Perfecting primaries – Another look at LED wavelengths for RA4 paper

Way back I explored the best wavelengths for exposing RA4 paper. In that blog entry, I pitched an idea of optimal wavelengths that I hadn’t tried yet. Well, I picked up the glove myself and I think the results could be pretty relevant. Read on, especially if you’re looking to purchase or build a LED exposure system for color enlargements.

Back when I wrote that article on LEDs and how integrated RGB LEDs suck for RA4 printing (I still believe/know they do, btw), what I didn’t have was a proper validation of my argument. I relied on a theoretical exercise involving paper spectral sensitivity curves, Kodak/Wratten filter absorption curves and some very informal/subjective testing.

In the meantime, I picked up a photospectrometer, which puts me in the position to do some more thorough testing. On a rainy Sunday, I figured I could apply that device to my color printing LED enlarger experiments, and you know, one thing led to another, and several rainy Sundays followed…but let’s start with that first one.

What I did was so simple that I wonder why I didn’t do it before. Here it is:

It’s literally nothing more than the same Stouffer T2115 step wedge enlarged onto FUJIFILM DPII RA4 paper. These are three separate prints (scanned in one go), and the only thing I changed was the exposure – the time, and the color of the light. The top strip was exposed with only the blue channel, the middle with the green channel and the bottom with the red channel. The result is three neat step wedges showing the primary colors of this paper.

Or do they? I was actually working on this pretty soon after my blog on how the color-forming layers in RA4 paper are in the opposite order than in color film: i.e. cyan on top in paper instead of yellow as in film. Based on that blog, I figured I could demonstrate the principle of color filtration. The demonstration I had in mind relied in part on the principle that if you expose the paper to very strong blue light, this will end up exposing all three color layers, since all three emulsions are inherently sensitive to blue light.

Well, sort of. The principle seems to be true for at least the yellow and magenta layers. Look at the yellow strip at the top of the illustration above. It’s not quite yellow, is it? Note how the darkest patches on that strip aren’t really a dark yellow (what would ‘dark yellow’ look like, anyway!?), but rather a dark rusty red-brown.

Let me skip ahead a few weeks when I followed up on that testing and I created some better strips. The image I showed earlier was made by actually enlarging a T2115 onto paper, but this brings all manner of problems with flare/halation at the strong light intensities used to print through the dense parts of the step wedge. In a recent test, I simply contact printed a step wedge (a T3110 this time), which creates an overall much cleaner result. Here’s the yellow one, exposed with the blue (450nm) light of my recent enlarger light source (I wrote 440nm on the strip, but it’s really 450nm):

Note that this one also trends from yellow into a pretty convincing blood red. Since I had that photospectrometer anyway, I figured I could take some measurements off of those strips. And with measurements, one can make plots, and plots sometimes help to understand things. From the yellow strip measurements, I took the Lab* values, ignored the L* for a bit and plotted b* against a* to get a feeling for how the hue of the patches changes:

The plot starts close to the origin (the ‘white’ steps are in fact slightly fogged due to overexposure) with a* and b* values close to zero. The steps then take off upwards, becoming more and more yellow. They also curve to the right a little but, until around step 11 or so, at which point there’s no more increase in yellow. Instead, the plot drifts off to the right, and downward. What does that mean?

Well, have a look at this CIELAB hue circle: yellow is on top, and if you go to the right and down, you go to…that’s right, magenta! So it pans out – if you overexpose RA4 paper with only blue light, you end up not just activating the yellow-forming layer, but also the magenta-forming layer. There might be a little cyan creeping into the mix, as well, but it takes a metric sh*t ton of exposure to get there, and it’ll mostly show up as desaturation and a dark tone anyway, since the yellow and magenta are there as well (and yellow + magenta + cyan is just black).

Funny, that. To complete the exercise, I also did step wedges for magenta and cyan in the same way, measured the Lab* values and plotted them in a similar way. Here’s what that looks like:

So here’s the yellow plot going off into the direction of magenta when dramatic overexposure with blue light is given, and magenta does a similar thing, but the opposite: it leans towards yellow given enough exposure. In fact, look at how similar the high densities on the yellow and magenta strips look like:

It’s really the same kind of red that they trend to, just approached from a different direction.

Cyan drifts off into the direction of magenta at some point, but not by much, even if you expose the heck out of this paper with red light. The cyan just turns very slightly purple, but this is actually not really visible with the naked eye and it only shows up clearly on the plot:

Forgive my trusty old Epson 4990 for not doing such a great job at rendering the cyan in particular -it’s far more vivid in real life. To be frank, I can’t be bothered to properly fix the color profile for this exercise – what’s important is that the cyan remains pretty much cyan, even on dramatic overexposure. It doesn’t really go very violet, as you’d expect it to do if there would be a lot of magenta mixed into it.

