Welcome, everyone. This is Andrea Wilkerson with the Pacific Northwest National Laboratory. Welcome to today's webinar, TM-30-what have we learned in the past two years, brought to you by the U.S. Department of Energy's Solid-State Lighting Program. Today's presentation will review the basics of TM-30 before exploring recent and ongoing research.
Our presenter today is Michael Royer. Michael Royer is a lighting engineer at the Pacific Northwest National Laboratory, where he supports the U.S. Department of Energy's Solid State Lighting program. Michael is a member of the IES Color Committee and was the chair of the IES Color Metrics Task Group that developed TM-30. Thank-you.
Thanks, Andrea, for the introduction. So as you all know, the topic of today's presentation is TM-30, and specifically what we've learned in the past two years. So I'm going to begin today by going over some of the basics of TM-30 as a quick instruction. Because I think one of the things that I've personally learned is that it takes a whole lot of outreach when you're introducing a new method. I'll follow that up by going over some research that we've done here at PNNL and elsewhere, and conclude by looking at some adoption issues and logistics.
Just another note, if there's some of you on the line that this is your first introduction to TM-30, this might go pretty fast for you. There were a couple webinars that we gave almost two years ago to the day, the first on TM-30. And those are available on the web page that you see linked here on this title screen. There's also the URL on the end screen that'll be up during questions.
So, to begin, what is TM-30? In a nutshell, it's a method for evaluating light source color rendition that has this core system of key components. And that would be the model of human color vision. It's called CAMO2-UCS.
And if you're an end-user, you may never need to know this. But this is sort of the black box behind everything. It relies on a standardized test of 99 color evaluation samples called CES. And it uses a system to establish a reference baseline, either Planckian radiation or a CIE D series illuminant. So if you're familiar with CRI but not TM-30, you can think of TM-30 as a modern version of CRI, plus a bunch of additional information, that you can use, if you need to, to better pick how colors will appear in an architectural space. The reference illuminants there.
So, from that system, then, we can calculate a whole suite of objective characterizations of light source color rendition. At the highest level of the average values, we have a fidelity index that calculates the average magnitude difference for all 99 color samples. And I'll go into detail more about each of these in just a minute. We have the gamut index, the average gamut area, and we also have 16 local chroma shift values and 16 local color fidelity values based on different hues.
Supplementing that numerical information, we also have the color vector graphic, which provides a quick visual representation of hue and chroma shifts across different hues. Now, things that we've learned in the past two years – it's important to convey what TM-30 isn't. So it isn't a test that can be passed or failed. You can't be certified according to TM-30. TM-30 doesn't inherently include a measure of color preference. But we're going to talk a lot about that later.
It doesn't inherently include a set of specification criteria or other guidance on how the numbers should be applied. It's not a way to address color consistency or color shift over time. So it's specifically just objective characterizations about how a given light source appears different from the reference illuminant.
So the third part – and this is a little bit of the how to use it – it's really intended to be used so that these objective characterizations are combined in various ways to predict perceptual outcomes. Ultimately, what we're trying to do is understand how a space will appear. Is it preferred, does it look normal or natural, vivid, saturated, is it acceptable?
And in my opinion, this can't be done with a single number. Because it's all based on the context of the architectural environment. That's the color palette, the application, the design intent, the viewer's adaptation states, to how long they're viewing it, the culture, and their background.
So I'm, I guess, an amateur photographer as well. And I think of this in terms of video editing. Now, we don't have one slider that we can use to just adjust in a magic way and make a picture or photograph more pleasing or better-liked. There's lots of additional tools if you're in Photoshop, or whatever software you're using, that you can adjust individual hues, get that more granular information to help you edit a photograph in a way that you would like to.
So, diving a little bit deeper into these metrics. So we can start with the average color fidelity. If I have my 99 color samples here squished into a two-dimensional plane, average color fidelity is telling me the average difference between those red and black dots. So it tells me nothing about which way they're moving overall, or for each individual sample.
We can start to supplement this. And one of the key things we're doing in TM-30 is dividing this plane into 16 individual bins. What that allows us to do is calculate the average coordinates in each of those bins.
The first thing we can do with those averages is we can calculate the area of a polygon formed by those 16 points. We can calculate one for the test source, one for the reference illuminant. So we can calculate the ratio of those two areas. And that gives us our Rg value, the gamut area, which is roughly going to tell us, on average, are we increasing or decreasing saturation? But as you can see from this graphic, I could be increasing the saturation, or chroma – use those synonymously here, probably, today – in one particular hue region, but be decreasing it in another.
