and an observer can describe the same lamp completely differently without either being wrong.

Why a vehicle lamp dims with PWM instead of a resistor

Linear dimming wastes the difference as heat, and in a sealed aluminium housing with no airflow that heat raises the temperature of the LEDs the housing exists to cool. The relationship between temperature, output and service life is the subject of the thermal management article; the point here is that PWM moves the loss into the switching device and removes the need for a second current set point.

The economics are documented. LED professional notes that PWM is a common dimming method in LED drivers, that it introduces moderate-frequency components typically at 400–800 Hz, and that typical operating dimming frequencies run from 200 to 400 Hz, up to 800 Hz (Flicker: Standards and Test Methods). DIAL puts the usual luminaire range at 100–400 Hz and notes that cost pressure pushes simple drivers toward the low end. Werner states the trade-off from the other direction: higher PWM frequencies cost more and lose efficiency, which is why cheap drivers sit low and why flicker returned with LED lighting after electronic ballasts had pushed fluorescent lamps past 40 kHz (Flicker Beyond Perception Limits).

Vehicle lamps have a specific version of that trade-off. A daytime running lamp, a tail lamp with bright and dim states, and a work light with two or three output levels all need one emitter to produce several brightness levels from one driver. On a single regulated current, chopping is the cheapest route. That is a design pattern rather than a rule, and the sources opened here do not include a vehicle-lamp dimming survey, so the defensible statement is that PWM is a well-documented method in LED lighting generally and that multi-level automotive lamps are exactly where it is economically attractive.

Duty cycle, frequency and average current: the three numbers that define a PWM waveform

Three parameters fix a rectangular PWM waveform completely, and only two are usually published.

Parameter	What it fixes	What it does not fix
Frequency f	How often the cycle repeats; period T = 1/f	Brightness, modulation depth
Duty cycle D	Average current and average flux, as D × peak	Colour, peak current
Peak current	Junction condition during the pulse, roughly spectrum and heat load	Average brightness

Two cases make the arithmetic concrete. At 200 Hz the period is 5 ms, so a 10% duty cycle means 0.5 ms on and 4.5 ms dark. At 500 Hz the period is 2 ms, so the same duty means 0.2 ms on and 1.8 ms off. Both deliver the same average current if the peak is the same, and both have a minimum of zero. They differ only in repetition rate, which two of the four metrics here cannot see at all.

What percent flicker measures, and what it ignores

Percent flicker, also called flicker percentage or modulation depth, compares the brightest and dimmest points in one cycle. LED professional states the definition in words: it is the measure of the maximum light versus the minimum light in a cycle, it accounts only for those two points, it does not differentiate between waveforms, and it is the simplest form of flicker to determine (Flicker: Standards and Test Methods).

As arithmetic, that definition becomes percent flicker = 100 × (max − min) / (max + min). The expression is worth reading twice, because it is not the ratio of minimum to maximum. Dividing by the sum rather than by the maximum normalises the result so that a waveform swinging between full output and complete darkness scores exactly 100%, not 50%, and so that any non-negative waveform stays bounded between 0% and 100%. The same arithmetic appears in the IEEE 1789 discussion as Michelson contrast, which DIAL reports as close to 100% for rectangular PWM, about 10% for a halogen lamp on mains, and near zero for an LED on direct current.

Waveform	Max	Min	Percent flicker
DC, no modulation	100	100	0%
Halogen on 50 Hz mains, typical	100	82	10%
Sine, fully modulated	100	0	100%
Square, 50% duty	100	0	100%
Square, 10% duty	100	0	100%
Square, 10% duty at 10 kHz	100	0	100%

The last three rows are both the point of the metric and its limitation. Percent flicker cannot distinguish a 10% duty pulse from a 50% duty pulse, and it cannot distinguish either from a version running a hundred times faster. It answers one question: how far does the light travel between its extremes within a cycle.

A second limitation is less obvious and more consequential. Because the metric is computed from two values, it is extremely sensitive to measurement bandwidth: a photodiode with a slow amplifier, or a lux meter averaging over milliseconds, reports a minimum higher than the true one and therefore a percent flicker lower than the true value. The standard method removes that ambiguity with sampling intervals no greater than 50 microseconds, at least 20 kHz, over a record of at least two seconds at 25 °C ± 2 °C. An LED PWM flicker number quoted without a bandwidth is not comparable with one quoted with a bandwidth, which is why the arrow in a supplier conversation should always point at the test method.

