I love talking about color rendering. As some of the other people in my lab will attest: I frequently say it’s the most important, or at least most interesting, problems in color science. I started my trajectory towards color science very young in the theatre. I started out singing and dancing in a few shows, but by the end of highschool I was doing lighting design for a dozen community productions every year. Events, Plays, Concerts, Cafe’s everything I could touch I tried to light. My friends used to comment “Every time we go someplace you just go in and look at the lighting.” I still do that and it’s given me a life long enamorment of color rendering.
What I love about theatrical lighting design, and I think it should be employed more heavily in environmental lighting design, is the distortion of color rendering through the use of gels. In my last post I made some remarks about how lighting designers choose gels because they distort the color rendering of the lighting fixtures in specific ways. I might want a sunset yellow vs. campfire yellow. Or a deep saturated pink which I’ll mix with a light green to create dramatic shadows in the scenic design, or make some subtle shades of red more prominent in the painting. As lighting designers, we get to distort how the world is seen by the selection of illuminant colors. Eventually I want to get to talking about gels. Before we get there, I think a good understanding of color rendering is needed.
So let’s start with a definition from the CIE International Vocabulary, Color Rendering is the “effect of an illuminant on the colour appearance of objects by conscious or subconscious comparison with their colour appearance under a reference illuminant.” The Illuminating Engineering Society (IES), whom developed TM-30, also copies this definition in their dictionary. I like to simplify it a little bit and use a definition from some more recent research. Color Rendering is the property of a lighting fixture which describes how the light it produces will affect the appearance of colored materials. So that’s what we are talking about: What will stuff look like under this light? Sometimes this is called color rendition rather than color rendering, but I think the terms should be one and the same.
Color Rendering Index, like TM-30 Rf that I am going to be writing about, seems to fit the more rigorous definition used by the CIE. Unfortunately using color rendering index as a generic term can be very confusing. Many years ago the CIE published a standard which they called “Color Rendering Index” which now refers to a specific method for calculating color rendering quality. The generic term we use today is color rendering fidelity.
Quantifying color rendering with a numerical method is hard. Initially, many years ago, the CIE had developed what was called the “Color Rendering Index.” The CRI score of a light source was computed by comparing the color of 8 reference color chips. If I was testing an LED light source at 3200K, I would compare these 8 paint chips lit by LED vs. these 8 paint chips lit by a real 3200K source. One of the advantages of this system was that the paint chips were selected from a standard color book called the “Munsell Color Atlas.” Many color practitioners and scientists had access to this and it was considered standard lab equipment many places. Therefore it was easy for people to make measurements with the materials they had on hand. This was in the era before using computational methods was as common as it is today.
There was a fear that 8 samples was not enough so CRI was then expanded to 14 color samples. Even considering this, these 14 samples do not really cover a wide enough variety of material colors and, since they all come from paints, they really only tell us how a light source makes those paints appear. What about the rest of the colored materials in my life? Like my foods or textile dyes or stone and earth materials? Furthermore, CRI only tells us that the test light source does or doesn’t match the reference light source, but it doesn’t tell us anything about how colors will change. The issues with CRI are numerous and I don’t want to dwell on them too long. Mike Wood of Mike Wood Consulting, LLC wrote a pair of articles describing the issues with CRI in 2010.
Fortunately, we live in a new era, marked by much greater understanding in the visual system and function of light. The Illuminating Engineering Society (IES) has developed a very well received new color rendering metric called IES Method for Evaluating Light Source Color Rendition – Technical Memorandum 30-18. Or just TM-30 for short. It uses much more modern methods of quantifying color, relies on real spectra measured from the world, uses many more color samples, and blah blah blah. It really is much better. I’d like to use it as the base for talking about abnormal color rendering with gels. In this post I’ll describe some just below the surface details of how TM-30 is calculated, then we’ll look at some theoretical examples and try to get an intuitive sense of what the rendering fidelity score is. Finally we will take a look at some real world examples and I’ll end it all with a few thank yous and what to expect next.
Color Evaluation Samples
As a matter of practicality, we only use computational methods to calculate TM-30. I’m going to go over what I feel is the basic theory. Just enough of a primer that I hope you feel comfortable talking about TM-30 and using the tools, which, by the way, are available for free in the IES bookstore. All YOU need, is to download the excel spread sheets included in the book store and then copy and paste the spectral data from your light meter into the tab titled “calculate.” From here on I’m going to talk about what happens after you copy that data in.