So, one of the most important things that I think TM-30 brings over some previously proposed metrics is that it has this third component. And I'll define that as gamut shape. And TM-30 includes a graphical way to explore that, the color vector graphic, as well as these local values, local chroma shift in particular being important.
So, now, just to think, if we want to know information about a particular hue and what's going on there, take for example hue bin one. And it was given the label one because it's sort of this true red color that we might be familiar with from CIE R9. And if I can calculate the color shift in that hue angle bin, I can make this into a graphical representation first by normalizing those black dots into a circle. And this has no real mathematical meaning. It's just a way to make this more visually interpretable.
So what I can do with that individual shift, then, is transfer the vector to the circle – I can do that 16 times for each of these vectors – connect the endpoints of those vectors to form this red shape with a test-source polygon. And this begins to tell me how color rendering is changing across the different hues.
So, specifically, interpretation of this graphic. When I see an arrow moving tangentially to the black circle, that's showing me there's a hue shift. So, in sort of the yellow-green region for this particular source, it would be shifting the colors from more green to more yellow. I can see when an arrow is pointing in from the black circle towards the red that there's a decrease in chroma. So the blues, blue-greens, and reds are being decreased in chroma for this particular source. When the arrow is pointing out from the black circle towards the red line, it's an increase in chroma. And we can see that occurring for yellows and blues for this particular sample.
Now, this numerical information – now, I've said this for a couple years now – I think it's very powerful. And it allows someone to quickly identify how a space might appear once you're familiar with interpreting this graphic. It can take a little while to get used to that. But I do think it's quite powerful.
At the same time, it's pretty hard to write a spec based on a picture. So there is a numerical component that goes along with this graphic. And in particular, that's the local chroma shift.
And I'll throw up a fancy equation here which isn't actually all that fancy when you get down to it. And that's just for reference. But the simple interpretation is that local chroma shift is the purity radial difference versus the reference. So if I looked at any one of those arrows and I drew a line to the center of the circle, it's the distance I'm moving in or out from that black line.
So if I'm moving in, it's negative. If I'm moving out, it's positive. And either way, it's represented as a percentage, which allows us to apply those values to any sample, regardless of the absolute chroma.
So let's look at two examples in sequence here, to show the importance of gamut shape and why it's a really important supplement to average color fidelity and gamut area. So this source here has the Rf value of 83, Rg value of 98. We can see that in the graphic, the area that's red is being desaturated a little bit. Yellow's being increased in saturation a little bit. And we can follow the local chroma shift values in the bar chart on the left and see that they mimic what is occurring in the graphic on the right.
Now, I'll change this to a second source. We can see that it has approximately the same Rf and Rg values, but it's distorting colors in the complete opposite way. So this one is increasing the saturation of reds and greens while decreasing the saturation chroma of yellows and – it's not really in the blues. It's just a small shift in the blues.
So I can go back and forth, and you can see that these two sources, even though they have the same average fidelity and the same average gamut area, approximately, would render colors pretty differently. And we've asked this in humans' experiments. And these are perceived differently. So it tells us we really need this additional information if we're going to have any hope of using color-rendering metrics to predict how a space is going to appear when it's illuminated.
So, just to orient you a little bit further to what you can expect to see in a color vector graphic and local chroma shift values, here are three example commercially available sources – one, a linear fluorescent lamp on the left, the middle, a ceramic metal halide, and the right, a phosphor LED, kind of your standard, low-80s CRI product. So we can see that all of these actually have a very similar distortion pattern, as seen in these color vector graphics. So they're decreasing red saturation and tend to increase yellow-green saturation.
And if I change this plot – so this is now looking at local chroma shift values for about 212 commercially available products in the TM-30-15 library – we can see that this pattern of distortion is very much typical of commercially available products. And that happens because this is the most efficient way to make a light source that's designed to meet a minimum CRI value. We can see, really, the only exceptions to this is one particular RGB LED, a neodymium incandescent, and one hybrid phosphor plus direct red LED. Almost all of the others are decreasing red saturation – that's the lines being below 0% for hue angle bin one – and they're increasing in the yellows and yellow-greens, decreasing saturation in the sort of true green, blue-green, and increasing saturation in the blues. Because we tend to get these elliptical distortions in the color vector graphic.
It's important, though, to know that that distortion pattern doesn't have to happen. So we can look at some experimental LED sources in this case. And we can see that we can distort colors in any way we really want to, at any given fidelity value or gamut area value, and at any particular chromaticity. Those are all dependent.
So we can see the bottom one on the left there is increasing reds. It has a relatively low fidelity. We can see in the middle, again, increasing red-green, a little bit higher fidelity, higher overall average saturation Rg. And the one on the right is still increasing yellows and sort of blue-purples, but not decreasing the saturation of reds. All have different possibilities that we can see.