What flicker index adds: the shape of the waveform

Flicker index answers what percent flicker cannot: given the same extremes, how much of the cycle is spent near the top? The definition is an area ratio. LED professional describes it as a metric that considers the area of the waveform above and below the average light output, requires more calculation than flicker percent, and takes waveform shape into account.

As arithmetic, flicker index = (area above the average) / (total area over one cycle). The ratio runs from 0 for a steady light toward 1 for a very narrow pulse. Three cases are worth computing because they show how differently the two legacy metrics behave on waveforms a lamp can actually produce.

Waveform, period normalised to 1	Mean	Area above mean	Flicker index
DC	1.0	0	0
Sine, min = 0	0.5	0.5	0.318
Square, 50% duty	0.5	0.25	0.500
Square, 10% duty	0.1	0.09	0.900

The last column is computed here from the definition, not quoted from a standard. For a fully modulated sine the area above the mean is 2 in radian measure and the total area is 2π, giving 0.318. For a 50% duty square the mean is 0.5, the area above it is 0.25 and the total area is 0.5, giving 0.5. For a 10% duty square the mean is 0.1, the area above it is 0.09 and the total area is 0.1, giving 0.9.

The pattern is the point. Percent flicker reports 100% for all three modulated waveforms; flicker index reports 0.318, 0.500 and 0.900 for the same three. If the question is how much of the time the lamp is actually emitting, flicker index is more informative. If the question is whether a person will see anything, neither is informative, because neither contains a frequency term.

LED professional states directly that these metrics do not quantify temporal light artefacts correctly and objectively, that they are not selective between flicker and stroboscopic effect, and that they do not account for frequency-dependent sensitivity or wave shape, so additional verification is required. The encyclopaedic summary is blunter: modulation depth, flicker percentage and flicker index are widely used, and none predicts actual perception, because perception depends on depth, frequency, wave shape and duty cycle together (Wikipedia, Flicker (light)).

Why percent flicker and flicker index disagree about the same lamp

The disagreement is largest exactly where vehicle lamps live: short, narrow pulses.

Consider a tail lamp that produces its dim state by running the emitter at full current for 10% of every cycle. Its percent flicker is 100%, the maximum the metric can report, because the light reaches full output and reaches zero within each cycle; its flicker index is 0.9. A second lamp produces its dim state by running the emitter at 20% current continuously with a 5% ripple from an imperfect filter. Its percent flicker is about 9% and its flicker index is negligible. On a datasheet both are described as dimmed. The first will band on camera and produce a stroboscopic effect on a rotating fan; the second will do neither.

The corollary is that neither metric can rank lamps that differ in frequency. A 100 Hz lamp at 30% modulation and a 1 kHz lamp at 60% cannot be ordered by percent flicker, because the lower-modulation lamp operates where the eye is most sensitive to the stroboscopic effect.

Where visible flicker ends and the stroboscopic effect begins

The boundary is a frequency rather than a modulation depth, and the literature converges on roughly the same number from several directions. DIAL puts the flicker fusion frequency at approximately 80 Hz and describes stimulus bands above it as no longer consciously visible but still detectable by the visual system. LED professional states that flicker is visible up to 70–80 Hz and that above this the discussion is about the stroboscopic effect. The encyclopaedic summary separates the artefacts by name: up to about 80 Hz the phenomenon is visible flicker, from about 80 Hz to 2000 Hz it is the stroboscopic effect, and from 80 Hz to about 6500 Hz a third overlapping effect, the phantom array, appears during eye movements (Wikipedia, Stroboscopic effect).

Frequency band	What an observer can experience	Typical source in a vehicle lamp
Under 8 Hz	Slow, obvious pulsing	Fault, not design
8–30 Hz	Photosensitive seizure risk band, most provocative near 15–20 Hz	Rare in lighting; seen in malfunctioning drivers
30–80 Hz	Directly visible flicker, reported as annoying	Very low PWM frequencies
80–2000 Hz	Stroboscopic effect on moving objects, not consciously visible as flicker	Common PWM dimming range
2–6.5 kHz	Phantom array during saccades	Higher-quality PWM dimming
Above 3 kHz	No evidence of effects found in the sources reviewed here	Recommended design target

Three cautions belong on that table. The boundaries are approximate and depend on modulation depth, waveform, light level and observer, so a lamp at 70 Hz with 5% modulation is not equivalent to one at 70 Hz with 100%. The 8–30 Hz seizure band is a population risk rather than a comfort issue: DIAL cites an incidence of about 0.1% for photosensitive epilepsy, identifies 15–20 Hz as the most provocative range, and notes that the 1997 Pokémon broadcast at roughly 12 Hz produced around 700 documented cases, 650 of them children. And the dismissal of everything above 80 Hz as invisible and therefore irrelevant is exactly the assumption the SVM work was designed to test.

