Flicker, stroboscopic effect and low-end dimming: what to ask for in an LED fitting specification

At a glance

Read on 6 November 2026. The UK limits are at full load only. California figures are a US precedent, not a UK requirement.

The terms, in plain words

Four effects get called flicker, and they are not the same. SI 2021/1095 defines the first two, and gives a number to each. The drawing sets them on a frequency scale.

• Flicker is the perception of visual unsteadiness caused by light that fluctuates with time, for a still observer in a still scene (the Regulations and CIE e-ILV 17-22-092). The metric is Pst LM, where st stands for short term and LM for light flickermeter. Pst LM = 1 means the average observer has a 50 per cent probability of detecting flicker.
• Stroboscopic effect is a change in motion perception caused by fluctuating light, for a still observer in a moving scene. The metric is SVM, the stroboscopic visibility measure, and SVM = 1 is the visibility threshold for an average observer.
• Phantom array is a trail of light spots seen when the eye moves past a fluctuating source. A CIE workshop deck hosted by ENERGY STAR gives its range as about 80 Hz to 2 kHz, and listed a metric for it as a gap in 2017. It has no limit in the UK Regulations.
• Camera banding is not on that list. The CIE workshop deck says the effects of fluctuating light on equipment such as cameras were outside its scope, so Pst LM and SVM, which describe human vision, do not say whether a restaurant will film cleanly.

Two older measures describe the waveform itself. Percent flicker is 100 x (A - B) / (A + B), where A is the peak and B the trough of the light output; the flicker index is the area above the average over the total area (CIE e-ILV 17-22-094). A US DOE deck says percent flicker ignores shape, duty cycle and frequency, and flicker index ignores frequency. So neither alone says how visible a swing is, and Pst LM and SVM, which weigh frequency, are the metrics the Regulations use.

The Regulations do not name the test standards. They say the metrics are defined in standards of an international standardising body. The IEC webstore lists IEC TR 61547-1:2020 (third edition, July 2020) as describing an objective light flickermeter, and IEC TR 63158:2018 (first edition, March 2018) as describing an SVM meter for general indoor lighting above 100 lux with movement below 4 m/s. The CIE's technical note on these effects, TN 006:2016, is shown by the CIE as superseded by CIE 249:2022.

Why dimming brings flicker

The DOE/RTI report says the temporal light artefacts of LED lamps are most often set by circuit design. In practice flicker is what the driver does to the LED to dim it, and what happens when driver and dimmer are not a pair. The US DOE and PNNL describe two ways of dimming.

• Constant current reduction lowers the current and keeps the LED on. It does not create flicker, and it lengthens LED life, but the DOE notes that regulation is more difficult at deep dimming levels, where the current is very low.
• Pulse-width modulation keeps the same current and switches the LED on and off, with the share of time on setting the average. The DOE says it regulates well at deep dimming, but the switching frequency is important: it can create unwanted flicker, set a minimum dimming level and cause noise. A DOE deck shows an example of a 0-10V driver using it with a flicker frequency of about 250 Hz.

The drawing shows both at full output, half and 10 per cent. The table puts exact numbers on the same example waveforms. It shows why one number is not enough: pulse-width modulation has a percent flicker of 100 at every dimmed level, but a flicker index that rises from 0.50 to 0.90, and neither says how fast the pulses are.

Exact arithmetic on idealised waveforms drawn for this sheet, not measurements. Pst LM and SVM weigh frequency and cannot be read off a drawing.

• The waveform changes with the load and the dimming level. The DOE flicker resources page says the current delivered to the LED depends on the driver loading and the dimming level. In the DOE/RTI dim-to-warm tests, flicker and stroboscopic readings differed at each power level, and one candelabra-style lamp had an SVM above 1.0.
• Phase-cut is its own source of trouble. PNNL lists LED products that do not dim, dim erratically, turn off long before a halogen lamp would, flash or stay on when off, show colour shift, make noise or produce objectionable flicker, often with 100 per cent modulation.
• Compatibility does not guarantee quality. PNNL says compatibility alone does not guarantee a lack of flicker, smoothness or a particular minimum level, and that these are best judged by observation, which is why it strongly recommends a mock-up of every lighting circuit tested over its full range.