The behavior of the cyan layer makes good sense; it’s exposed with red light, and the paper has no sensitivity to red light in its magenta and yellow forming layers:

The plot above is the spectral sensitivity for Fuji Crystal Archive paper – it’s inconsequential, because the actual silver halide emulsions used for the Fuji papers are all the same, so the plot above is true for DPII as well.

But there was this one thing I wrote back in that initial blog about using LEDs for RA4 exposure:

At the time of writing, if you’re in an experimental mood, I would dare you to create a light source based on the following peaks, because that would theoretically be the best solution:

  • Red: 690nm
  • Green: 550nm
  • Blue: 480nm

The strips I’ve shown so far were all made with the light source I constructed a few weeks ago and have been using since. The wavelengths used are the same I’ve been using since my first successful LED head, which is 450nm for blue, 525nm for green and 650nm for red.

I just couldn’t resist the temptation to order some more LEDs.

So I went online and I found that by now, 550nm green LEDs are actually available at reasonable cost and power levels. This wasn’t the case when I started out with these LEDs and you could essentially only get 525nm green (at least, I never found anything else back then). Also, the selection of red LEDs has expanded dramatically, probably because these are now quite popular in ‘horticultural applications’ (cough *weed* cough). And yes, blue can now also be had in pretty much any color you like, as long as it’s some kind of blue!

So I got me some 480nm blue, 550nm green and 680nm red LEDs. Pretty much smack on the peaks of the sensitivity of these Fuji papers (well, quite nearly so):

550nm green is really exactly on the peak sensitivity of the paper. 680nm red misses the peak by about 15nm, but seems like a very decent match, too. And blue at 480nm is really close to the peak at 490nm as well. Compared to the previous set of 450nm blue, 525nm green and 650nm red, it looks like this:

I put together another light source, which is pin compatible with the one I had already been using, but now with these new LED wavelengths:

Like the previous design, it’s an aluminium-core PCB populated with 3535 form factor SMD LEDs and an RMS total power (all channels on) of around 100W. It’s of course mounted to a suitable cooler (not shown above).

Now, the big question is – how does this new choice of LEDs affect the color rendition of the primary color step wedges? During a single session, I created similar step wedges with the previous and the new LED exposure units, so with the ‘old’ and the ‘new’ wavelengths. This was possible because I can fairly easily swap out one light source for another; it just involves lifting out one unit and putting in the other, and re-plugging three cables.

From the step wedges, I made the same Lab* plot I’ve shown earlier. This is it for the 480/550/680nm LED head:

To tell you the truth, I was surprised by the result. In case you’re not seeing it, yet, let me give you a visual clue:

See it, now? That’s right – the magenta is actually magenta, and stays that way, even upon massive overexposure! This I had not expected, even though it makes perfect sense. Look at the spectral sensitivity plots above and notice how 525nm (which I used for green up to now) is really a near-miss with the blue-sensitive/yellow-forming layer. In fact, it turns out to be not a miss at all and 526nm LED light does in fact (quite evidently) trigger yellow dye formation in the paper.

But with 550nm green, the magenta stays pure and no yellow (or cyan) dye is formed. Note how in the Lab* plot, the magenta line starts at the origin of the plot, moves outward to magenta, and then simply stays at the same hue even if additional exposure is given. Neat!

At least as interesting is that cyan/red and yellow/blue seem to behave exactly the same way as before. There seems to be no significant difference between 450nm and 480nm blue, or 650nm and 680nm red, at least in terms of hue purity. There are distinct differences in speed; while the new set of 480/550/680 looks dimmer to the naked eye, the overall printing speed at a similar color balance is about a stop faster. I suspect that the 550nm and 680n, LEDs are rather inefficient, but that this is offset by the increased sensitivity of the paper at these wavelengths, which makes it all balance out in the end.

Something else stands out in that new Lab* plot, as well. If you look at the yellow curve, you can see how it starts as a pure yellow, then goes off into the right and downward as magenta is added to it, and then it goes back towards the center again. This is evidence that indeed, cyan is being formed and that you can in fact expose all layers of color RA4 paper with blue light – it’s just that there are a couple of stops of difference in speed between the layers. This trend towards cyan wasn’t visible in the earlier plot because I either didn’t expose it quite long enough for this to happen, or the cyan/red layer is a little less sensitive to 450nm light, which makes the effect kick in a little later.