The upper limit is softer than it sounds. Wikipedia’s summary notes that steady foveal vision rarely detects flicker above 90 Hz but that during saccades it can be perceived beyond 1 kHz, and records Werner’s point that most viewers stop reporting a flicker impression above roughly 120 Hz. Neither statement says a 400 Hz waveform is doing nothing: Werner’s review finds no suggestion that frequencies above 800 Hz affect humans, but enough evidence that flicker up to 400 Hz is not harmless under long-term exposure (Flicker Beyond Perception Limits).

What SVM measures that the older metrics cannot

The stroboscopic effect visibility measure starts from a different premise. Instead of describing a cycle’s shape it asks how visible the stroboscopic artefact would be to an average observer, and it answers by decomposing the waveform into harmonics and weighting each by a measured threshold.

Both the threshold function and the summation are published. For sinusoidal modulation the visibility threshold as a function of frequency f is T(f) = 2.865 × 10⁻⁵ × f^1.543 + 0.225. The measure is the 3.7th root of the sum of weighted harmonics: SVM = ( Σ (C_m / T_m)^3.7 )^(1/3.7), where C_m is the relative amplitude of the m-th Fourier component of the relative illuminance, relative to the DC level, and T_m is the threshold at that component’s frequency (Wikipedia, Stroboscopic effect).

Two features do all the work. The exponent 3.7 makes the measure behave like a maximum rather than an average, so one dominant harmonic decides the result. The division by T_m makes high frequencies cheap: the threshold rises faster than any realistic harmonic amplitude, so a 5 kHz component contributes almost nothing even when its absolute amplitude is not small.

The acceptance criterion is equally specific. If SVM equals 1, the modulation produces a stroboscopic effect that is just visible, meaning an average observer detects the artefact with a probability of 50%. Above 1 the probability exceeds 50%, below 1 it falls short. The threshold does not guarantee acceptability, because some applications tolerate more and others require less. The meter specification and test method are published as IEC TR 63158, the terminology and models as CIE TN 006, and acceptance guidance for applications as NEMA 77-2017.

The applicability limits matter on a vehicle. SVM as specified applies to typical indoor light levels above 100 lx and to moderate movement of the observer or a nearby handled object below 4 m/s. A work light illuminating a workshop at 200 lx fits that envelope. A driving lamp at 50 m does not, because the angular velocities are far higher and the sources reviewed here do not extend the validated model to that case. Using SVM to compare two work lights is defensible; using an SVM figure to predict what a driver sees at speed is not.

Working the arithmetic on one waveform shows how much information the frequency term carries. Take a 50% duty square wave with 100% modulation, the textbook output of a PWM dimmer at half brightness. Its Fourier components relative to the DC level are C1 = 4/π = 1.273, C3 = 4/(3π) = 0.424 and C5 = 4/(5π) = 0.255. Applying the published threshold function:

Repetition frequency	T at fundamental	C1 / T1	SVM, computed here
100 Hz	0.260	4.90	4.9
500 Hz	0.644	1.98	1.98
1000 Hz	1.444	0.88	0.88
2000 Hz	3.786	0.34	0.34

The 100 Hz result is the check on the method, because it reproduces the value published with exactly this waveform, SVM = 4.9. The rest of the column follows from the same function at other frequencies and is computed here rather than quoted. Percent flicker stays at 100% across all four rows while SVM falls by a factor of fourteen and crosses the visibility threshold just below 1 kHz. A lamp that is unacceptable at 100 Hz becomes acceptable at 1 kHz with no change to its modulation depth, duty cycle or optics. Only the switching rate changed.

That is why the 3 kHz design target is not arbitrary. With the fundamental already at 3 kHz the threshold is about 6.9 and the fundamental ratio near 0.19, which places the measure far below unity before any harmonic is counted. The cost is the trade-off Werner describes: higher switching frequency means more switching loss, more expensive magnetics and, in a sealed housing, more heat to remove.