Which numbers appear where, and where they stop

Several numbers are in circulation. Only the first is law in Great Britain, and it has three limits that matter to a decorative fitting.

1. Full load only. The Regulations measure at full load, the condition of maximum, undimmed light output, and at the reference control settings, the factory defaults. Nothing we found in the text limits flicker at a dimmed level, which is where the sources above find the problems.
2. Mains light sources only. The flicker and stroboscopic rows of Table 5 name LED and OLED mains light sources. The Regulations define a non-mains light source as one that needs a separate control gear to operate on the mains. On our reading a module fed by a separate driver, which is how many decorative fittings are built, is not a mains light source, and neither Table 5 nor the control gear rows of Schedule 3 set a flicker limit for the driver. A compliance adviser should confirm. See UK ecodesign rules for lighting products and LED drivers for decorative luminaires.
3. The limit is the threshold. Pst LM = 1 and SVM = 1 are the points at which an average observer can detect the effect, by the Regulations' own definitions. A source that just meets them is not flicker-free. The DOE advises conservative specifications for critical applications.

Where flicker matters: a DOE deck of 2014 lists general and task lighting, hospitals, classrooms and offices, and says it matters less for accent lighting on artwork and very low-intensity lighting. A hotel bar or restaurant is a place where guests stay for hours, near the lights, and may be filmed: so ask for a figure, and for the figure at the level in use.

Low-end dimming: what to ask for

Specify the lowest level as a number, and say what is allowed to happen on the way. The DiiA guide to DALI describes levels from 0.1 per cent to 100 per cent and curves that follow the eye; how low a particular driver and light source actually go is a data sheet and a test result, not a feature of the protocol. California's table asks for a minimum dimming level of 10 per cent or lower for its residential sources, and a PNNL phase-cut example checked whether lamps became unstable at or below 10 per cent of light output and used a low-end trim so lamps went out before the point of instability. Evening scenes in hospitality often sit well below 10 per cent, so write your own figure.

How to test on site

Measuring Pst LM and SVM takes a calibrated meter and a laboratory. On site you can only find out whether there is a problem. The DOE advice, which dates from 2014, is to see the product in person with the same driver and dimming setting as the final installation and to learn to test for flicker.

1. Use the real chain. One fitting, with the production driver and the dimmer or controller that will be installed, on the control method in the dimming schedule.
2. Step through every level in written steps (for example 100, 50, 25, 10, 5, 1 per cent and the lowest), and look at each for steadiness, steps, hum and the lowest level. Then command it off and look for glow.
3. Try the simple tests. The DOE deck suggests a flicker wheel or a spinning top, waving your fingers in the light to look for a strobe effect, and notes that a digital camera will sometimes pick up flicker. They show a problem; they do not size it.
4. Film it if it will be filmed. Use the camera and settings that will be used, including slow motion, at the evening scene and the lowest level. Banding that shows only on the camera is still a result worth recording.
5. Treat a phone app as a hint. In a DOE study of handheld flicker meters against a reference meter, the smartphone app tested differed from the reference by 17.31 on average in percent flicker, and by 33.20 at dimmed levels; most dedicated meters differed by 0.2 to 3.1, though one differed by 19.10. The DOE says even a free app can indicate that a problem may exist, and should be followed up with more precise equipment.
6. Send for a measurement where a number is wanted: the laboratory reports the waveform, frequency, percent flicker, Pst LM and SVM at each stated level.

