Reading a Red Light Spectrum Report: Peak Wavelength, Tolerance, and the Normalised Graph Trap

RedVance designs the LED boards and selects the bins that determine these numbers, so this is written from the side that reads spectral data before a product exists. The format is deliberately generic: report layouts vary between labs and factories, and the aim is a reading method that works on any of them. This is the spectral companion to our guide to reading an irradiance test report, which covers the intensity side.

Two axes, and both are frequently misunderstood:

• Horizontal: wavelength in nanometres (nm). Red LEDs appear around 630–660nm, near-infrared around 810–850nm. Note that wavelength is measured in nm — hertz applies only to pulse frequency, never to wavelength itself.
• Vertical: intensity. This is the axis that decides whether the graph is informative or decorative, depending on whether it is normalised or absolute. Section 03 is about exactly that.

Each LED type produces a peak, and a dual-wavelength panel shows two. The useful information is not that the peaks exist — it is where they sit, how wide they are, and how much energy each one carries.

“660nm” sounds unambiguous. It is not — a single spectral measurement yields several different figures:

These can differ by several nanometres on the same LED. So two suppliers measuring identical hardware can honestly publish different numbers, in the same way that LED counts can be honestly stated two ways — the parallel ambiguity in our LED count guide. Ask which figure a spec sheet is quoting; a report that specifies “peak wavelength” is being precise, one that just says “660nm” is not.

This is the most important section on the page, and the mechanism almost no buyer knows about.

Most spectrum graphs published in marketing material are normalised: the tallest peak is scaled to 100 percent, and every other peak is drawn relative to it. Normalisation is a legitimate convention for showing where peaks sit. It is useless for showing how much energy each carries — and it can make a decorative wavelength look like a headline feature:

A panel with two main channels and four token wavelengths can produce a normalised graph showing six confident peaks. The graph is not falsified; it simply answers a different question than the one the buyer is asking. This is the measurement mechanism behind the “bonus wavelength” pattern described in our multi-wavelength guide.

A peak has a width as well as a position, and width is almost never discussed in this category. FWHM (full width at half maximum) measures how broad a peak is at half its height, in nanometres:

• Narrow FWHM: output concentrated close to the stated wavelength.
• Broad FWHM: a meaningful share of the energy sits well away from the nominal figure, even though the peak label reads correctly.

The practical consequence: two panels can both report a 660nm peak while delivering noticeably different amounts of energy actually near 660nm. A label saying “660nm” tells you where the summit is; FWHM tells you how much of the mountain is nearby. The wavelength-selection reasoning this feeds into is in our 660nm vs 850nm guide.

LED production is not perfectly uniform, so chips nominally rated 660nm emit across a small spread. That is why a credible spec states a tolerance:

660nm ± 5nm → measured peak falls between 655nm and 665nm

Tolerance is not a defect; it is physics plus procurement. The tighter the tolerance, the more selective the LED bin — and more selective bins cost more. So a stated tolerance is quietly a statement about component sourcing, which is why “which bin do you use?” and “what is your wavelength tolerance?” are the same question asked two ways. The binning mechanics are in our LED count and packaging guide, and how bin choice interacts with lifespan is in our L70 and LM-80 guide.

A spec sheet with no tolerance at all is either omitting it or has not measured it. Both are worth asking about.

One more reason a spectrum report needs its conditions stated: LED emission wavelength shifts with junction temperature, typically toward longer wavelengths as the device warms. The shift is small, but it means a spectrum captured at switch-on is not identical to the same panel measured after a full session.

So thermal state belongs on a spectrum report for the same reason it belongs on an irradiance report — and if a supplier states warm-up conditions on one document but not the other, that is worth noticing. The output-decline side of the same phenomenon is in our thermal drift guide.

1. No instrument named, or no calibration date. Same fatal gap as any test document.
2. Normalised graph presented as evidence of per-wavelength power. Wrong data type for the claim.
3. No tolerance stated. Either unmeasured or omitted.
4. No FWHM. Peak position without peak width.
5. No thermal state. Suggests a switch-on capture.
6. Model or serial does not match your device. The report covers something else.
7. Vague spectral language such as “full spectrum healing light” or “seven-colour spectrum” in place of numbered peaks.
8. A graph image with no underlying data table. A picture is not a measurement.

As with irradiance reports, none of these individually proves bad faith — but the response to asking is diagnostic, which is the engineering-depth test in our factory audit checklist and the broader pattern in our spec trust gap overview.

What a manufacturer can do about it: publish absolute spectral power distribution alongside any normalised graph, state which wavelength figure is being quoted (peak, centroid or dominant), give tolerance in ± nm and FWHM for each channel, state the instrument, its calibration date, and the thermal state at capture, provide the per-wavelength irradiance contribution in real units, and include the underlying data table rather than only an image. A factory that selects its own LED bins holds all of this before production begins — publishing it costs nothing but candour. Terms used here are defined in our glossary, and the wider verification discipline is in our spec validation guide.