Again, to illustrate the range of possible values, here we're looking at the huge angle bin local chroma shift. And so we can see over a set of about 5,000 sources, the range of possible values for any of these local chroma shifts is quite large. Now, if I limit that simply to sources with an average color fidelity Rf value of 70 or greater, we can see that the range is somewhere about negative 25% to 25%. And it varies a little bit by the specific hue angle bin. So there's more possibility to increase or decrease saturation of reds and greens than there is yellows and blues.
So that's all for the basic review of TM-30. And again, there's some papers on this topic and a lot more information on the DOE web page that you can look into on your own to try and get a better understanding of the metrics themselves. I'm going to shift my focus now to experiments that we've conducted at PNNL and a review of some that have been conducted elsewhere. So I've looked at trying to understand how these objective characterizations provided by TM-30 relate to our perceptual experiences in an environment.
So this first experiment – this is published in Lighting Research and Technology, and it's also available from the DOE SSL web page – was completed in 2015. We abbreviated here, CREX1 – color rendering experiment one. And you'll see a CREX2, and a 3, and a 4. This one particularly looked at 3500 K.
And on the Planckian locus, 26 different lighting scenes in this environment shown on the right side of your screen. We have four levels of Rf, five levels of Rg. And at any given of those combinations, we were trying to minimize or maximize the red chroma shift in hue angle bin one in order to vary the gamut shape.
So we varied all three of these key parameters within TM-30. We asked the question, on a rating scale from one to eight, is it dull or saturated, normal or shifted, and do you like it or dislike it? And subjects viewed this experimental room for a minimum of 30 seconds before providing a response, the lighting scene was changed, they went back into the room and repeated.
So just to illustrate exactly how these lighting scenes were divided again, we have these four levels of Rf, five levels of Rg, and at any given point, we were shifting this gamut shape to be increasing or decreasing red saturation. So I have a photographic illustration here of mostly what was seen. Again, you're going to see this, probably, a little differently on your screen than what was actually present for the participants in the space.
But I'll go back and forth a little bit between these two. I like to look at the pumpkin or the orange pepper, sometimes the tomato or the red Coke box. And you can see some of the shifts that can occur, even at the same fidelity and same gamut area values.
So, jumping into the results, we had these perceptions that we were interested in. And then we'd look at different parts of TM-30, or even CRI, and see how they were correlated with the participants' responses. So you can see here, perceptions of normalness – which we also described to the participants as naturalness – is not really related in any way to color fidelity, not CRI or IES Rf. And it's pretty weak correlations here in both cases.
So that's really just a limitation of characterizing only the average difference. When we walk into a space, we're not taking an average of all the colors, and seeing how different they are, and making an assessment based on that. That's not our perceptual process.
We can see, however, that if we look at the local chroma shift value for hue bin 16, RCS H16 here on the left, that's correlated a little better with our perceptual experience as it relates to normalness or naturalness. However, you can still see that at a given level of red chroma shift, it's pretty significant variation in the ratings. And that happens because those that are rated better at a given value of that local chroma shift had a higher average fidelity value than those that were rated more poorly.
And so we can actually combine Rf and RCS H1 to produce the best predictive model of perceived normalness. In this case, our R squared, very high, 0.85. And we can see that that gives us a very good indication of whether the source would be perceived as normal or shifted. So again, that's identifying that reds are very important to our perception.
And if we look at color psychology literature, sort of outside of architectural lighting, there's a whole body of research about why red is very important to us perceptually. If you look in a candy or consumer goods type of aisle in a store, it's no coincidence that you'll see a lot of red products. It really grabs our attention. And that's the case, too, in this experiment. And we got additional data that's not in this presentation to illustrate how strong that identification with red objects was.
So if we move to the second perception that we studied, that of saturation or vividness, we can see that Rg does a fairly good job of capturing that perception. And you can, again, see a vertical spread in the responses at any given level of Rg. And so we can see that a very, very strong predictor for perceived saturation is again this red chroma shift, almost a perfect correlation with perceived saturation across these 26 different lighting conditions. So that, again, indicates that this perceived saturation level is tied to our perceptions of both normalness, and, as you'll see on the next slide, of preference for the different lighting conditions.
So, moving to preference, again, CRI is a terrible predictor of color preference. In this case, it's actually negatively correlated with color preference. Same for IES Rf. It's just not an indication of whether we're going to find the color in a space pleasing or not pleasing.
