A common trend in fixture design among LED lighting fixtures targeted at the entertainment lighting market is the inclusion of a lime LED. The lime LED is a blue pump phosphor converted LED which results in a broadband emission. This broadband characteristic is typically avoided by most LED buyers, who are interested in producing products with a wide gamut and high maximum saturation. However, for the goal of improving color rendering, broadband LEDs are critical. In this blog post I provide several renderings which represent different types of lighting fixtures. In these renderings you will see that the fixtures which include lime LEDs, among other colors, have dramatically better color rendering than their counterparts.
What makes the addition of a lime LED so impactful in theatrical lighting fixtures? And why lime? Why not mint, or amber, or a broadband cyan? The answer is due partly because of commercial availability of these lime LEDs but there are important colorimetric factors as well.
In the early days of LED fixtures coming into the entertainment market, many devices only used red, green, and blue LEDs. There is a certain amount of intuition towards this; color is three dimensional, and we have three cone types in our eyes. Display devices had been operating in color for years using red, green, and blue phosphors or a range of color filters. Where the observer is looking directly at the light source, three primaries is just enough to create a wide range of colors for the viewer and those colors can look as real or as vivid as looking at some objects themselves. Three LEDs could therefore be enough to provide a wide range of color options for lighting designers AND the colors should look as good as any monitor or television. However, there is a key difference between displays and the illumination of objects or a stage.
In the case of illuminating different objects, the observer is not directly looking at the illumination but rather they are looking at the colors produced by the interaction of the illumination and some reflectance properties. The color that the observer sees can be estimated by understanding the color sensitivities of their retina, AND the reflection of some of the light from the illumination. The objects’ reflective properties are critical to producing color in this scenario. In the case of a modern display, the observer is directly viewing the illumination from the color filter array or OLED matrix and there is no object reflectance to be concerned about. In order to quantify the performance of different light sources, then, we must use a metric which includes some quantification of how different objects reflect light and the resulting appearance when illuminated by different sources.
Color Rendering Index
Lighting designers and consumers have known for years that they must compare the appearance of different objects in order to evaluate lighting quality. The CIE CRI metric was initially designed to solve this problem by averaging the color difference of 8 color samples illuminated by a test and a reference source. The reference source is selected to be the black body radiator or daylight spectrum with the same correlated color temperature (CCT) as the test illuminant. Later analysis showed that for a large range of illuminant spectrums (58 tested) just 5 of these color samples can be effectively calculate the Ra metric (colloquially, the CRI score)1. As colorimetry has advanced and new light sources with dramatically different spectra have become available CRI has become outdated and the CIE itself no longer recommends its use2.
Recently, the Illuminating Engineering Society has pioneered an effort to develop a new color rendering index called TM-30. TM-30:18, the latest edition of the standard, includes several metrics and recommended visualizations. TM-30 contains many other useful metrics in addition to a color fidelity metric dubbed Rf. Most importantly it is based on the color appearance of 99 color samples instead of 8, or 5. Many manufacturers are adopting these new metrics for evaluating light sources designed for theatrical or architectural usage. For the visualizations presented in this post, I have included the 99 color evaluation samples (reflectance spectra) from the TM-30:18 standard. When looking at the different light sources evaluated, pay attention to how these samples change color. This is the change that is quantified by the Rf, Rg, and other scores and graphics presented in a TM-30 vector graphic. For more information on TM-30 I highly recommend this keynote lecture by Dr. Kevin Houser hosted by ETC during “Workshop” 20163.
The Color Checker
Color Checker Classic
The top left portion of the digital color checker is based on reflectance measurements provided by the Munsell Color Science Laboratory4. This target was selected because of its common usage in imaging and printing research. In theory, it provides colorimetric matches for common objects such as skin tone in the top left, flowers, plants, sky color, and printer inks. While a useful target for real world applications, in this case we are not bound by what traditional paints and pigments are capable of and so having real samples of reflectance data in our digital checker is more useful. The other two sections are comprised of those samples.
NIST Skin Tone Data
The top right of the digital color checker contains 12 reflectance spectra synthesized from 100 real skin tone reflectance measurements provided by the National Institute of Standards and Technology5. From these 100 reflectance samples, 10 were synthesized by k-means sampling. The final two skin tones, the two at the bottom right of the skin tone samples, are the minimum and maximum total reflectance of the NIST dataset. Figure 3 contains the final 12 skin tone reflectance factors plotted against wavelength.