What’s the practical implication of all this? Well, I tried testing that by making prints from the same negatives with both LED heads. However, the color balance between those prints is a little different, which is simply because it’s really tricky to color balance two prints to an identical result with narrowband light sources with such distinct differences.

I have a feeling that the practical differences aren’t in fact that big. I would expect that if you were to print a very saturated and dark magenta, you would see a difference with the 525nm LEDs making the magenta become more red, while with 550nm LEDs it would remain pure magenta. The question is to what extent it’s likely you encounter such a scenario. And, in a similar vein, how likely it is that you’ll spot the differences as a very subtle color shift in less outspoken colors. In any case, the 550nm LEDs demonstrably give less (or in fact, no) color crossover on the magenta layer and under certain conditions, this would somehow show up in prints – especially when e.g. burning & dodging.

Lab* curves of primary color step wedges for the 480/550/680 combination overlayed on top of the old 450/525/650 wavelengths

I did some printing with the new head last night and all I can say is that after adjusting the color filter algorithm in the exposure controller, printing is as straightforward as with the previous generations of the LED head. The prints come out fine, too; the colors look natural and clean. And I get about an extra stop in speed for the same electrical power, which is always nice.

One more thing – in the Lab* diagrams, you may have noticed the diamond-shaped points in yellow, cyan and magenta. These are the ab* values I obtained from a Fuji engineer for this DPII paper. It shows I’m right on the money for the magenta and yellow hues. I could hit the cyan benchmark as well if I overexpose the red layer a little more, but note that Fuji’s benchmark value is itself already shifted a little bit towards magenta. At this point, you’ve already passed the purest cyan hue, which is a little more to the left. So I’m not too worried about that one.

The one thing I have not yet tested, is printing black and white. And I suspect that this is where the combination of 480nm blue and 550nm green might actually be a poor performer. I think so for two reasons:

1: The 550nm green LEDs seem fairly inefficient. When printing on variable contrast paper, it’s the green channel that needs to deliver the most power. I therefore suspect that for slow, warmtone papers (I’ve just received some more Fomatone MG to play with!), I’ll be wanting for more oomph on the green channel.

2: 480nm blue is mostly ‘blue’ because it’s nominally that. In reality, 480nm ‘blue’ is more of a cyan – it’s a very greenish kind of blue. And I suspect that in a variable contrast B&W paper, this wavelength will activate not just the fast blue layer, but also the slower green layer a bit. This will result in the maximum usable contrast grade not being as high as with a shorter wavelength blue.

I’ve yet to verify this. But that’ll have to wait for another rainy Sunday. For the moment, I think I’ve seen enough step wedges.

8 thoughts on “Perfecting primaries – Another look at LED wavelengths for RA4 paper”

  1. Hello, that’s some cool research again! I am glad to see how magenta improves with real 550nm LED. Unfortunately my LED head, is still not finished yet. I rejected the 525nm leds already in the start and when I was beginning with this head 2 yrs. ago, I could only find 560nm LEDs with sufficient power. They are rather green-yellow. But because the sensitivity of cyan/red layer starts quite far and lazily, I think that the hint of the red in these LEDs will not make some big trouble.
    But I can try some real 550nm LEDs, if they are already available? Where did you order them from?
    As to the choice of blue LED, which give similar results with different wavelengths, I think the reason for this problem is, that they always have some side spectra. If you choose 480nm LED, you are very close to the peak of blue sensitivity, but its side band already reaches quite much into the steep part of curve of the magenta layer.
    If you choose for instance 460nm, you are farther from green peak, which is nice, but on the other hand you are in the less sensitive part of the blue curve and the magenta layer is still very sensitive here, too.
    If the sensitivity peak of the yellow layer would be of standard 450 nm how it was in “for analog” paper, we would be fine. But these “for digital” papers have the peak closer to the sensitivity peak of magenta layer.
    The only solution for it to use narrow band blue light source. Then you can get close to the blue peak without exposing magenta layer too much.

    1. The 550nm LEDs I got from AliExpress; there aren’t many sellers that offer them, at least the powerful ones, but you should be able to find them. I searched for “3535 smd led 550”. The ones I use are nominal 3W types, but I run them at less than half that at the moment.