I'm going to take one little aside here. From an experimental perspective, when we have these 26 lighting conditions, those are the results we found. This is squishing those two graphs on the previous slides together. If, however, I had only looked at sources that didn't vary gamut shapes, so that they all desaturated red chroma, then my correlations change. And it looks like color fidelity is a pretty good metric. So it's just important to understand, and as we learn more about color rendering, this third element of gamut shape, how important it is, both when we're designing experiments and developing new metrics, to be able to account for these differences in chroma shift.
OK. Back to understanding the different metrics that are related to preference. If we plot this two-dimensionally and look at where the values would align in trying to identify criteria for preference, we can see that we could do that on the left with Rf and Rg. However, the sources identified with the black circles there, those are the ones where we get the most opposition in gamut shape.
And so we can see, again, at that same Rf and Rg value, we can't distinguish between preference just using Rf and Rg. We have to move to the graph on the right, where we're using Rf and red chroma shift. And there, we can separate out the better-preferred sources versus the ones that were not liked very well.
So this is an extremely complex chart. And I want to use it just to illustrate one point – that it is really the red chroma shift that is important in identifying these perceptions. And you'll see that the correlation for the red hue angle bins – predominantly red – is much higher than for any of the other hue angle bins. Just to wrap this up again, and look at the combined models for multiple metrics, we can see that we can get very strong correlation with perceived preference using models that combine average color fidelity, Rf, and red chroma shift in either hue angle bin 1 or 16.
So those numbers are all well and good. It's pretty complex when you look at it. And I think it's important to try and boil this down into a specification that could actually be used in practice.
So here, what I did is used the rank order of preference to try and identify some criteria that could be applied for any of these perceptual correlates. Again, I'm always careful when I present this slide to indicate that this is for this specific context. And applying this specification criteria to other contexts, you have to be careful doing that. Because you don't know, in particular, if that other context is going to be perceived in the same way as this one.
For example, if you had a space with no red in it, designing specification criteria around red would probably not be that useful. But in the generic consumer environment, what we're looking at is if we want normal, we're looking for a modest increase in red chroma with reasonably high fidelity values. If we want to increase perceptions of saturation, just jack up the red saturation as much as possible. If you're looking for the preferred zone, you can extend that increase in red saturation a little further, which necessitates a little bit of further range in average color fidelity.
And I put the third spec in there for Rg. It's not critical. But I think I'd rather overspecify this than underspecify it.
So, what about other chromaticities? That was our next question. And we completed an experiment in 2016 to look at that. We examined five different chromaticity groups, both on Planckian and below Planckian, negative 0.007 PUV.
Within each of those groups, we had essentially the exact same 10 color rendition conditions, based on their Rf, Rg, and gamut shape, for a total of 50 different lighting scenes. We asked the same questions as before. And in addition to that, we asked, is it unacceptable or acceptable?
When we were varying chromaticities here, it was really important that we had the participants chromatically adapt. So we did that in a separate room. And then we showed each block of 10 color renditions for each group together in a randomized order. Again, minimum of 30 seconds in this experimental room before making an evaluation.
So, a summary of the results from this experiment really defines these same factors that are being influential across all these different chromaticity groups. And that's the red chroma shift, which is really tied to our ratings of preference, and sort of this ideal spot where it's sort of neutral to modest increases in red chroma shift. And that's directly related to our perception of saturation in the environment.
Again, very strong correlation using the different TM-30 metrics. So we could see, if we relate the actual perception ratings to one another, that preference and normalness are very highly correlated. And both of those are very highly correlated with saturation. But there is the slight difference that our ratings for preference are actually maximized at a slightly higher saturation than our ratings for normalness.
And again, this is exactly the same finding as in the first experiment, where I think it identifies that the participants were able to recognize that it was increasing saturation, and that's what they liked about the environment. So, again, another graph with a lot of information here. I'll try and distill it quickly here in this slide. And this is looking at the chromaticity effect relative to the effect of color rendition condition.
So we can see across these five different groups, at different chromaticities, there was some variation in the mean rating across those 10 conditions. In particular, the two groups at 2700 K, the sources that were on Planckian on average, were rated more disliked than the sources that were at a negative PUV. But at 4300k, there was no effect from PUV.
And overall, there's no effect from CCT. So we just have this interactive effect where PUV matters at certain color temperatures but not at others. At the same time, this graph shows that describing the differences in rated perceptions was really the color rendition conditions. So there's more variation within each of these blocks than there is between the blocks.
Now, again, boiling this down to specification criteria – because I think this is something that I've learned over the past two years is really what a lot of the industry is veering for – an easy way to apply these values, to understand, this is what I need to look for. So I isolated the various color rendition conditions and combined those. And I found that I could, with the set of criteria of IES Rf greater than or equal to 75, Rg greater than or equal to 100, and RCS H1 red chroma shift between negative 1% and 15%, that isolated the sources that had greater than 89% acceptability ratings. In this case, acceptability and rated preference were very highly correlated.