At each of [100, 50, 25 and 10 per cent and the lowest level] of full light output, on the control chain in the dimming schedule, the fitting is to have Pst LM not more than [1.0] and SVM not more than [0.4], measured on the supplied model by [accredited laboratory], and reported with the waveform, the modulation frequency and the percent flicker at each level. These are the specifier's limits: the UK Regulations apply Pst LM 1.0 and SVM 0.4 at full load only [where they apply]. In [space where guests stay for hours] not more than [ ].

Lowest stable level [1] per cent of full light output on the specified control chain. From full output to the lowest level the light is to dim smoothly, with no visible steps, flash, drop-out, buzz or colour jump, and the fitting is to go fully off, with no glow, when commanded off. The dimming method of each driver [constant current reduction / pulse-width modulation at not less than [ ] Hz / other] to be stated on the shop drawing.

Where [space] is to be filmed or photographed, no visible banding or flicker is to appear in video at [ ] frames per second and [shutter setting] at the evening scene and at the lowest level, shown on the mock-up and recorded.

The partner is to supply the driver data sheet with flicker data at the stated levels. One fitting on the real control chain is to be stepped through every level and approved in writing by [lighting designer] before production. Any change of driver, dimmer, module or dimming method after approval is to be tested again.

Standards and sources

• SI 2021/1095, Lighting Products Regulations 2021 (Sch 1, Sch 2 Table 1, Sch 3 Table 5 and para 11, Sch 9): revised text, read 6 Nov 2026.
• IEC TR 61547-1:2020; IEC TR 63158:2018: IEC webstore scope read; not read in full.
• CIE e-ILV 17-22-092, 17-22-094; CIE TN 006:2016 (superseded by CIE 249:2022): definitions read.
• IEEE 1789-2015 (Inactive-Reserved); NEMA 77-2017: named, not read; check the current status.
• US DOE and PNNL: LED Dimming (2012); SSL Flicker Fundamentals (2014); Fighting Flicker (2019); PNNL-22945: dated, US-centred.
• California Energy Commission, 2019 Energy Code: Residential Lighting (JA8-2019): JA8-2022 not read.

Questions we are asked

What flicker limits apply in the UK?

For LED and OLED mains light sources, SI 2021/1095 sets Pst LM at 1.0 or less at full load, and SVM at 0.4 or less from 1 September 2024 (0.9 before). They apply to a product placed on the market, at full load and at its factory default settings. We found no UK limit for the dimmed state, and on our reading the rows do not obviously reach a module that is fed by a separate driver.

Does Pst LM of 1.0 mean no flicker?

No. By the Regulations' definition, Pst LM = 1 means an average observer has a 50 per cent chance of detecting flicker, and SVM = 1 is the visibility threshold for the stroboscopic effect. A source at the limit is at the edge of what people can see. The US DOE advises conservative specifications for critical applications.

Why does my LED fitting flicker only when dimmed?

Dimming changes what the driver does. Reducing the current keeps the light steady but is harder to control at very low levels; pulse-width modulation switches the LED on and off and creates flicker if the frequency is low, with a depth that can be 100 per cent. Driver and dimmer pairings add problems. The cure is to ask for the figures at the dimmed levels and to test the real chain.

How low can a dimmed LED go?

It depends on the driver, the dimmer and the light source. The DiiA guide describes DALI levels from 0.1 per cent to 100 per cent, but a driver's real minimum is in its data sheet and in a test. Specify the lowest stable level you need as a percentage of full output, and test it on the real control chain.

Can I test flicker with my phone?

Only as a hint. In a US DOE study the smartphone app tested differed from a reference meter by 17.31 on average in percent flicker (33.20 at dimmed levels), where most dedicated meters differed by 0.2 to 3.1. A slow-motion camera can show banding that matters if the room is filmed. Send for a laboratory measurement when a number is needed.

Does flicker matter in decorative and hospitality lighting?

A 2014 DOE deck says it matters less for accent lighting on artwork, but guests in a bar or restaurant sit near the lights for hours, and rooms may be filmed. A level and its test method cost little to write into a specification. We give no medical advice: for anyone with a health concern, ask a qualified adviser.