      As to the blue LEDs – there’s just inherent sensitivity of the other layers to blue light. The spectral plots on the Fuji datasheet are incomplete; for some reason they just truncate the plot of each color channel, but in reality, there is a response in areas that are literally uncharted. This is especially true for the blue and green channels. Keep in mind that silver halide as such is inherently sensitive to UV light and this extends into blue, even for silver chloride (color paper is mostly silver chloride with a little silver bromide). Moreover, it seems that the dye sensitization of the green channel gives a fairly broad peak that extends especially downward. My conclusion so far is that it’s simply impossible to expose only the blue/yellow layer. You always expose the green layer a little, too – and probably the red one as well. Since red is so slow, it would take massive exposure to build cyan density with blue light, but the green emulsion is much closer in speed to the blue one, so yellow tends to transition into magenta/red anyway. I’m quite positive this is even true for narrow-wavelength light sources such as laser. It’s just inevitable. And it’s probably perfectly OK too for regular printing, so more something out of academic interest to look into.

      Let me know how your LED head progresses!

      1. Yes, the green sensitive layer is always exposed. All we can do about it is to find a spot where the green sensitive layer is exposed the least compared to blue sensitive layer exposure. This place lyes of course in the sensitivity peak of blue/yellow layer. But the broader is the light source spectrum, the more it will touch adjacent steep part of green sensitivity and the benefit of being on the sensitivity peak will be cancelled.
        In such situation only the laser on the peak would give the best results.

        But interestingly according to your graph, the yellow looks very clean till some point, and after this point abruptly decays. However it is not clear, where this breaking point lies in your test print. To my eye, in your test print the yellow starts to get orange already somewhere in the middle of the scale. (But maybe in this point the yellow is already fully dense and wouldn’t get denser anyway, am I right?)
        On the other hand magenta and cyan get contaminated according to your graph very quickly, but your test prints show that their maintain their purity even at severe overexposure. How come?

        I have almost no practical experience with color rendering in “for digital” paper in analogue printing, but something tells me that yellow purity is quite crucial – it takes part in rendering important colours like green, orange, red. But green is special there. If red or orange gets some magenta more, it is not big deal, but eye is very sensitive to green and if yellow is contaminated by magenta just a little bit, you cannot properly render bright fresh greens of spring nature etc. Or am I wrong?

        1. About the magenta crossover due to unwanted exposure of the green layer with blue light: note that the left slope of the spectral sensitivity of the blue and green sensitive layers is always parallel, with the blue layer riding a couple of stops above the green layer. This suggests it doesn’t matter that much which wavelength you pick, and not even if it’s a very narrow-bandwidth light source, because the difference between green and blue sensitivity doesn’t vary very much. The thought behind using 480nm is/was that the difference between the green and blue sensitivity in absolute terms is the highest at this point. But evidently, this doesn’t make much of a difference in the actual print – or in fact, it makes absolutely no difference at all.

          As to the test strips vs. the graph: the graph is made from measurements performed on the strips you see, so it’s the same data presented in a different way. The point where the yellow curves peaks and then bends to the right and becomes more magenta is at step #16 for the 440nm light source and on step #18 for the 480nm light source. This is just left of the series of the steps in the chart that look very much the same.

          > magenta and cyan get contaminated according to your graph very quickly
          Not the way I read the graph; in fact, they stay quite pure and radiate out from the center in a straight line. It’s the yellow curve that bends right from the start. I think that bending of the yellow curve at least at the low densities is not due to crossover, but an inherent property of the dye itself. I’ve done quite a few measurements on pigments and you often see arcs like these, where the hue actually varies a bit in accordance with density.

          You’re right that contamination of certain hues will have more visual impact than others. The reasons are complex, and involve such factors as “is there a visual reference somewhere that makes a color problem stand out”, “what kind of subject matter is pictured” (a Caucasian face going slightly green we will notice, but a flower of the same hue doing the same thing we probably won’t experience as unnatural at all), etc. However, regardless of which hue is contaminated, as soon as there is cross-contamination, the total gamut you can image on the paper will reduce.

          As to digital – I think digital is ultimately more forgiving since so much can be compensated for through ICC profiles and LUTs.

          1. I knew that you took measurements with spectrometer probe from your prints, I only didn’t know which steps in your prints match which dots in your graphs. If the yellow deviates from step 18, then it should be OK under normal circumstances, because according to your graph yellow don’t get any denser from this point.

            Maybe I don’t just know how to read your graphs, but I see yellow is deviating no more than 20 points in its maximum, but magenta and yellow are deviating almost 40 points in their maximum density. So magenta and cyan should be less pure than yellow. Which is quite possible, because magenta and cyan dyes show quite heavy stray absorptions themselves, while yellow dye itself is almost ideally pure.

