CRI vs. alternatives with measurement examples

[maukka]

Hi all, I just bit the bullet and bought a highly regarded BabelColor CT&A analysis software that provides much more information than I’ve been able to provide regarding color rendering. Research has shown that CRI is quite inaccurate in representing the color fidelity with LED lights. CT&A is also able to extract some replacement indices, such as CQS, CRI2012 and MCRI.

I gathered here some information on metrics.

CRI: Color Rendering Index (1995)
The only widely used standard for quantifying color rendering.
Ra: Average of the first 8 (pale) color samples in the test
R9: Saturated red sample, difficult for LEDs and often negative
R10-R14: More saturated color samples

CRI is not very accurate with LED light sources and it often penalizes lights which enhance chroma but are actually preffered by human observers. The problem can be alleviated by calculating the average of Ra and GAI (Gamut Area Index). The system can also be gamed, since a light can be optimized for high Ra, while the actual color rendering is bad.

CQS: Color Quality Scale
NIST proposed replacement for CRI to better estimate color fidelity with modern light sources.
Qa: General index (RMS average of 15 samples, Q1-Q15)
Qf: Color fidelity, takes into account only hue, ignores saturation effects
Qg: Gamut area, saturation of colors

CQS Qa is calculated from 15 samples and makes the gaming of the calculation difficult. Research shows that lights with high Gamut Area Scale (Qg) are often preferred. CQS takes into account the error of increased chroma but penalizes less from it than CRI.

CRI2012 (Ra,2012)
Another proposed CRI replacement that comprises of 17 color samples, including several darker tones. Still actively worked on.

MCRI
A color rendering index based on the memory of colors in which a person is shown a set of 9 familiar objects. Research has shown that MCRI is a better indicator of color fidelity than CRI (Ra) or CQS.

Sources:
http://vision.uni-pannon.hu/LeCV/docs/2012_P01_Rajendra_Correlation_between.pdf
http://cie2.nist.gov/TC1-69/Princeton/Explanatory%20notes%20on%20CQS.pdf

CT&A_Help.pdf

https://lirias.kuleuven.be/bitstream/123456789/261922/1/CRIVienna_KevinSmet.pdf

Here’s a sample screenshot of all the data BabelColor CT&A’s CRI function provides.

This is the data I measured out of my own lights with an X-Rite i1Pro in a table and graphical form. I will update these into my existing reviews later.

[ReManG]

Where do I send you donor LED’s for testing?

Your testing is some wonderful stuff for those of us that are wanting more data on color reproduction and appearance. I am glad the Astrolux A01 scored so well, this shows that they are using genuine Nichia 219B LED’s…

Do you have the ability/skills to reflow LEDs for testing, or are you just testing lights that you currently have as purchases?

Thank you for these contributions here in BLF…

[djozz]

I agree with ReManG, this data are missing from what is done on led-testing thusfar (I sure do not have the equipment to add this to my output tests), and is interesting to know.

[THE_dAY]

Thank you maukka for the excellent work.

I wanted to note that there looks to be an update to the old CRI standard coming because of the LED inaccuracies you mentioned.
http://www.cie.co.at/index.php/index.php?i_ca_id=981

Basically, it hasn’t been updated since 1974.
“However, with the rapid uptake of LED lighting, which has greater freedom in spectral design, the need to update the CRI has significantly increased.”

[maukka]

@ReManG

Do you have the ability/skills to reflow LEDs for testing, or are you just testing lights that you currently have as purchases?

I have never done that so can’t promise anything. I would have to look up what’s required in terms of skills and equipment.

These are all just lights I have bought myself so the amount of data is naturally limited.

[Gunga]

Wow. Amazing. Can you state what LED is in each light too? Great info.

[maukka]

Updated table with manufacturer info about emitters.

[higbvuyb]

Thanks for the great quality posts. Looking forward to more.

[Evgeniy]

Thanks, excellent article

[The_Miller]

Nice work!!!

Can I send you some stars with a LED on them?
Then all you need is a cell and two wires to light them up.

[maukka]

@The Miller

Can I send you some stars with a LED on them?.

Sent a PM.

[1dash1]

Nice work. I don’t claim to understand any of the test methodologies (or what goes on behind the numbers), but I am surprised that the average rating of all the test items correspond directly with their CRI (Ra) index. It seems like you could just do that one test and skip the rest.

[djozz]

@1dash1

Nice work. I don’t claim to understand any of the test methodologies (or what goes on behind the numbers), but I am surprised that the average rating of all the test items correspond directly with their CRI (Ra) index. It seems like you could just do that one test and skip the rest.

I see what you say here, but there’s a few differences here and there. It tells you that CRI is is a working indicator for colour rendering. But it costs nothing extra to use an indicator that is even better, and it helps the industry setting the right target for the products instead of a slightly wrong target.

In general, when you start testing things you have not tested before, often after a while you see a pattern: already from initial observation you can predict how it most likely will test. From that moment on, the tests will be less informative, but if you stop actual testing you will also miss the surprises: occasions where specimens test very different from your prediction.

[The_Miller]

Nice
then I will send you
A Nichia 219C
A LatticeBright with not so bad color
A star with a 1w warm white led we have used in the kitchen and I wanna mod into a flashlight
All on stars no soldering needed.
Nice to have the 219C measured for a more complete overview.

[The_Miller]

Now wait when we are at it I will also send the XHP50 and XHP70.
Really appreciate that this is possible and you are sending them back afterwards.
Let’s build a comprehensive list right?

[djozz]

Every individual emitter will test a bit different, so there will not be ultimate results. You could detect false claims by sellers though.

[sixty545]

Another (uknown) factor is the applied reflector. The reflector will only direct a part of the light rays into the hot spot. The direct (bluish) rays Will miss the reflector and go into the spill. A measurement of the LED without some (standard?) reflector can give misleading results.

[WalkIntoTheLight]

Fascinating. I’d be interested to see the values for a regular neutral-white XML2 when you test one, since it’s used on so many of my lights.

R (9) looks so bad on all those lights, with the exception of the Nichia 219B. What kind of Nichia 219 5000K emitter gives the bad R (9) value? Is that just because it’s a bit cooler, and so there’s less red in its spectrum, or is it a 219C that is generally less good at red?

What does a negative R (9) value actually mean?

[The_Miller]

Well testing some will give more data then testing none
Maybe each emitter is slightly different, but there must be enough similarities in tint that is close enough in the same bin, otherwise all reviews of leds have little use right?
For weight sake I will not send the S70 (also removing that is more difficult then the easy to access S50 XHP50) and for my eyes the tint of the two is similar, of exactly the same.

[maukka]

Here are the results for the bare unspecified emitters The Miller sent me. I was surprised how cool white the XHP50 and Latticebright turned out to be without a reflector.

I didn’t have any specs for the emitters so the voltage and current weren’t set for maximum brightness, because I didn’t want to destroy them and there was no cooling except for the stars. I did test them on different output levels and didn’t notice significant change in CRI or tint after reaching a certain threshold of usable output. The XHP50 was run at 6 volts and 1,5 amp and it was blinding.