Now, there's nothing really to say that 89%, or 90%, or 80% is a magic number. So we could expand that range and say, if we want to isolate those with greater than 84% acceptable, I can relax those criteria a bit, particularly for the red chroma shift and a little bit in Rg, to create another set of criteria. I could relax those even further if I wanted to.
So, in particular, I think it's important to consider where are sources that we're using now going to fall? Color rendition condition one is probably at the lower end of the range for what you would see in a CRI 80 type of product. So that had about 65% acceptability in this case. That particular color rendition condition one in this experiment had red chroma shift of about negative 13%. If we look at a large library of sources and isolate those with CRI 80 or above, or 80 to 86, we see that red chroma shift is, on average, negative 9% with a range from negative 6% to negative 13%.
So, again, if I looked at the criteria B here that I'm specifying, there are a few commercially available CRI 80 products that would meet criteria. B. But most wouldn't. A vast majority wouldn't. But they are available today. You could write a specification that's even more relaxed than B – call it C if you want. That would allow more decreases in red saturation if you're looking – if that's acceptable in a particular environment.
So, our next step, experiment three, is going to look at further exploration of chromaticity effects, trying to refine these criteria even further, focusing on where that interactive effect occurs. Experiment four, we're going to look at an investigation of illuminance effects. So it's my working hypothesis that these preferences for increased red chroma versus a reference, when we're looking at interior illuminance levels, are actually occurring because we have a visual phenomenon known as the Hunt effect, where our perceived saturation is decreased as illuminance decreases. So really, these forces that are increasing saturation versus the reference are actually making us think that the colors are being rendered more like they would in an exterior, daylight environment.
So I want to include a brief review of a couple other research studies, done completely independently of any work here at PNNL. This is some research from Tony Esposito, who is completing his dissertation at Penn State University. This really mirrors the first experiment done at PNNL, looking also at 3500 K conditions. But it was done in the hueing booth with a completely different set of objects.
What I think was pretty fascinating when I saw the results is that the models of different perceptions were almost identical to what we found at PNNL. So perception was really driven by average color fidelity and red chroma shift combined together. It was very predictive of preference and normalness with saturation completely tied to red chroma shift.
Another experiment – Tommy Wei – I think this one was also done at Penn State. Tommy Wei is now at Hong Kong Polytech. The key conclusion here I drew in on was that the graphic of gamut shape is an important adjunct to average measures of color fidelity and gamut. So again, we need all three aspects of color rendition to really be able to understand how a space will be able to be perceived.
Another experiment, Xu et al. This looked specifically at color fidelity metrics and what components there led to the most accurate predictions of color differences. You see there is the 1964 10-degree color mapping functions in CAMO2-UCS, which is what is used in Rf.
So, we have all this research. Now, what is the status of TM-30 in the industry? I get some questions about this.
Because I think a lot of constituents are sitting around waiting for what's going to happen next. Manufacturers are waiting for specifiers to be using these metrics and understand them. Specifiers are waiting for manufacturers to provide the information so that they can use it. Efficiency programs are kind of sitting back and going, when is anyone going to use this? Should I include this on my own? And so there's kind of this waiting game going on.
So I did a little searching to see what manufacturers were providing the data. And I looked at LED Lighting Facts and some web searches. And there's a pretty healthy list of manufacturers that are providing this information.
Also, at the IES Conference last month, I was in a room giving a presentation. And Ron Steen was moderating the panel. And he asked, are you using TM-30? And there was a healthy number of hands in the room from manufacturers that were providing this information. Also, working with LED Lighting Facts – that's part of the DOE SSL program – we're working to make that information more accessible by trying to encourage the submission of spectral power distributions, in which case, TM-30 and any other color metrics can automatically be calculated for any of the products that are submitted.
I also have had some conversations with some of these companies that are actually designing products specifically to take advantage of TM-30 and some of this research that's been ongoing. Another question is, are there meters and software available to do these calculations? The IES had a calculator that comes with TM-30. But it was really intended as a scientific reference instead of a commercially usable type of software package.
But there are now TM-30 calculations integrated into commercially available handheld meters and laboratory-grade spectrometer software. Another important point that's always useful to convey is that all you need is an SPD to calculate TM-30. It doesn't require any new measurements than you would have had if you had already calculated CRI.
So, what's the status in terms of standards organizations? CIE TC1-90 issued report 224 earlier this year, where they essentially adopted TM-30 Rf with a few minor tweaks that I'll go over in the next slide. However, they still supported simultaneous use of Ra, the CRI that has been in use for about 50 years. Personally, I think that's a little bit challenging. But it was the stance they chose to take.