Let’s start by looking at what are called the Color Evaluation Samples or CES. The CES are a set of 99 reflectance factor curves which were carefully selected from a database of over 100,000. These are the 99 samples that are replacing that set of 14 from CRI. I’m not going to go over how or why these in particular samples were selected but I can say that I have reviewed the methods used and they were very fair and good. Here are the 99 CES reflectance spectra:
Then, to get a sense of how objects with these 99 reflectance spectra might look, we compute their color with a lightsource. This is done by taking the spectral power distribution (SPD) of the lightsource from 380nm to 780nm (roughly the visible range) and multiplying it by the reflectance factor for each wavelength. So lets say that some light source produces .03W, or 30mW, of photon radiation (light) at 620nm. If one of these objects has a reflectance factor of 50% at 620nm. The reflected radiation (light) is 15mW.
In addition to computing the reflected spectrum from our test light source, our definitions above tell us that we have to compare the light source to a reference. We need to pick a reference light source and compute the reflected spectral power distribution for the reference source as well. For low color temperatures, like the 3200K of a modern incandescent theater lamp or 2700K of candle light, the reference source comes from what’s called a black body radiator. For higher color temperatures, like 5000K and 6500K, the reference source is a daylight-like spectrum.
After calculating the reflected spectrum for the test light source and the reference light source we can compute the color of each of these samples starting with CIE XYZ, 10˚ Observer. This is another spectral calculation where we take an estimate for the sensitivity of the human visual system and multiply it by the reflected spectral power distribution. The difference between the 10˚ XYZ functions and the original 2˚ functions is a bit subtle and too under the hood for this discussion but I at least want to point out that TM-30 uses the 10˚ version, which is more modern and more correlated to our perception of large areas.
Finally, the last calculation, which is a very significant and improvement over CRI, is to calculate how our visual system compensates for lighting and adjusts to what we are seeing. There are three types of cone cells in your eyes that are responsible for color vision. The long, medium and short wavelength (LMS) cones are each sensitive to a different proportion of the visible spectrum. When a photon of the correct wavelength (energy) interacts with these cells, a chemical reaction takes place and an electrical signal eventually connecting to the brain is adjusted. During this process, there is a shift in the availability of chemicals sustaining the reaction, and, over time the cone cells get tired and they do so independently. In simple terms, this means that when you go into a room that is very yellow, the long and medium wavelength cones are more active and get tired more quickly. Rebalancing the overall color back towards white. This is called chromatic adaptation and it is similar to how the white balance setting on your camera works. Just biological and much more complicated.
Thankfully, due to chromatic adaptation, we don’t experience such a dramatic change in color as the sun sets, when switching on our incandescent or “warm” led lights at home, or after we sit in the theatre for a few minutes while our brains adjust. Largely, white paper still looks white both during the afternoon sun and the evening. Rather than just comparing the XYZ values, TM-30 uses a very complex model of vision that is known to be much more accurate and accounts for chromatic adaptation among other things. This is perhaps the single most important improvement over the mathematical system used in CRI. Chromatic adaptation is really amazing, really complex, and still an active area of research at Munsell Color Science Lab.
The very last step of TM-30 is to actually take all of those calculations for the 99 color patches and then compare them between the two light sources. In the figure to the right each arrow represents one of the color evaluation samples. The size and direction of the arrow tells you how different that CES patch looks when you compare the test and the illuminant light source. The average size of all the arrows (in 3D) relates to average color change caused by the light source. We call that average the TM-30 Rendering Fidelity score, or Rf. TM-30 also gives us a bunch of other scores, which I’ll discuss later on.
RGB TM-30 Example
Ok. So let’s start looking at some colors. All of that math stuff above is really just how we compute what the colors of the 99 samples looks like, but I still haven’t gotten to the part where we get to see some nice looking pictures. Here is the first “color render checker” of a perfect 3200K light source. It’s got a lot of squares on it and I’ll briefly describe them.
On the top of my color render checker (which I’m just going to start calling the color checker) are two areas with big squares. The left section is generated by using the reflectance data for the MacBeth ColorChecker Classic which is very popular in photography for color correcting your photos to the lighting. To the right of that there are 12 patches whose color is calculated from a skin tone reflectance database published by the National Institute of Standards and Technology (NIST). Considering that we often have performers on our stage, I thought it would be useful to include these. On the bottom of the color checker are 99 small rectangles representing the 99 TM-30 color evaluation samples.
You will see that some of the squares have a dark red outline. Although I can compute what these colors should be, most computers and web browsers are not capable of displaying them. So I’ve had to adjust the color while maintaining the same hue. If you have a reasonably new or calibrated monitor the rest of the colors are all correct. Otherwise, you shouldn’t spend too much time thinking about this particular nuance of computer displays.
