← HCL Intelligence
Research & Evidence·13 min read

Flicker-Free Specification: Why SVM Matters

"Flicker-free" is not a number. This guide explains SVM (stroboscopic visibility measure), the EU's 0.4 limit, and the clause to write into a lighting specification.

Flicker-Free Specification: Why SVM Matters

"Flicker-free" is not a specification. It carries no number, no frequency and no test method, so two luminaires both described as flicker-free can differ by an order of magnitude in what a camera, a moving object or a peripheral glance actually sees. SVM — the stroboscopic visibility measure defined in CIE TN 006:2016 and applied in IEC TR 63158:2018 — closes that gap by weighting every modulation component in the light waveform against the frequency-dependent sensitivity of the human visual system. That weighting is the whole point: 10% ripple at 1 kHz is unremarkable, while the same 10% at 100 Hz sits exactly at the limit the European Union now enforces. This page sets out what SVM measures, the three values in play today — the EU's 0.4, NEMA 77's 1.6 and a 0.15 design target — where LED systems actually fail, and the clause to write into a specification.

Key Facts

  1. Three different artifacts, three different bands, three different metrics. Direct flicker (below roughly 80 Hz) is measured by PstLM; the stroboscopic effect (roughly 80 Hz to 2 kHz) is measured by SVM; the phantom array effect (from about 80 Hz upward, measured to 20 kHz) has no regulated metric at all. Source: CIE TN 006:2016, with test methods in IEC TR 61547-1 (PstLM) and IEC TR 63158 (SVM).
  2. The only mandatory temporal-light-modulation limit is the EU's. Ecodesign Regulation 2019/2020 requires PstLM ≤ 1.0 and SVM ≤ 0.4 at full load; the SVM limit was 0.9 from September 2021 and tightened to 0.4 from September 2024. Everything else — WELL, tender documents, brand standards — is voluntary. Source: Regulation (EU) 2019/2020, Annex II Table 4, as amended.
  3. SVM is frequency-weighted, so a limit in SVM is not a limit in percent flicker. For a single dominant modulation component, SVM equals modulation depth divided by the contrast threshold at that frequency, and the measured threshold is about 0.25 at 100 Hz. SVM ≤ 0.4 therefore corresponds to roughly 10% modulation depth at mains-related frequencies, but roughly 20% at 400 Hz. Source: Perz et al. (2018), JOSA A 35(2) 309–319.
  4. Percent flicker on its own is frequency-blind. The IEEE 1789-2015 low-risk line is depth ≤ 0.08 × frequency, i.e. 8% at 100 Hz but 100% at 1,250 Hz; California's market-based JA10 table accepts 16% at 200 Hz and only 1.0% at 40 Hz for the same depth limit. Source: IEEE 1789-2015 as summarised in US DOE/PNNL SSL material; McHugh Energy comments to ENERGY STAR (2018).
  5. Compliance collapses in the dimmed state. In California's JA8-2016 database of 8,749 inseparable luminaires, 68% met the IEEE 1789 low-risk line at full output but only 51% at 20% output; for light engines the drop was 89% to 30%. EU law tests only at full load, so a dimmed-state figure has to be asked for explicitly. Source: McHugh Energy comments to ENERGY STAR Luminaires 2.1 (2018), Tables 2–3.

"Flicker-free" is a marketing phrase, not a measurement

Flicker is not one phenomenon. CIE TN 006:2016 separates three temporal light artifacts, and each one is visible in a different frequency band and needs a different metric:

ArtifactFrequency bandMetricRegulated?
Direct flickerbelow ~80 HzPstLM (percent flicker and flicker index are also used)EU: PstLM ≤ 1.0 at full load
Stroboscopic effect~80 Hz – 2 kHzSVMEU: SVM ≤ 0.4 at full load; NEMA 77 recommends SVM ≤ 1.6
Phantom array effect~80 Hz upward, measured to 20 kHznone standardised (PAVM proposed in research)No

Two legacy metrics appear on almost every datasheet and neither answers the question. Percent flicker is (maximum − minimum) ÷ (maximum + minimum) × 100: a depth ratio with no frequency term. Flicker index adds a shape term but still no frequency term. Both are, by definition, blind to the difference between a waveform that is imperceptible and one that makes a lathe appear stationary — because that difference lives entirely in the frequency, not in the depth.

What SVM actually measures

SVM takes the light waveform, decomposes it into modulation components, and divides each component's relative amplitude by the visibility threshold for its frequency. The components are then combined with a Minkowski exponent of 3.7:

SVM = ( Σ (C_m / T_m)^3.7 )^(1/3.7)

The scale is anchored: SVM = 1 is the visibility threshold for an average observer (50% probability of detection), SVM below 1 is not visible, SVM above 1 is visible.