There's another CIE committee, TC1-91, that focuses on issues other than color fidelity. The last report I have, there was no recent progress there. Although the report that eventually might come out of that group could include the other parts of TM-30 among other proposed color rendering metrics in their report.
So one of the things that's going on with the CIE is they have these two separate committees that are really focused on different aspects of color rendition, whereas TM-30 covers all of that comprehensively out of one system. And so I think it was a little bit hard for the CIE to immediately react and be able to accept or not accept TM-30. Now there is another TC expected to convene this year, from what I understand, to continue to investigate color preference.
And what is the IES working on? The IES Color Committee is considering changes right now – we actually have a meeting next week to discuss this – in harmonizing TM-30 to reflect the small changes that were made by CIE, so that we can have one unified Rf metric and not competing ones. The group is also considering working on recommended practices and specification sheet guidance to have some of this other complementary information to go along with the TM, which is specifically a description of a calculation method.
So, again, I said I'd explain the differences between IES Rf and CIE Rf. And this goes a little bit off the deep end at the very end of the presentation here. But I'll explain it to anyone who is interested.
Change A was a change in the extrapolation method. Essentially, we see no difference in the Rf values due to that. There was also a change in the blending region for the reference illuminant from 4,500 to, say, 5,500 to 4,000 to 5,000. Again, that has no significant impact on scores randomly distributed maybe one to two points.
The third change was a change in the scaling factor – to use a colloquial term, I'll call this grade inflation – where the scores of all products are going up to some degree. And we can see that in chart C. And if we combine all those changes together for a compositive effect, we see in chart D. Now, if I compare both of these to Ra, you will see some of the difference.
So here is the original IES TM-30-15 Rf versus CIE Ra. So one of the things the CIE was concerned about is that a lot of sources with CRI values above 80 saw their Rf value lower than their CRI value. By changing the scaling factor, we can see that those sources are now more in line with their CRI value, although sources with much lower fidelity values now are more likely to have a higher Rf value than Ra value.
The practical impact of this is insignificant to me. The only real changes you could consider are that specification criteria, the Rf values would need to be a little bit higher when we make this change to the CIE's Rf value system. It's about two points at a CRI of 80.
OK. Another thing that I've learned in the past two years is there was enough of an argument that Rf and Ra weren't different that it should really be addressed. And so this chart shows that there's about a 40-plus spread in Rf at an Ra value of 80. Now, that is a pretty significant value. And I think it's important to understand why that occurs.
So this is some recent analysis that I've submitted for publication in LEUKOS exploring how that is related to specific types of distortions. So the key chart here is the one on the upper right, showing that the difference between Rf and Ra is strongly tied to what's going on with red chroma shift. So when red chroma is increased, as experiments have shown is preferred, Ra scores tend to go down much more, and they're much lower, than Rf scores for the same product.
And if you're familiar with R9 and its odd scale, you can understand how this might occur, because of the color spaces and non-uniformity used for calculating Ra. And that ties back into R9, that the scale differences are present in Ra as well. Other ways to look at these differences, if we calculate Rf using the eight samples from CRI, we can see in chart A there, it shows that you can't score better on the 99 samples than you would – let me make sure I'm saying this right. You could score better on the eight samples than you would on the 99.
So you can specifically tailor your source to those eight samples. But you can't do that for the 99. So you can't pass the test with 99 but not pass the eight as well. And you can show chart B on the right there, again, the differences due to the different color spaces that are used to calculate Ra and Rf.
So, back away from the technical cliff for the conclusions. Should the lighting industry make the effort? Who's going to step up to the plate and start providing this information, and using it, and have it really become part of our everyday language in lighting? I think some key things to think about when making that decision is CRI is really an anti-preference or naturalness measure. It has this negative correlation with our preference and naturalness perceptions because of this treatment of red saturation.
Still, even if it didn't have that, using average fidelity alone is, in my opinion, mostly useless, as it's unrelated to any perceptual attribute on its own. Now, it's an important counter-part when we specify red chroma shift. Because we can't just specify red chroma shift alone. Because you could have very large distortions in other places as long as you have a good red chroma shift if you didn't also include Rf in that specification.
I will say, at this point, there is probably more research available to support specifications based on TM-30 than there ever was to support CRI 80 or 90 values. In my own reviews, I've never found an experiment that directly found the result that CRI 80 led to a reasonably acceptable and preferable product. It's my understanding at this point – and if anyone has any research otherwise, I'd like to see it – that those values essentially came out of technology capabilities with fluorescent lamps moving to triphosphor, implemented more or less in the Energy Star program by DOE at the time, as a way to encourage both better color quality and those more efficient triphosphor fluorescent lamps.