What IEEE 1789 does and does not give you

IEEE 1789-2015 is the document most often cited in this field, and its current status is the part almost nobody quotes. The IEEE Standards Association record describes it as IEEE Recommended Practices for Modulating Current in High-Brightness LEDs for Mitigating Health Risks to Viewers, approved by the board on 26 March 2015, published on 5 June 2015, ANSI-approved on 7 September 2016, and now carrying the status Inactive-Reserved with an inactivation date of 26 March 2026 (IEEE SA record). A document in that state has not been withdrawn, but it is no longer an active standard, and a supplier citing it in 2026 as current practice is describing a ten-year-old recommendation.

The scope is narrower than the citations suggest. The abstract lists a definition of the concept of modulation frequencies for LEDs, a discussion of their application to LED lighting, a description of applications where modulation frequencies pose possible health risks, a discussion of dimming LEDs by modulating the frequency of the driving current or voltage, and recommendations for modulation frequencies to help protect against known potential adverse health effects. What the abstract does not contain is a single number.

The numbers exist but are in the paid text, and the open summaries reviewed here describe them qualitatively rather than reproducing them. DIAL’s account says the recommendation plots modulation against frequency, that concrete values are frequently not mentioned, that the flicker index is not taken into closer consideration, and that the recommendations are conservative. LED professional makes the same point from the standards side: IEEE 1789 is not a standard, and its recommendations are conservative to the point that some traditional lamps cannot meet them. This article therefore does not restate a numeric IEEE 1789 risk band, because none of the sources it could open does so; a reader who needs the bands needs the document.

What can be said with the sources in hand is structural. The recommendation is organised around frequency and modulation depth, it treats flicker and the stroboscopic effect as different problems, and one widely repeated reading of it is that above about 3 kHz no evidence of effects has been found, which is the origin of the 3 kHz design rule. None of that is a pass or fail criterion, and a claim that a lamp meets IEEE 1789 without a test report naming the document, the revision and the measured waveform should be treated as marketing.

The limits that are actually written down

If the influential recommendation is qualitative in its open form, several other instruments are not. They come from general lighting rather than vehicle lighting, which is the most important caveat in this section, but they are the numbers that exist.

Instrument	What it constrains	The number	Status in the sources
California Title 24, Joint Appendix 10	Flicker of high-efficacy sources	Under 30% flicker below 200 Hz	Low flicker operation, effective January 2017
Japan DENAN, JIS C 8105-1	Whether a product is deemed to flicker	100 Hz or more within 5% of peak, or 500 Hz or more	Deemed not to create a flicker sensation
UL marketing claim verification	Flicker percentage of tested products	10% at 60 Hz mains, 8% at 50 Hz mains	Claim mark reads Low Optical Flicker Less than X%
IEC TR 61547-1	Flicker severity via a light flicker meter	PstLM below 1 acceptable, above 1 unacceptable	1 is the perceptibility threshold
CIE TN 006 and IEC TR 63158	Stroboscopic effect	SVM of 1 is the visibility threshold	Method, not a limit
NEMA 77-2017	Acceptance criteria	Guidance by application	Referenced, numbers not restated here

The pattern explains why a single flicker specification is so hard to compare. California constrains flicker percentage in a frequency band; Japan offers two routes, one a frequency floor with a tight amplitude limit and one a much higher floor with no amplitude limit; UL verifies a percentage at mains frequency; the IEC metric is a perceptibility index. Each answers a different question, and a product can pass one and fail another without any test being wrong.

The caveat is not small. Those instruments were verified through LED professional’s summary, which is explicit that they are general-lighting documents; this article did not verify whether FMVSS 108 or the ECE lighting regulations contain a flicker limit, and makes no claim either way. The practical consequence is that flicker compliant on a vehicle lamp listing is unverifiable unless the seller names the instrument and the test house. The general-lighting limits are reference points for what the industry considers low, not a legal standard for an auxiliary lamp.

Why a camera records banding your eye never sees

A camera is a sampling instrument, and it samples differently from the eye. Most consumer image sensors use a rolling shutter: the frame is not captured at a single instant but scanned across the scene, line by line, so different parts of the image are recorded at different times (Wikipedia, Rolling shutter). Point that sensor at a lamp switching at a few hundred hertz and the top of the frame may be recorded during an on-pulse while the bottom is recorded during an off-pulse.

The arithmetic is simple enough to do in the head. If the sensor takes t_read to scan the full frame and the lamp switches at frequency f, the number of on-off cycles inside one readout is approximately f × t_read. A 200 Hz lamp and a 1/60 s readout give about 3.3 cycles across the frame, which appears as three or four bands. The same lamp at 1 kHz gives about 17 bands, which at normal viewing size collapses into a fine texture or a colour cast rather than a legible stripe. That is why one lamp can look terrible on one phone and acceptable on another: readout time, exposure time and in-camera processing all differ.

