The Colour of Light: LED Light Spectrum Tests

In my previous post I was trying to order my thoughts on how much power I needed for the LED lights I’m about to build. This post is going to look at the quality of that light. For LED lights this is mostly quoted as the CRI, or colour rendering index. This is effectively a measure of colour accuracy and is measured on a scale from 0 to 100. To get a score of 100, the light source has to render colours as well as natural sunlight. For the CRI of a light source to be considered ‘high’ it has to score over 90, and generally speaking for photography we’d ideally be looking for a CRI of at least 90 to reproduce colours accurately without having to ‘fix’ them in post processing. For video work the recommendation is that ideally we’d want the CRI somewhere in the low- to mid- 90’s.

Well that all sounds pretty straighforward: just look for LEDs with the appropriate CRI value and buy them. If you’re using a specialist supplier there’s a pretty good chance that LEDs with a decent CRI value will have it quoted in the product details. The problem is that lots of LEDs don’t have their CRI stated in their specs, particularly the ones you’re liklely to find on the likes of ebay and Amazon. To make matters worse, depending on the individual supplier I’m not convinced that the CRI numbers being quoted are always entirely (or at all) accurate, so my assumption would be that in these cases the LED’s CRI values are not going to be high enough to give decent colour reproduction in photography or video work.

CRI Measurement

Of course if the manufacturers or suppliers of LED’s aren’t making the CRIs for their products available, it should be possible to test them yourself. However, the method to measure CRI as defined by the International Commission on Illumination (or CIE) is unfortunately a bit tricky: it involves measuring the light reflected from a set of standard samples using the test light source, comparing them to the reflections from each of the samples using a perfect black body radiator, fiddling about with some complicated maths to find the Euclidian distances between the two results, based on those coming up with a number on a scale of 0 to 100 for each of the standard samples and finally taking their average. Note: this is a paraphrase of the process, but I think it captures the notion that this isn’t something you’re likely to be trying at home!

Simple Colour Testing

The better news is that in the absence of a home CRI testing method, you can go a way towards finding out how good the colour rendering of your LEDs is going to be using spectrum analysis. This that may sound like an expensive option, but there is a DIY spectrometer that can be put together on the cheap – see this post.

A good CRI rating should be reflected in a reasonably gap-free colour spectrum being produced, and by the same token a colour spectrum with gaps in it would indicate that the light source in question is going to give poor colour rendering and would have a low CRI if it was tested.

Tests

As a reference, the colour spectra produced by daylight (at a latitude of 55 degrees North) are shown below. The first of these was taken in the early afternoon on a clear sunny September day, and the second early on a clear morning, both taken facing North:

Not a lot of difference between them apart from there being more blue/violet in the early afternoon. Moving on to artificial light, first up is a 100W incandescent tungsten lamp, which produced the spectrum below:

As you’d expect there’s a lot more energy at the red end of the spectrum (and beyond in the IR region), but very little at the blue/violet end.

CFL

The lamp that was used to calibrate the spectrometer was a CFL, as they have a very distinctive (uneven and gappy) colour spectrum with peaks at a pair of very predictable wavelengths. This makes them good for calibrating spectrometers, but very poor for using as photography or videography lights. The spectrum I measured using a standard CFL bulb looks like this:

As can be seen there are several peaks in the spectrum but, importantly, the troughs between them are deep and there’s very little light up in the red area. If there’s hardly any light being transmitted in an area of the spectrum, there’s going to be very little reflection from the subject you’re trying to photograph and the rendering of colour in that area is going to be poor. Note also the green peak is the biggest, hence the green cast you get with CFL lighting if you don’t set your camera’s colour balance to match.

Getting on to the ‘photography’ lights, the first I checked was my Neewer CN-160, which gave the following spectrum:

This looks pretty decent: it’s a bit lacking up in the red band and down in the violet but it’s fairly even from blue through to orange with no major dips.

The Good, The Bad and The Ugly

Bad and Ugly

Now to the home grown solutions: some of these aren’t quite as nice as I might have hoped. First up, the spectrum for my 100W COB LED chip gave:

and my old 5050 LED strip panel looks like this:

I found a couple of ring LEDs that were bought for a project that never happened lurking amongst my miscellaneous components, so felt I had to put one of them to the test as well. Here’s what it produced (probably good news that I never used them!):

Note, these are all fairly characteristic of what I’d expect to see with cheap LEDs – the big notch in the cyan wavelengths and again not much violet or red. The 5050 LED panel is the best of the bunch, with a reasonable amount of energy in that cyan band, but it’s still not great and has been reassigned to lighting the bench in my workshop. The COB chip is never going to be used for anything more critical than a (very powerful) flashlight.

The Good

The cheap 3W SMT COB LED I tested for intensity in the previous post fared surprisingly well given it’s price and the lack of CRI information provided by the supplier:

By way of comparison with the cheap LEDs, and following my own advice in the second paragraph above, I bought some Nichia COB LED chips with a quoted CRI of 93 from a specialist supplier. Putting one of these to the test, in this case a 9W daylight balanced (5000K) version, gave the following spectrum:

To my eye, this looks pretty similar to the spectrum produced by the Neewer light I tested earlier. I also have some 37W versions of a similar Nichia LED (the NFCWJ120B-V2, 5000K, R9050) which has a quoted CRI of 95. The spectrum for this looks a bit different:

So still a little notch between the blue and the green, but a lot more energy up in the red region, with a much slower tail-off at that end of the spectrum than the other LEDs I’ve tested. They also appear to produce an astounding amount of light and are quite small: I’ll definitely be building something with them in the near future!

More …

I’m also planning to get hold of a high CRI warm white LED or two, to see what sort of spectrum they produce – maybe a combination of warm white and daylight balanced would be worth experimenting with. It’ll also be interesting to see if I can find anything that throws out a bit more light in the violet region, as this appears to be the band where the LEDs are leaving a gap compared to daylight.

I’m going to be buying a bunch of other cheap LEDs for some experiments where CRI isn’t a factor, so I’ll capture their spectra to see if they produce any surprises. At the end of the day I suspect you get what you pay for, and I’m not expecting many of the cheap ones to come close to giving the sort of spectrum I want for photo lights.