Another point, it's not more complicated. So I can specify TM-30 with just two or perhaps three numbers. And it can be quite effective. Today, you might see CRI and R9 still using two numbers.
I think it's important, this idea of color preference versus color fidelity. It's much easier to have light that is preferred and perceived as more normal and natural and have it be efficient too than if you're only considering fidelity alone. To get higher and higher fidelity values, you have to have more continuous spectral power distributions, said to be less efficient.
There's this question of whether we're going to reach international agreement and make everyone happy. I don't see a reason at this point to not use a system that is more scientifically accurate, that is no harder to use, and that can lead to better results, while we wait for international agreement that may never come. CIE has been working on this for 30-plus years and is still working and forming a new committee. I think what the IES has done can be accepted as it is today and used quite effectively. I also think, from a manufacturer's perspective of trying to deal with different regional requirements, there's no additional measurement requirements here. And it's not as if other requirements, be it electrical safety or any other criteria, does not vary internationally as well.
So, ultimately, I'd say, the science is there. Color quality is a choice for specifiers, manufacturers, EE programs, consumers, and everyone in the lighting industry. And it's going to be up to those different entities to really take the step, learn about the system, and try and start implementing it as best as possible at this point.
And I think we're starting to see that momentum building, and we saw the large number of manufacturers providing the data. It's integration into the tools of lighting software and the meters. And I think more and more as we go along, I'm hoping, my goal is that we see products released that are taking advantage of this perceptual research and providing something different than what is commercially available today, which is pretty much looking the same, still going down one path. Because that's where our past metric CRI had us.
So with that, I will be finished and open up to any questions. I will answer as many as I can live online right now. And if I don't get to your question, I will try to respond via email sometime this week. So, thanks.
Thank you, Michael. And we have received several questions during the presentation. I'm going to start with some of the questions related to the actual color experiments. So, first question, were participants evaluating the color rendering of the space looking at the room from the outside of the room or from entering the room itself?
They always entered the room itself. I think that's really important to make the experimental experience as closely related to your actual experiment in the space as possible. So that relates to the chromatic adaptation and your actually being physically in the space.
And also, they never saw the changes. So they came out of the room to see the change. Because we don't experience those type of changes when we're going through architectural spaces.
And did they make their judgments immediately, or did they wait?
They were instructed to wait at least 30 seconds before making their judgment and they were allowed to take as long necessary to make their judgment. Again, this was to address as much as possible in a short-term experiment their adaptation to different conditions. Now, of course, what they previously saw is going to affect their judgment, even though they didn't see the actual change. So that's why we randomized this over all subjects, so that we can counterbalance that order effect.
And would you expect different results over longer periods of time, or do you have any comments about the time period of two minutes versus, let's say, 30 minutes?
Yeah. This question has been raised. I've had some good discussions with different researchers about whether our long-term perceptions are different than our short-term perceptions.
Do we adapt more to those changes over time or not? I think it's an interesting question and one I'd like to explore in additional research. I have no indication at this point that it would be any different. But it certainly could be.
I can give you some examples – for example, the neodymium incandescent lamp. That's the classic product example of a source that increased red-green saturation. That was developed experimentally, developed into a commercially available product. It was widely successful. It sold at a premium to standard incandescent lamps and it still sold millions of units and was around for 20 years, until LEDs more or less have made it obsolete from an energy perspective.
So I think the other important line of thinking related to that question is, how do we experience spaces and how do we make judgments? Not all of our experiences, especially in color-critical applications, are made over a long term. If we walk into a store or viewing an artwork exhibit in a museum, that's almost always a fairly short-term experience. And so, in not all cases do I think we should think that the long-term experience is the most important consideration.
OK. Thank you. Why were normalness and saturation rating correlated with hue bin 16 versus hue bin one?
It's not that they weren't correlated with one or the other. If you look at the actual numerical results, they were both very highly correlated. One would be, say, 0.94 if I built the model and the other would be 0.92. So it just happened to be that one was better in one perception than the other by that slight different amount.
So there was that big bar chart I showed you. It had lots of different information on it. And so you could see that bins 16, one, two, and even – I think 16, one and two would be the three that would be the most related to one another. And because they're highly correlated with one another themselves. So you can't really isolate and just increase saturation in hue bin one without also doing it in bins that are adjacent to it. With typical sources, the pattern of distortion is really continuous around the hue circle.
OK. And a question about the ages. Was that looked at as being a possible factor in the results of these experiments?