Exposure time interacts with the same geometry. The rolling shutter article describes the partial exposure case directly: if a flash lasts for only part of the exposure, the illumination may be present for only some rows of the frame, so the top third of a picture can be brightly lit while the bottom two thirds is dark. The same mechanism applies to fluorescent lighting and to any fast burst of light, which is exactly what a PWM-driven LED produces several hundred times per second.

The conclusion is a discipline point. A camera is excellent evidence that modulation exists and a poor instrument for measuring it. The CIE-derived summary of temporal light artefacts says so explicitly: photographs and video of modulated light show temporal light interference, the artefacts vary with shutter type, frame rate and in-camera mitigation, different cameras show different artefacts, and using a common camera is not a valid or objective means of assessing the artefact from lighting equipment (Wikipedia, Temporal light artefacts). A phone video is a screening test, not a measurement.

PWM lamps and the cameras that point at them

Once cameras are part of the vehicle rather than in the driver’s hand, the same physics becomes an engineering requirement. Sony’s automotive sensor group describes the problem in terms of traffic lights and road signs: LED light sources in those applications switch on and off at a frequency that cannot be detected by human eyes, and if the exposure time is shorter than that on-off timing the image may be taken during the off period, so the emission is not captured at all (LED Flicker Mitigation Technology for Automotive Use).

Sony states the consequence that matters for driver assistance: there is a risk that a lit LED signal is mistakenly recognised as a flashing one. The fix has two halves. An exposure longer than the emission period makes it possible to capture the source, but a longer exposure risks pixel saturation, so Sony describes sensor designs intended not to saturate even when the exposure exceeds the emission period, which allows the source to be captured in every period with no flicker in the image signal.

Three boundaries belong on that paragraph. The page concerns traffic lights and signs, not auxiliary lamps, so applying it to a light bar is an inference from the same mechanism rather than a statement Sony makes. It describes one supplier’s mitigation rather than an industry requirement. And it says nothing about what a dashcam records, because a dashcam is a different sensor with a different readout and its own processing.

PWM as a signal, not just a dimmer: the daytime running lamp case

On a vehicle a PWM waveform is not always a brightness command. It can be a message. A granted US patent describes a combination low-beam and daytime-running-lamp headlamp that uses PWM detection to decide which of its two internal LED sources is powered: if the lamp detects a PWM signal on the low-beam voltage input it switches to driving the DRL LEDs, and if it detects a steady-state input voltage it drives the low-beam LEDs (US10647244B2, LED headlamp with daytime running lamp).

The same document’s background explains the practice that produced the convention: depending on regulation and equipment, the daytime running function may be implemented by dedicated lamps, by operating the low-beam headlamps or fog lamps at full or reduced intensity, or by operating the high beams at reduced intensity. Reduced intensity has to come from somewhere, and on a halogen lamp the cheapest source is a duty-cycled supply.

Two consequences follow. A lamp that behaves perfectly on a bench can behave differently on a vehicle whose body computer signals dimming by chopping the supply, because the lamp is now receiving a command as well as power. And PWM in a vehicle context describes at least three different things: a dimming method inside the lamp, a signalling convention on the input, and a by-product of a regulator that is not dimming at all. This patent describes one embodiment and should not be generalised into a rule about how all vehicles work, but it establishes that the signalling interpretation exists in production practice.

For anyone replacing a lamp, that is why the vehicle-specific checks in the CANbus and headlight flickering guide start with the vehicle rather than the lamp. This article does not repeat those checks; it supplies the mechanism behind them.

Why invisible is not the same as no measurable effect

The claim that a lamp is fine because nobody can see the modulation rests on an assumption the perceptual literature does not support, and it is worth separating what is established from what is inferred.

The strongest evidence for an effect below conscious perception predates LED lighting and used fluorescent lamps. Werner’s review cites Wilkins and colleagues, who examined long-term exposure at 100 Hz and found that a switch from substantially modulated light at 60% modulation to lightly modulated light at 6% at the same frequency reduced headaches and eyestrain within a few weeks, with statistically significant results. The same review is careful about the direction of evidence: the opposite effect could not be demonstrated within the four weeks the study allowed, much of the lighting industry ignored the result, and the research used fluorescent sources whose waveform differs from an LED’s and may be less harmful.

