Why RGB LEDs suck for a color RA4 enlarger

Not many people are crazy enough to build a light source for a color enlarger. From time to time, someone picks up the challenge and posts about it online. Most of the time, the concept revolves around some form of RGB LEDs, whether that be led strips, Chip-On-Board (COB) LEDs or even addressable LED ‘beads’ such as the popular WS2812. All of these are in my opinion doomed to fail miserably to produce quality RA4 prints. Let me explain why.

Before we begin, a small disclaimer: I’m not an engineer of any kind, nor do I work for any company involved in the photographic industry – in short, what I know, I learned by reading, experimenting and making it up as I went along. On the other hand, I did do quite extensive experiments with an RGB COB led design for a color enlarger and on that basis, I think I have a fairly good understanding of its fundamental shortcomings, and also why these translate 1:1 to RGB led strips and addressable RGB led beads such as the WS2812(b). (After those experiments, I built two generations of discrete-LED RGB heads that I used extensively; the second generation is my ‘daily driver’ enlarger, with which I have made hundreds of very nice color prints already.)

First, let’s have a look at what color RA4 paper responds to. In other words, its spectral response curve. Here it is, for Crystal Archive Supreme, the color paper I use 99% of the time:

Fuji Crystal Archive Supreme spectral response, taken from its datasheet.

In case you’re wondering, other papers’ datasheets are eerily similar, even Kodak’s. (Although Kodak doesn’t really produce any color RA4 paper anymore, and Sino Promise who took over the business maybe never will anymore either, but let’s put that matter aside for a bit.)

Immediately apparent are the three sensitivity peaks for blue, green and red light. These translate into the yellow, magenta and cyan dyes that will be formed in the emulsion upon processing after exposure. Furthermore, notice the distance between the red and the green peaks, and the much smaller distance between green and blue, with considerable overlap between these two. These factoids will prove relevant later on.

So how do you get that red, green and blue light onto the paper? If you’re using a color enlarger, you’re most likely using a subtractive method employing dichroic filters. (Unless you happen to be using a Philips PCS 2000, which is the only commercially successful enlarger for the general public that I know of to use an additive topology.) Be sure to click that subtractive filtering link, because it contains a very nice illustration that explains it better than any description could. So one way is to use a wide-spectrum light source, typically a halogen bulb, and then selectively block out large swathes of the color spectrum using the yellow and magenta filters (cyan isn’t used for RA4 printing).

Now, most of the RA4 paper that’s consumed today comes never close to any dichroic exposure systems. Nearly all of those big (and smaller!) rolls of color paper go through digital photofinishing equipment that actually uses an additive method. These use red, green and blue lasers, and thereby directly hit the sensitivity peaks of the paper. Nice – no waste (i.e. lower power draw and dissipation, which is a good thing in a high-volume setting), and the opportunity of a perfect match between the paper’s sensitivity and the light source used.

When building a LED color head, we adopt the second approach that is the more common one today, and use an RGB additive scheme as opposed to a CMY subtractive one. This means that we need to select the Red, Green and Blue components to match what the paper needs. And this is exactly where things get tricky.

Take your average RGB led product, whether it’s an RGB led strip, an RGB COB led module or an addressable RGB led bead, and observe what kind of colors they actually produce. Here’s a typical specification – I took this from the WS2812b datasheet, but you’ll find virtually identical numbers on (nearly?) all RGB led products:

WS2812b addressable led bead peak wavelengths

Ok, that looks nice, right? Well, let’s plot this onto the sensitivity chart of our RA4 paper:

WS2812b peak emission wavelengths plotted onto CA Supreme sensitivity

Two things stand out:

  • All emission peaks of the LEDs miss the sensitivity peaks of the paper. The distance is particularly big between the red sensitivity peak (at around 690nm) and the LED’s emission peak (at 625nm).
  • The blue led exposes the green emulsion as well.

Then, there’s another complicating factor: the LEDs peaks are really that – peaks of a bell-curve shaped plot, and to make matters even worse, there may be secondary peaks as well. Here’s an example of a random red LED spectrum show the problems of many/most red LEDs for our application:

Typical red LED spectrum. Note the extension into the green area, below 600nm. [Source: Apogee Instruments]

There’s a small amount of energy emitted below 600nm, and depending on how much that is, this may be (and in practice, *will* be) sufficient to excite the green-sensitive layer in RA4 paper. We can solve this problem, at least sufficiently so it ceased to be a problem, by selecting red LEDs with a much longer wavelength. Although these too suffer from a wider emission spectrum than we would ideally have, the problem doesn’t stand out as much. Partly because we hit the peak sensitivity of the paper much more efficiently, so we can get the same density without throwing more than necessary lower-wavelength secondary emissions at it, and partly because the LED spectrum tends to be spread around its peak, and hence, less of the secondary emissions will actually end up being in the green-sensitive area of the paper.