I made this chart because I wanted to start looking at how the patches would change if I swapped out the reference illuminant (perfect 3200K) and various theatrical lights, like a cheap RGB LED. Notice that this light source has an Rf (Color Fidelity) score of 48.4. The color changes should be extremely noticeable at this level. Here’s the animation between the reference 3200K and the test RGB fixture.
Yep. We can see that there is a big color change here. Keep in mind that the RGB fixture is perfectly calibrated according to traditional calibration measures, meaning white is still the same perfect white as the reference source would show. Now let’s look at the spectrum and the TM-30 report that you might see on a fixture data sheet.
In the top graph, the reference spectrum, 3200K in this case, is shown in black. The spectrum from my RGB LED light is shown in red. Arguably the most important part of the TM-30 report is the big colorful square, called the Color Vector Graphic. The CVG shows you several key pieces of information. Starting on the bottom, the Correlated Color Temperature (CCT) is shown as 3200K. To the right of that we have the Delta u, v score. Often this is written as Duv or ∆uv; Duv value tells us the difference between the white of the test fixture and the white of the reference light source. For LED fixtures with three or more primaries, this will always be 0 if the fixture is calibrated correctly.
Moving up from there we see the big hue circle with 16 divisions. The 16 sections are called the hue bins. Around the circle are several arrows which tell us the average color change for each hue bin. Each arrow itself is made from an average of some number of samples in that hue bin. Large arrows moving out from the circle tells us there is a big change in chroma, or saturation. Arrows moving sideways tells us something about the hue shift. Lastly, the average of these 16 arrows (in 3D) that tells us the Rf score.
At the top of the CVG the overall average change in color is reported as the Rf score. The average change in gamut, similar to saturation, is shown as the Rg score. A perfect RF score is equal to 100. Gamut, as seen here, can be greater or less than 100 depending on whether colors are increasing or decreasing in chroma / saturation on average. A score of 122 tells us that colors are greatly increasing in chroma under this light source. If you look at the rendered color checker above, I think you can see that too.
To the right of the CVG we have the “Local Chroma Shift” and “Local Hue Shift” bar charts. These tell us the exact values for the arrows for each hue bin. It’s very important to remember that most of the time you will see the local hue shift reported in radians, not degrees. So if you want to know how much each hue bin is shifting in 0-360 degrees, you need to convert. As a rough estimate you can look at the left / right direction of the arrows and compare it to the width of the hue bin. Each of the hue bins on the CVG is 22.5 degrees.
Putting Rf Into Perspective
Now I know how to the read the TM-30 report, but if you are like me then you are wondering “Ok. What’s a good enough Rf score.” That’s a really hard question, and of course it depends on application too. The Rf score of “flash and trash” fixtures probably doesn’t matter that much, but if I’m lighting a painting gallery I might only accept a TM-30 Rf score of 95+. But what does an Rf of 80 look like compared to an Rf of 90? What about 70 vs. 90? I still don’t really have an intuitive sense about what these numbers mean.
To help us get things into perspective I decided to take a look at several possible fixtures with an RF score ranging from 90 all the way down to 30. I used the spectral data from a set of ETC Source4 LED Series 2 Lustrs to generate the possible light sources. What I’m about to show you are all real possibilities with that particular 7 LED fixture.
Wow! Quite a significant change when the animation loops back around to Rf=100 from Rf=30. Individually, each step seems kind of small by itself. The cyan square near the skin tones changes the most dramatically to me. I also find it interesting to watch this in my slightly peripheral vision. What do you see?
Let’s take a look at the TM-30 report for this progression. At the top, I’ve modified the spectral graph a little bit. The dashed lines show the power of each of the LEDs that are active. I find it interesting to see the composite spectrum and each component spectra. Lastly, keep in mind that all of these spectra are calibrated for 3200K and 0 Duv!
I’m obligated to end this section with a little bit of a disclaimer about what this means for other illuminants. The series of spectra here, from Rf 100 to Rf 30, are just some possible spectras. There are an infinite number of possible spectra that have the same Rf scores but very different hue or chroma distortion. There is no guarantee that if you find an Rf = 80 fixture it will match my color checker rendering above. In fact, that’s almost impossible unless you use the same LED Source4 as me and calculate the same control values as I used. It is not the case that all Rf 30 illuminants would produce the same color checker or color vector graphic.
In phosphor converted white LEDs, which are very common for consumer lighting fixtures, you will typically actually see a decrease in red saturation. Unlike the increase in my Rf series above. There is an ongoing effort to define a lighting preference metric. Preliminary evidence has shown that for low Rf scores an increase in Rg, and particularly an increase in red saturation, is preferable. This preference is not well understood in the lighting science community yet but you may see it come to fruition in the next several years.