We did look at age and gender and demographic factors in these experiments. I can't remember specifically the exact results of that now. So I don't want to say it. I believe it is discussed in both of those journal articles.
I don't think it's always easy to develop products for a specific age group. I mean, there are certain instances where that could be done. But in most cases, you're going to have people of different ages in a space. You're also going to have people with different color-matching functions viewing a space.
I've been reviewing some different journal articles looking at that specific effect of interobservability on color rendition metrics. And like you would with perceptions of CCT or lumens, our visual systems all differ. So we really end up having to design to these averages as a way to prefer practicality in commerce and design.
OK. Thanks. And why was the light level chosen for colors for experiment two? And what was it, first? And then, why was that chosen?
Both one and two were around 210, 215 lux. More or less, that's the capability of the system and the maximum illuminance we could create with the luminaires we had. It's also pretty typical of an interior architectural environment, about 20 foot-candles. Perhaps a little bit on the lower side, but not too different than, say, 25 or 30 foot-candles.
OK. And now looking at other experiments, have there been other color preference experiments that include tasks? And if so, does preference vary by task type?
There's certainly been color perception studies related to some tasks. You know, sorting Munsell chips. I'm not sure of any that specifically asked about preference while performing different tasks. It could be something interesting to look at. But I don't know of anything in particular.
There's been lots of different environments that have been studied. But it's more or less people just experiencing the environment and not necessarily doing something in the environment. Of course, because when you have a task, you're asking someone to focus narrowly on something, and if that's a typical task we would do in a space, like reading, they're not going to be looking at a colored object.
Now, if it's color matching or distinguishing between colors, that's a different task than would be related to color preference. So if we were trying to minimize metameric mismatches or increase the ability for someone to discriminate between colors, we wouldn't be trying to optimize their preference in that space. So we'd be looking at different correlations there.
OK. Thank you. And are you seeing the TM-30 data being provided at the LED-component level? Or is this something that's mainly being provided by fixture manufacturers?
I mean, ultimately, the spectral data and the spectral power distribution is, I'll say, the responsibility of the component provider. They're the one who engineered that spectrum in their product. And so I think the most simple solution would be for those component suppliers to provide that information that could then be used by the luminaire manufacturer. And I think, as far as I understand, that's typically what's occurred with other photometric information.
Yes, the integration of that component into a luminaire might change that spectral power distribution a little bit. And if the luminaire manufacturer knew that was going to happen, photometering the product again and providing new information would be important. But I think, really, all along the supply chain, and anywhere we've traditionally required a photometric test, it's the same point where you would get the TM-30 information. Again, the test requirements are no different than anything we've had in the past.
OK. Thank-you. And will you create a spreadsheet that calculates preference based on spectral power distribution?
Sure. That's pretty simple to do. So from both of the models that are presented in the journal articles, they're just based on the RF and the red chroma shift values. So if you know those two values, you can create a third column in the spreadsheet and just type in the formula that's given in those journal articles to get the preference value.
Now, of course, what we tried to do in the second study is relate those values to some acceptability level. You could always try and maximize preference. But usually, in lighting, you're going to have some kind of trade-off between color rendering and energy efficiency or color rendering and another aspect. So you might not always be able to maximize that color preference.
But again, those calculations should be pretty simple. If you're interested in a spreadsheet, I can provide that. I don't think that's something that would come in the standardized version of TM-30 when it's revised at this point, although that's something the committee could consider.
OK. And could you address a little bit more about the state of CIE and IES in regards to Rg? So will there be any updates related to Rg?
So, I'm not involved with the CIE Committee. So I can only speak from second-hand information. But again, Committee 191 has focused on photometrics other than color fidelity.
Now, they're compiling a list of different proposals, as far as I know. And there will be a report issued that describes all of those different proposals without necessarily recommending one or another. If the CIE would wish to change Rg in some way, that would come back to the IES and we would consider whether we felt those changes were appropriate. I hope we could all be unified, as I expect we will be with Rf. But that's just speculation on my part.
OK. And we're almost out of time. So are there any other comments that you have before we finish today's webinar?
I just thank you all for tuning in. If there's another question you have or something you want to see, you can reach out to me. We're going to keep trying to provide the information that's hopefully helpful to you to help get these – really, my goal in this is to see a greater variety of products so that specifiers have a choice of what to use to create the best-lit environment and meet their design goals. I hope manufacturers can start talking to those specifiers, looking at the research, and identifying new engineering targets, and specifiers will then have these choices available to them to improve our luminous environment. So, thanks, everyone for listening.
Thank-you, Michael. And thank-you again for participating in today's webinar, brought to you by the US Department of Energy Solid-State Lighting Program. You may all now disconnect.