TM-30 Color Evaluation Samples
The final section of the digital color checker is made up by the 99 color evaluation samples (CES 1-99) of the TM-30:18 standard. These samples were chosen very carefully from a large database of reflectance measurements so that 1) they evenly sample the CAM02 color space, and 2) uniformly sample spectral features. The selection of the 99 CES started with a large database of 105,000 reflectance spectra including thousands of skin tones and other real objects. The process for selecting the final “large set” was based around the two previously mentioned characteristics. With those goals in mind, 4,900 spectra were selected for gamut and fidelity computations. Wanting to reduce the number even further, to make it easier for end users to calculate, the final 99 CES were selected to sample the large set where for some 5,000 test illuminants the difference in Rf and Rg scores was minimized when calculated with the full or reduced set of reflectance spectra6.
Rendering for 5 Lighting Fixtures
Selection of the Fixtures
In order to visualize the difference that the inclusion of a lime LED can make, 5 lighting fixtures were simulated. For each simulated lighting fixture multiple primary spectra were selected and could be varied by a scaler value. The final illumination spectrum can be regarded as the sum of each spectral power distribution multiplied by its assigned scaler. Or, more simply, the fixture output is the sum of some amount of each primary channel. This type of model is a simulation of the way that a lighting console might control the output of each LED group of a lighting fixture.
The first fixture is a simulated laser light engine. For this lighting fixture, three spectral power distributions for consumer grade lasers were simulated. The laser SPDs have peak wavelengths corresponding to the wavelengths required for a ITU Rec. 2020 compliant laser projector. These wavelengths are 630, 532, and 467nm.
The second lighting fixture utilizes red, green, and blue LEDs. These 3 primary spectra were selected from the 7 primary spectra of the Electronic Theatre Controls Source 4 LED Series 2 Lustr. Similarly, the RGBL lighting fixture consists of the deep blue, green, red, and lime LEDs from the Series 2 Lustr LED measurements. In addition to the ETC Source 4, measurements from a Phillips Sky Ribbon LED lighting fixture were used for a set of renderings. The Sky Ribbon LED matrix consists of red, green, blue, white, and mint LEDs. The mint and white LEDs are also a broadband phosphor converted LED and serve the same purpose as lime LEDs in the spectral composition. Its chromaticity lies slightly above the daylight locust, slightly farther “south” the lime LED.
The lights and their basic measurements used for this blog post have previously been characterized in Murdoch 20197. Similarly, the method for calculating the power / scaler value for each fixture channel used a “colorimetric-plus” methodology as proposed in that paper.
Calculating Per Channel Intensities
For any fixtures with exactly three primary spectra or primary colors there is exactly one ratio of channel intensities that can create a target color and brightness. For these demonstrations the target illumination color was D65 white with a luminance of around 450 candelas per square meter. For the purposes of this rendering, there was no upper bound on channel values. Accordingly, each simulated fixture was capable of reaching the target brightness. This obviously does not translate well into the real world, however it made the math and rendering much easier. In a real world scenario, it’s imaginable that you might have a lower brightness, but if the simulated light source represents the overall illumination you would adapt to that brightness in situ.
For the simulated fixtures with more than 3 independent emitters each channel intensity was calculated by using a “colorimetric-plus” solution as coined in Murdoch 2019. A computational linear optimizer, MATLAB’s fmincon, calculated the channel intensities with the constraints that the resulting color of a white reflector match to D65 white, and that all of the channel intensities were positive. That is, there cannot be negative light. The optimization goal was to be a match to D65 spectrum weighted by the Viggiano function for L* 768. Viggiano proposed that each wavelength has different value in solving the metamerism problem and proposed a method for calculating weights for each wavelength based on its importance for the target color in CIE LAB color space. Using white as a target color for the Viggiano weights may not be the best choice for weighting a spectral difference. This is one possibility for further research, however in the absence of a a more specific target this approach is recommended by Berns9.
The final illumination spectrums calculated using the above methods can be seen in Figure 1 where each is compared to the target, D65, in red. Each illuminant is normalized to have exactly the same chromaticity and luminance as the D65 target spectrum.