What TM-30 Rf is good enough?
We’ve seen that TM-30 is a very powerful new tool and the TM-30 intermediate report (which I’ve used above) gives us a ton of new information about our lamps and light sources. For someone who was previously used to looking at just CRI scores, this new information can be really intimidating so I thought I’d talk about what I would do as a lighting specifier.
First off, unfortunately with a color rendering metric that has greater power comes greater responsibility to use it. Using TM-30 is no longer as simple as looking at a single number and saying “Yes this light source has good enough color rendering.” Particularly for fixtures that have lower scores in the 70s and 80s. You might be evaluating a fixture with Rf = 78, which really desaturates red colors, like some phosphor converted LED systems. Next to it you might have a fixture made with RGB + W LED tape and a poor Rf of only 72 for your target color temperature. But the RGB fixture boosts the saturation of red and blue colors which is preferable in your lighting plan. You will be able to see that by looking at the color vector graphic and seeing where the hue or chroma shifts are.
For example, in a dining experience with lots of earth tones you really want to highlight the wood grains and architecture. This is not task lighting and a lower Rf fixture that has a chroma boost in the red-ish hue bins might outperform, in terms of the designer preference, the higher Rf fixture. Above the tables, where people need to read menus and see the chef’s glorious creation, you might want a higher color rendering fixture which shows off the lavender and mint garnish on top of lemon roasted finger potatoes and medium rare sirloin. Hmm… I’m hungry.
Finally, in theatrical application: I’m not sure Rf score is the be all end all. The Rf=30 settings shown above provide a really striking look while still maintaining the white point. In some design, I might want that. My thoughts for theatrical design are that I want a fixture which is capable of very beautiful white light, but has control options available so that I can find settings which really distort the color vector graphic. Theatrical lighting design is just peculiar enough that we need the highest color fidelity available, and we need the flexibility to not use it. More on this idea in future blog posts.
If you have the time and want to learn more, I highly recommend this video lecture given by Dr. Kevin Houser at ETC Workshop in 2016.
Real World Examples
I thought after all this reading and writing it might be nice to end with a few images and real examples of lighting fixtures I’ve seen. I’ll present the Source4 LED Series 2 Lustr and Philips SkyRibbon with RGB + Mint and Amber LEDs. If you have a fixture or lighting plan you’d like me to comment on or do some color rendering calculations, feel free to email me or leave a comment.
Source4 Lustr 2
At 3200K the Source4 Is a very capable fixture. It’s almost impossible to notice the color shift in the color checker at all. The worst hue shift is in the 13th hue bin, lavender colors. Here, when illuminated by the Source4, the hue bin shifts by about 5.1˚ towards blue. Unsurprisingly, bin 12 has the worst chroma change, with a very small 5% increase.
Also a capable fixture at 3200K the SkyRibbon has a very good Rf score of 91.4. The SkyRibbon in model used here has mint and amber LEDs in addition to RGB. The worst hue shift is in the 11th and 13th hue bins. Here, when illuminated by the SkyRibbon, the hue bins each shifts by about 3.4˚ towards blue. Less overall hue shift than the Source4. Chroma is most significantly increased for the 6th hue bin, about 7% for greenish yellow objects.
I hope that you have learned something new from all of this. Like I said at the beginning, color rendering is one of the most interesting aspects of color science. We use artificial lighting specifically so we can see things and it’s interesting to think about how the choices and trade-offs in the design of those lighting fixtures really impacts quality of what we see. I’ve only just barely scratched the surface too. Next, I’d like to start thinking about how chromatic illumination affects color rendering and how well these LED fixtures can recreate or provide greater performance and flexibility than my Rosco and LEE gel swatches.
I’d like to thank ETC and Philips Lighting (now Signify) for their equipment donations to Munsell Color Science Lab where I work, and Dr. Michael Murdoch for sharing measurement data as well as several important lessons about composite lighting spectrums.
Lastly, I hope I’ve helped you gain a greater understanding of the TM-30 metrics for color rendering, but I don’t expect I’ve done so perfectly. I would love to address your specific questions as I have time and would really appreciate your comments, questions or feedback. Learning is not a singular endeavor and I hope that by talking about your thoughts on the subject we will both benefit. I will be hosting a reddit AMA about color science and entertainment lighting technology on July 7th on /r/techtheatre. If there’s anything else you’d like to ask me it’s a great opportunity to ask in a public forum that will benefit many people and young designers.