            Spectral width of the light source matters – see this image Upper graph shows exposure with ideal monochromatic light in the peak of blue sensitive layer. Exposure ratio blue:green is the best possible. Bottom graph shows light source with broader spectrum, with similar distribution like LED. Its peak matches the sensitivity peak, but side bands hit more sensitive parts of the green sensitive layer, causing its further exposure. The exposure ratio will be less ideal, as green sensitive layer will be exposed more than in case of monochromatic light.
            This is why laser digital printers take advantage here. And you are right, with the help of correction LUTs, the purity of hues can be further improved, if there’s a headroom for it of course.

          2. Yes, I realize that I didn’t communicate clearly the relationships between the density plots and the Lab plots. For most purposes this isn’t so relevant, but your question makes sense. And indeed, the hues remain pure up to their maximum density. What happens beyond the exposure needed to achieve that density is where the funny business starts. The conclusion one could draw from this is that everything is just fine as long as the negative printed fits within the exposure latitude of the paper. If not, and/or if areas are burned that also affect shadow regions, pushing the paper beyond the dmax of any of its color layers (especially yellow), those areas will start to cross over significantly. I may come back to this at a later moment, but I think this shows up subtly in normal prints.

            > but magenta and yellow are deviating almost 40 points in their maximum density
            Deviating from what? The a* vs b* plots of a step wedge of an ideal dye or pigment would start at the center and radiate out to the extremes of the plot in a straight line. A curve to the line denotes a hue shift. Any straight light means the hue remains the same and only the saturation (chroma) changes. Magenta behaves like a virtually perfect dye. The cyan layer is a little more wayward with its twist at the end of the curve; this actually reminds me of Pb15:3 pigment which does pretty much the same thing in my experience.
            If by deviation you mean how far cyan and magenta end on the a* and b* scales in an absolute sense, while yellow hits close to b=100, then yes. This is also something you see most of the time with cyan and magenta colorants, with magenta often hitting higher maximum chroma than cyan. Mind you, chroma and density are different things, although related.
            Does this help?

            As to spectral width: yes, I’m familiar with the argument you’re giving and that was exactly the reason why I tested 480nm. And the net result is…it doesn’t matter. So theory doesn’t hold up in this case. The yellow layer responds exactly as well (or poorly) as it does to 480nm as to 450nm blue LED. You could argue that the 450nm LED is more spectrally pure than the 480nm one, but I really doubt that. Again, it would be cool if I could test the width of the spectral peak. Something with a prism and a digital camera might work.

          3. Erm, I read your graph wrong, sorry, I am new to this CIELAB. Now I understand. Yes, yellow seem to deviate already in the middle of its density range, question how is it perceptible. C and M seem to be OK, only magenta itself consistently leans towards yellow, which is the property of the dye, which itself is not perfectly magenta. But yes it is linear, which means that it doesn’t get contaminated with yellow from yellow layer throughout it density range.

            You have spectrometer, so you can measure the LED spectrum and apply it to the Fuji’s graph to see how much will be the magenta layer exposed with your blue LED vs. ideal monochromatic light. You can then estimate integral exposures of both yellow and magenta layer and calculate their ratio. With monochromatic light the exposure ratio would be ca. Y:M = 4:1.
            But I took a little effort and plotted some typical blue LED spectrum into the Fuji Graph and it doesn’t seem to be such a big disaster like I thought. (My previous plot was way too pessimistic as spectrum width of blue LED light concerns):
            The increasing part of green curve starts to be little touched, but not as much as I thought, so the ratio of integrated exposures of yellow and magenta layers is gonna be better then I thought, but IMAO never as optimal as with monochromatic light.
            If you have some so-popular blue laser pointer/gun with some suitable wavelength on your hands, you can try to expose the paper with it, of course with decolimated light beam, or you will make a hole in your paper : D

          4. Yes indeed, the magenta is in fact somewhat red. In the end, it shouldn’t matter too much; as long as te chroma is nice and high, the gamut will still be large. If the magenta had been a little less yellow, the paper would have expanded gamut in the violets – but at the cost of oranges. I think this was probably a deliberate compromise, especially since Caucasian skin tones are in the orange corner.

            My photospectrometer doesn’t allow direct measurement of incident light. I could try and hack it in some way (technically, it’s evidently capable of doing this since it will also profile a monitor), but it would still be not very optimal as it only allows data to be collected in 10nm-wide bins. That’s pretty coarse.

            I’ve been playing around with some LED spectra as well and arrive at the same conclusion as you; it’s not all that bad and in fact, the match with the paper’s sensitivity is quite good in most cases. I’m writing a followup blog in which I may touch on this as well.

Leave a Reply

Your email address will not be published. Required fields are marked *