The above slide show should hopefully demonstrate the color rendering properties of these different illuminants. You may also download these images on black, white, and transparent backgrounds here. I highly encourage you to load them into your favorite image previewer and click back and fourth between each simulated fixture and the daylight target. To address the question, why lime? I suggest looking back and forth between the red green blue and red green blue lime illumination renderings. Here you can easily see where the inclusion of the lime LED helps give much of the spectral power necessary to improve fidelity. Finally, it should be quite clear that using highly saturated and efficient lasers as the illumination source will never be satisfactory for cases where color fidelity is a concern.
Discussion and Future Work
One thing that the above slide show also makes clear is that improved fidelity does not require a lime LED specifically. The Phillips Sky Ribbon rendering looks even better than the RGBL rendering, or at least it has a higher Rf score, and it substitutes lime with mint and white. However, this is heavily impacted by the choice in target spectrum. If instead these renderings were based on a tungsten or traditional incandescent target spectrum then perhaps the yellowish-ness of the lime LED would be more important than the broad band power provided by mint and white. Future work could address this by performing the same renderings with CIE illuminant A as the target spectrum and including a chromatic adaptation model in the rendering process. After all, even though we think of incandescent light as “yellow” in fact, as your visual system adapts to it, it becomes just as white as the sun, and a rendering demonstration like this should reflect that.
Additionally, the usage of the perceptual Viggiano weights when calculating spectral difference is of some interest. Perhaps averaging the Viggiano weights for all of the CES samples would provide a general index of metamerism and be much easier / faster to calculate than the colorimetric appearance of each color sample. Would optimizing the illuminant spectrum based on this index result in the same per channel intensities as optimizing for TM-30 Rf?
I would like to thank the Munsell Color Science Lab and Dr. Michael Murdoch for providing the time, resources, and data required for this project. Dr. Murdoch provided the measurements of the Phillips and ETC lighting fixtures used. MCSL provided much of the standard observer data, D65 spectrum, and other useful color data. Finally, many of the matrix methods used for calculating each color patch can be found in Billmeyer and Saltzman’s Principals of Color Technology, 4th ed. This book also provided the idea of using Viggiano weights as the objective function during optimization. I highly recommend it to anyone interested in learning more about color science.
Munsell Color Science Lab has several available funded PhD positions opening in the 2020-21 school year. The application deadline is January 15th and it is a very fun place to work.
- Guo, X., & Houser, K. W. (2004). A review of colour rendering indices and their application to commercial light sources.
- CIE. (2015). Position Statement on CRI and Colour Quality Metrics. Retrieved January 9, 2020, from http://cie.co.at/publications/position-statement-cri-and-colour-quality-metrics-october-15-2015
- Electronic Theatre Controls (2016). Keynote – Dr Kevin Houser edit – YouTube. Retrieved January 9, 2020, from https://www.youtube.com/watch?v=5fQkau7i51k
- Unknown. (n.d.). Useful Color Data. Retrieved January 6, 2020, from RIT MCSL Educational Resources website: https://www.rit.edu/science/munsell-color-science-lab-educational-resources
- Cooksey, C. C., Allen, D. W., & Tsai, B. K. (2017). Reference Data Set of Human Skin Reflectance. Journal of Research of the National Institute of Standards and Technology, 122. https://doi.org/10.6028/jres.122.026
- David, A., Fini, P. T., et al (2015). Development of the IES method for evaluating the color rendition of light sources. Optics Express, 23(12), 15888. https://doi.org/10.1364/oe.23.015888
- Murdoch, M. J. (2019). Dynamic color control in multiprimary tunable LED lighting systems. Journal of the Society for Information Display, 27(9), 570–580. https://doi.org/10.1002/jsid.779
- Viggiano, J. A. S. (2002, June 6). Perception-referenced method for comparison of radiance ratio spectra and its application as an index of metamerism (R. Chung & A. Rodrigues, Eds.). https://doi.org/10.1117/12.464650
- Berns, R. S. (2019). Billmeyer and Saltzman’s Principals of Color Technology (4th ed.). Hoboken, NJ: John Wiley & Sons Inc.
Disclaimer: In 2014 and 2016 Tucker Downs was employed by Electronic Theatre Controls.