The threshold function T(f) is the part that makes SVM useful. Measured contrast thresholds for the stroboscopic effect are approximately 0.23–0.27 at 100 Hz and rise with frequency; a published fit over 50–800 Hz is:

T(f) = 2.865 × 10⁻⁵ × f^1.543 + 0.225

For a single dominant component, SVM = depth ÷ T(f). That converts a limit written in SVM into a depth limit that a driver engineer can act on:

Modulation frequencyDepth at SVM = 1Depth at SVM = 0.4 (EU limit)Depth at SVM = 0.15 (SUMLINK design target)
100 Hz~26%~10%~3.9%
200 Hz~33%~13%~4.9%
400 Hz~52%~21%~7.8%
800 Hz~109% (not reachable)~43%~16%

Two cautions belong with that table. First, it is a single-frequency approximation: real waveforms contain several components, and the 3.7-exponent sum makes the result more sensitive than a single-component calculation suggests. Second, SVM is defined for a situation with illuminance above about 100 lx and moderate movement speeds (up to about 4 m/s, i.e. hand movements and machinery); in dimmer spaces and for static tasks, direct flicker — PstLM — is the artifact that matters, not SVM.

The three numbers in play today

ReferenceRequirementStatus
EU Ecodesign Regulation 2019/2020, Annex II Table 4PstLM ≤ 1.0, SVM ≤ 0.4 at full loadMandatory for the EU market (SVM was 0.9 from September 2021, tightened to 0.4 from September 2024)
NEMA 77-2017Pst ≤ 1.0, SVM ≤ 1.6Recommended practice for general lighting — four times the EU SVM limit
IEEE 1789-2015%flicker ≤ 0.08 × f (low risk), ≤ 0.0333 × f (no observable effect)Recommended practice; no SVM metric
WELL v2, Light feature L07 (Electric Light Quality)≥ 90 Hz at every 10% step from 10% to 100% output, or IEEE 1789-2015 low-risk modulationVoluntary certification, one point
SUMLINK platform design targetSVM < 0.15; percent flicker < 1% across the dimming rangeDesign target published on the product pages; not an independent laboratory result

The spread between 0.4 and 1.6 is the practical problem. A luminaire can be perfectly "NEMA 77 compliant" and still be four times over the limit that applies to the same product placed on the EU market. A specification that says "low flicker" without naming the reference is therefore not a specification. Name the metric, name the limit, name the frequency band and name the dim level.

The IEEE 1789-2015 line is worth reading carefully for a second reason: it reaches 100% modulation depth at 1,250 Hz. The same 100% depth that is unacceptable at 100 Hz is unremarkable above about 1.25 kHz — which is why a claim like "under 5% flicker" can be simultaneously true and irrelevant if the frequency is not stated. The requirements in this table are quoted from published summaries of the standards and of the regulation; the paid standard texts and the amending regulation text were not read (see §K and the verification table, where the CIE TN 006, IEEE 1789-2015, IEC TR and EU staging-date rows are all marked 未驗證).

Where LED systems actually fail

Failure mode 1 — the dimmed state. The largest published dataset on this comes from California's JA8-2016 database, analysed in comments to the ENERGY STAR Luminaires 2.1 specification:

PopulationFull output20% output
Inseparable luminaires (n = 8,749)5,920 comply (68%)4,444 comply (51%)
Light engines (n = 410)366 comply (89%)124 comply (30%)

The mechanism is simple: dimming changes the driver's operating point, and the residual ripple, the control-loop behaviour and the switching pattern all change with it. EU law does not catch this, because the requirement is written "at full load". If a project dims to 1% in the evening, the compliance number that matters is the one at 1%, and it has to be requested.

Failure mode 2 — mains-related ripple at 100 Hz. On 50 Hz supplies (Hong Kong, Singapore, most of Asia and Europe) a poorly filtered driver leaves ripple at 100 Hz; on 60 Hz supplies it lands at 120 Hz. This is the worst place to have ripple: stroboscopic sensitivity peaks around 90–120 Hz, and the SVM threshold is at its lowest (T ≈ 0.26 at 100 Hz). A mere 10% residual modulation at 100 Hz is SVM ≈ 0.38 — essentially at the EU limit — while the same 10% at 800 Hz would be SVM ≈ 0.09.

Failure mode 3 — PWM dimming. Pulse-width dimming reaches 100% modulation depth by definition; only its frequency decides whether that matters. PWM in the 200–500 Hz range produces SVM values well above 1. PWM above roughly 1.25 kHz is permitted at 100% depth under the IEEE 1789-2015 low-risk line, but may still produce a visible phantom array in slow-motion capture or fast eye movement. Constant-current (analog) dimming reduces depth instead of chopping it, at the cost of driver efficiency and dimming range. The control protocol — DALI-2, 0-10 V, Matter over Thread — sets the command path, not the waveform: the same protocol can drive a clean or a poor waveform depending on the driver behind it.

Failure mode 4 — measurement conditions. The IEA 4E SSLC Platform's IC 2023 interlaboratory comparison, involving 24 laboratories across 15 countries, found material lab-to-lab variation in temporal light modulation measurement, specific issues with SVM measurement and instrumentation, and a large influence of AC power supply on PstLM. The practical consequence is that the same luminaire can fall on different sides of a limit line depending on who measured it and how. A report without its measurement conditions — instrument, sampling rate, capture duration, filtering, dim level — is not verifiable.