DIAL adds a dose relationship from a different study: increased headache risk at 100 Hz with modulation above 35%. It also gives the mechanism that makes the claim plausible rather than mystical. Above the flicker fusion frequency the modulation is no longer consciously perceived, but the optic nerve and cortex can follow stimuli up to about 160 Hz depending on modulation and waveform, and responses have been identified in the retina up to about 200 Hz.

Three limits belong with that. No source reviewed here establishes a health threshold for a vehicle cabin, where exposure is intermittent, light levels are low at night and the observer works under vibration. None establishes that a lamp at 400 Hz with 5% modulation is harmful. And the quantified dose-response relationships sit at 100 Hz, which is what a cheap driver produces by rectifying a 50 Hz supply. The defensible position is that deep modulation at low frequency is a design choice with a documented history of complaints, that evidence at higher frequencies is thin in both directions, and that a buyer who wants to avoid the question should look for high frequency and shallow modulation rather than for the word flicker-free.

How to measure a lamp you already own

Five methods are available without a laboratory, and they differ enormously in what they establish.

Method	What it can show	What it cannot show
Wave a pen rapidly in front of the lamp	Whether a stroboscopic effect is visible at all	Any number; a negative result at one speed proves little
Record with a phone in slow motion	That modulation exists; approximate band spacing	Modulation depth, true frequency, or a comparable figure
Photograph with a manually set short exposure	Band count, from which frequency can be estimated	Anything about waveform shape
Photodiode and oscilloscope	The optical waveform, from which percent flicker and flicker index can be computed	Perception-weighted metrics unless you implement them
Laboratory flicker meter or SVM measurement	PstLM or SVM to the published method	Whether you personally find it acceptable

The pen test is the one Perz recommends in the TU/e announcement of her thesis, and it is a better screen than it sounds. Swing a pen up and down quickly in front of the lamp, then do the same in front of a window. If there is no difference, the driver is designed so that no stroboscopic effect is visible; if the window gives a continuous blur while the lamp gives a series of still images, the lamp has a visible stroboscopic effect (TU/e).

Band counting is the cheapest quantitative method, and the arithmetic is the camera section rearranged: frequency is approximately the number of bands divided by the exposure time, so twenty bands in a 1/100 s exposure implies about 2 kHz. It fails when the count is too high to read or the exposure is long enough for motion blur to smear the bands.

An oscilloscope measurement is the only one of the five that produces a waveform you can defend, and the sampling requirement is not optional. The standard setup specifies a photodetector matched to the CIE photopic curve, a sampling interval no greater than 50 microseconds, a record of at least two seconds and a controlled 25 °C ± 2 °C ambient. Two seconds at 20 kHz is 40,000 samples, which is what makes low-frequency modulation visible in the record.

Phone light-meter apps are not on the list for arithmetic rather than opinion reasons. Percent flicker requires resolving a minimum that may last a fraction of a millisecond, and the standard method sets the sampling interval at 50 microseconds or better. An app that samples a few times per second, or that reports a smoothed illuminance, cannot resolve the minimum of a 200 Hz waveform, so the number it reports is an average with unknown smoothing and is not comparable with any published figure. Measuring LED PWM flicker properly therefore starts with a photodiode, not a phone.

Reading a supplier claim: the numbers to ask for

Almost no vehicle-lamp listing publishes a flicker figure, and the reason is not conspiracy. The measurement needs a photodiode, a fast recorder and a controlled setup, and the result is a number most buyers cannot interpret. What follows is the short list of questions that produce a usable answer, in order of how often the answer turns out to be that the data does not exist.

Ask for	Why it changes the answer	What a usable answer looks like
Switching frequency at each brightness level	It is the dominant variable in visibility	A number in Hz per mode, not high frequency
Duty cycle or modulation depth at the dimmest level	Dimming usually increases modulation	A percentage, with the mode named
Which mode the figure applies to	A lamp can be clean at full output and poor at 10%	One figure per mode
Test method and instrument bandwidth	A number without bandwidth is not comparable	Sampling rate, record length, photodetector type
Electrical description of the input	PWM on the supply may be a command, not a level	Whether the lamp expects steady or chopped input

The fourth row is where most conversations end. A supplier who says flicker 3% and cannot say at what bandwidth, at what brightness and with what photodetector has given a number that cannot be compared with the Title 24 or UL figures above, because those instruments specify their own conditions. The honest form of the answer is a waveform or a test report; the second-best form is a frequency and a modulation depth with the method named.

LED PWM flicker that is invisible in still use becomes visible in three situations that matter here: it bands on video, it produces a stroboscopic effect on rotating equipment such as a fan or a wheel, and it is at its deepest in the dimmest mode, which is often the mode used while parked.