The clause to write into a specification

Copy-paste clause:

Temporal light modulation. At 100%, 50%, 20%, 10%, 5% and 1% of maximum light output, and at each CCT setting used in the design, luminaires shall have PstLM ≤ 1.0 and SVM ≤ 0.4. Measurement shall be made on the light output waveform in accordance with IEC TR 61547-1 (PstLM) and IEC TR 63158 (SVM). For each dim level the supplier shall report SVM, PstLM, percent flicker, flicker index, dominant modulation frequency and the measurement conditions (instrument, sampling rate, capture duration, filters). SVM ≤ 0.15 is preferred.

Six checks to apply when reviewing a submission:

  1. Is a metric named, or only the phrase "flicker-free"?
  2. Is the limit named, and against which reference — EU 0.4, NEMA 77's 1.6, or a project-specific target?
  3. Are the dim levels stated? A single figure at full output does not cover the state the building spends its evenings in.
  4. Is a frequency stated alongside any percent-flicker figure? Without it, the number cannot be judged.
  5. Are the measurement conditions attached, per IC 2023's finding that they change the result?
  6. If WELL v2 applies, is there evidence of ≥ 90 Hz at every 10% step from 10% to 100% output, or of IEEE 1789-2015 low-risk modulation?

What SUMLINK can and cannot state today

The BY001 Minimalist Linear Wall Light product page carries two temporal-light-modulation rows:

MetricPublished valueNote on the page
Percent flicker< 1% across the dimming rangeIEEE 1789-2015 no-risk region — design target
SVM< 0.15Stroboscopic visibility measure — design target

Both are labelled Design target on the product page, and that label is deliberate: they are engineering design values from the driver and LED platform specification, not a published independent laboratory result. In the internal bill-of-materials record, the BY001 configurations (7 W to 13 W, CRI 90, 2700–6500 K) are built on the LTECH SE-12-100-500-W2MA constant-current Matter-over-Thread driver, and SVM and percent flicker are still listed as open data gaps pending measurement. Where a driver datasheet lists IEEE 1789 or CIE-SVM among its approvals — as the SE-40-300-1050-W2MA datasheet does — that is a supplier listing for the driver, not a luminaire-level test result. Driver and control-gear documentation for the rest of the range is collected under drivers and controls.

So the honest position for a project team is this: the design intent is a platform whose modulation sits well inside the EU limit and inside the IEEE 1789-2015 no-observable-effect line, and the documentation to verify that at luminaire level is what we produce per project, on request, with the measurement conditions attached. We do not publish a health claim and we do not call a design target a certificate.

Platform specifications, driver bill of materials and dimming protocol data for the linear and architectural range are listed in the HCL linear light catalog.

Frequently Asked Questions

What is the difference between percent flicker and SVM?

Percent flicker measures only the depth of the modulation — (max − min) ÷ (max + min) × 100 — with no frequency term. SVM weights each modulation component against the visibility threshold for its frequency, across roughly 80 Hz to 2 kHz. That is why a waveform with 10% depth can be invisible at 1 kHz and borderline at 100 Hz, while percent flicker reports the same number for both.

What SVM value should a lighting specification require?

For products placed on the EU market the legal requirement is SVM ≤ 0.4 at full load (Regulation 2019/2020, tightened from 0.9 in September 2024). NEMA 77-2017 recommends SVM ≤ 1.6 for general lighting — four times looser. For dimmed operation, require the same limit at every dim level, not only at full output, and ask for the report at 100%, 50%, 20%, 10%, 5% and 1%.

Is "flicker-free" a technical specification?

No. It has no metric, no limit, no frequency band and no test method, so it cannot be verified or compared between products. Replace it with a metric (SVM or PstLM), a limit, the frequency band, the dim levels and the measurement standard — for example "SVM ≤ 0.4 at every dim level from 100% to 1%, per IEC TR 63158".

Does the EU Ecodesign flicker requirement apply to projects in Hong Kong or Singapore?

It applies to products placed on the EU market, so it is a market-access rule rather than a local code. In Hong Kong and Singapore the same numbers usually enter a project through a WELL v2 feature, a tender specification or a client standard. Because it is the only mandatory limit that exists today, it is the most defensible number to specify even where it is not legally required.

How is SVM measured, and why do reports differ?

SVM is computed from a captured light waveform using the method in IEC TR 63158, with PstLM measured separately per IEC TR 61547-1. The IEA 4E SSLC Platform's IC 2023 comparison found significant lab-to-lab variation, specific difficulties in SVM measurement and instrumentation, and a large influence of the AC supply on PstLM. Always ask for the instrument, sampling rate, capture duration, filters and dim level behind any reported figure.

Request BOM & Spec Sheets

Platform specifications, driver bill of materials and dimming protocol data for the linear and architectural range. Request BOM & Spec Sheets →