When PWM ripple is normal and when it is a fault

The distinction is not subjective, and it is the boundary between this article and the troubleshooting articles. Normal PWM ripple has four properties: it is stable over time, consistent with the mode, predictable when the mode changes, and present on every emitter in the lamp at the same depth.

Observation	More consistent with designed PWM	More consistent with a fault
Appears only in the dim or DRL mode	Yes	Unlikely
Changes with temperature or after a few minutes	No	Yes
Changes with vibration or a wiggle test	No	Yes
Affects one segment or one pod only	No	Yes
Present at full output, steady and deep	Rare	Possible driver or supply problem
Accompanied by smell, discolouration or heat	No	Stop using the lamp

The last row is not a diagnostic question. A lamp that smells, discolours or runs hot should be disconnected rather than investigated, and this article stops at that point. The systematic checks for a lamp that flickers, including supply, ground, relay and connection faults, are in LED Light Bar Flickering: Causes and Safe Checks, and the vehicle-side checks for a headlamp conversion are in the CANbus guide. Neither is repeated here, because repeating a diagnostic sequence in a theory article is how two pages end up competing for the same question.

Matching the dimming method to the job

The same lamp can be acceptable in one role and unacceptable in another, because the role decides which artefact matters. A DRL accent seen from outside at a distance is judged differently from a work light illuminating the surface a mechanic is looking at, and a lamp used as the primary illuminator should not be relying on duty cycle for its brightness at all.

Role	What the artefact costs	What to specify
DRL or position accent	Camera banding, stroboscopic effect seen by other drivers	High switching frequency, shallow modulation
Work light for inspection or repair	Eye fatigue, stroboscopic effect on tools and fans	Constant-current dimming where available, or PWM above 1 kHz
Scene or camp lighting	Visible flicker in peripheral vision, camera banding	Frequency above the stroboscopic band, low modulation
Signalling or warning modes	Legibility of the flash pattern	A designed flash rate, not a by-product of dimming
Primary driving illumination	Everything above, plus glare	Full output operation, no duty-cycle dimming

Two product observations follow from the catalogue rather than from theory. Lamps with several modes, such as the Pinnacle Series 40 inch curved spot beam bar with its amber and white DRL and its strobe function, necessarily contain mode-switching electronics, and the low-brightness modes are where a PWM implementation would appear if there is one. Work lights sold for inspection use, such as the heavy-duty LED work light range, are the category where a visible stroboscopic effect on a rotating tool is most likely to be noticed by the person holding it. Neither product page states a switching frequency, and this article claims none for them; the point is a question to ask, not a specification to assume.

A specification checklist before you buy

Eight items, ordered by how much uncertainty each one resolves.

• Ask for the switching frequency in hertz at each brightness mode, not the phrase flicker-free.
• Ask for the modulation depth or percent flicker at the dimmest mode, with the test method named.
• Ask whether brightness in the main mode is set by current or by duty cycle; a primary illuminator should be current-controlled.
• Ask whether the lamp expects a steady input voltage or can accept a PWM-dimmed supply, and whether it detects one.
• Confirm the input current at full output, because a lamp that dims by chopping still draws peak current while it is on.
• Note whether cameras point at the lamp, and treat a camera-visible band as a design constraint rather than a curiosity.
• Check the mounting location for airflow, because a hotter driver is one more likely to fold back or behave inconsistently.
• Decide in advance what you will do if the lamp flickers only in one mode after installation, and confirm the seller’s position before ordering.

Frequently asked questions

Is PWM dimming bad for LED lights?

Not inherently. PWM is a standard way to obtain several brightness levels from one regulated current, and it does not damage the LED when peak current and thermal design are correct. What varies is the artefact: a 100 Hz dimmer at 100% modulation produces a waveform measuring 100% percent flicker and an SVM near 4.9, while the same circuit at 2 kHz produces the same percent flicker and an SVM near 0.34. LED PWM flicker is a switching method whose consequences depend almost entirely on frequency and depth.

Why does my LED light bar flicker on camera but look steady to me?

Because the camera samples faster than the eye averages, and most sensors use a rolling shutter that exposes the frame line by line. A 200 Hz lamp recorded with a 1/60 s readout gives roughly three cycles across the frame, which appears as bands; at 1 kHz the bands are too fine to resolve and look like texture or a colour shift. The camera shows something real that the eye integrates away.

What is a good flicker percentage?

There is no universal number, but the reference points are California Title 24 at under 30% below 200 Hz, a Japanese DENAN route of 5% variation at 100 Hz or a 500 Hz floor, and a UL verification limit of 10% at 60 Hz mains. Below 5% above 1 kHz is unremarkable in a good driver. A figure of 100% is normal for a full-depth PWM dimmer and tells you nothing until you know the frequency.

Is 100 Hz enough to avoid visible flicker?

It is enough to be above the flicker fusion frequency for most observers in steady gaze, which is why 100 Hz and 120 Hz became de facto floors. It is not enough to avoid the stroboscopic effect. A 100 Hz square wave at full modulation is close to the worst case for a moving object, and it is also where the headache literature is concentrated.

What is the difference between percent flicker and flicker index?

Percent flicker compares only the maximum and minimum of a cycle, so a 10% duty pulse and a 50% duty pulse both measure 100%. Flicker index compares the area above the average with the total area of the cycle, so the same two pulses measure 0.9 and 0.5. Percent flicker answers how deep the modulation goes; flicker index answers how much of the cycle is spent near the peak. Neither contains a frequency term.

What SVM value should a vehicle lamp have?

SVM is defined for indoor levels above 100 lx and moderate movement below 4 m/s, so there is no validated target for a driving lamp at speed. What can be said is that an SVM of 1 is the visibility threshold where the model applies, and below 1 the detection probability is under 50%. For a work light used at close range an SVM below 1 is a reasonable target; for a driving lamp, frequency and modulation depth are more useful figures than a single index.

How can I check a lamp without any instruments?

Use the pen test. In a darkened space, swing a pen rapidly up and down in front of the lamp, then do the same in front of a window. A continuous blur in both cases suggests no visible stroboscopic effect. A series of still images in front of the lamp but a blur in front of the window indicates a visible stroboscopic effect, which places the modulation in the 80 Hz to 2 kHz band at a depth the eye can resolve.

Editorial note

Written by the SAEC Lighting Engineering Team, Foshan. This article separates three kinds of statement: facts quoted from the sources below, arithmetic performed here from published definitions and labelled as computation, and boundaries where the sources do not settle a question. The IEEE 1789 numeric risk bands, any flicker limit inside FMVSS 108 or the ECE lighting regulations, and any switching frequency for SAEC products are not stated because they were not verified. Product links answer specification questions and do not imply a measured flicker figure.

Last updated: 2026-10-18

Sources

• IEEE Standards Association, IEEE 1789-2015, Recommended Practices for Modulating Current in High-Brightness LEDs for Mitigating Health Risks to Viewers, record and abstract, status Inactive-Reserved: https://standards.ieee.org/ieee/1789/4479/
• DIAL GmbH, “IEEE 1789: A new standard for evaluating flickering LEDs?”, 23 June 2022: https://www.dial.de/en-GB/news/ieee-1789-a-new-standard-for-evaluating-flickering-leds
• LED professional, “Flicker: Standards and Test Methods”, 28 November 2016: https://led-professional.com/artikel/flicker-standards-and-test-methods
• LED professional, “Flicker Beyond Perception Limits: A Summary of Lesser Known Research Results”, 9 August 2016: https://led-professional.com/artikel/flicker-beyond-perception-limits-a-summary-of-lesser-known-research-results
• Wikipedia, “Flicker (light)”: https://en.wikipedia.org/wiki/Flicker_%28light%29
• Wikipedia, “Stroboscopic effect”, SVM definition, threshold function, IEC TR 63158 and CIE TN 006 references: https://en.wikipedia.org/wiki/Stroboscopic_effect
• Wikipedia, “Temporal light artefacts”: https://en.wikipedia.org/wiki/Temporal_light_artefacts
• Wikipedia, “Rolling shutter”: https://en.wikipedia.org/wiki/Rolling_shutter
• Eindhoven University of Technology, “Stroboscopic visibility measure: understanding how people experience LED-light fluctuation”, 1 February 2019: https://www.tue.nl/en/news/news-overview/01-02-2019-stroboscopic-visibility-measure-understanding-how-people-experience-led-light-fluctuation/
• Sony Semiconductor Solutions Group, “LED Flicker Mitigation Technology for Automotive Use”: https://www.sony-semicon.com/en/technology/automotive/lfm.html
• US Patent US10647244B2, “LED headlamp with daytime running lamp”: https://patents.google.com/patent/US10647244B2/en
• U.S. Department of Energy, “Flicker Research”, Solid-State Lighting programme: https://www.energy.gov/cmei/ssl/flicker-research