In practice, it turns out that 660nm LEDs are a good compromise since they are (at least, since a few years) widely available in higher power ratings. Since recently, even longer wavelengths at higher power levels are available, so maybe today it’s even better to try and find LEDs with an emission wavelength close to 690nm. I did not test these as they weren’t yet available when I built the LED RGB head I currently use.

On the blue side of the spectrum, the issue is a bit more complicated. Here, it turns out we can’t just shift the wavelength further away from the center, because that gets us close to the UV spectrum and my experiments with low-wavelength LEDs (particularly 425nm ones) suggested that the green- and red-sensitive emulsions also tend to be excited by these smaller wavelengths. In other words: problems actually increase if we choose this route. Experimentally I found that royal blue LEDs with a peak emission closer to 450nm work better than regular blue ones with a peak around 460-465nm, and also better than shorter-wavelength LEDs.

Theoretically, ‘ice blue’ LEDs that peak around 480nm could be an interesting solution, as these would exploit the difference in sensitivity between the green- and blue-sensitive emulsions, which seems to be the greatest around the blue-sensitive peak at this wavelength. I have not (yet) tested these, also in part because they weren’t widely available at the time of doing my validation experiments for RGB LEDs.

How about green? Well, it turns out that if you search for green LEDs, they pretty much all peak out at 525nm. Theoretically, a 550nm LED would be nice, but I did not find any when I was looking for them. Perhaps this has changed, because over the past few years, the choices in LED colors has positively exploded due to enduring technological innovation in the field of semiconductors.

So based on a bit of theory and a bit of experimentation, I concluded that the selection of a feasible set of purchasable and affordable RGB power LEDs would be something like this:

  • Red: 660nm
  • Green: 525nm
  • Blue: 450nm

Besides the argumentation I already offered, there’s another line of reasoning behind the selection I presented above and that I actually use in practice on a daily basis. It’s briefly summarized in this post on Photrio by the late Ron Mowrey where he mentioned they used Kodak Wratten #98, #99 and #70 filters back in the day at Kodak. Reasoning that was good enough for Kodak probably was good enough for me, so I included the filtering cutoffs of these Wratten filters in my selection. Here are those cutoffs with the 660nm, 525nm and 450nm LED lines overlaid on top of them:

My RGB LED selection peaks overlaid on Kodak Wratten #98, #99 and #70 cutoffs

As you see, it’s a fairly good match on the blue, and I noted earlier that a shorter wavelength for the blue induced more problems than it seemed to solve. For green, the match is suboptimal, but limited by product availability at the time. The red selection is also a fairly good match and indeed the primary reason why I hunted down the still-scarce 660nm power LEDs when I sourced the LEDs for my color head. Note that if you were to plot a 620nm line into that chart, it would end up smack in the middle of the Wratten #99 and #70 cutoffs. In other words, 620nm LEDs expose RA4 paper in a spectral zone where Kodak didn’t when they did additive exposures, and I take that as a bad sign for this wavelength.

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

I’m hesitant on the blue ones, not in the least because of their bad match with the Kodak Wratten #98 filter and also because I’ve got a feeling a 480nm blue LED might not be spectrally pure enough to get the job done. If the blue ones err too much on the UV-side or the green side, severe problems will pop up. It’s a matter of experimentation to see if this approach might work. For the 690nm red ones, a lot can be said; they hit the paper’s peak and fall smack in the middle of the red zone passed by the Kodak Wratten filters. The only drawback will probably be limited luminous efficacy. Green should be an improvement at 550nm if a suitable product is available, as green LEDs appear to be a challenging bunch; see below for more on this.

Back to the problem we started out with: especially on the red side of the spectrum, all RGB LED products I’ve seen over the years consistently use red LEDs of 620nm. And those just don’t work for a color enlarger. This is the major reason why I titled this post in such strong wording. So far, I can only conclude that integrated RGB LED products just aren’t suitable for RA4 printing because they all suffer from the same, fundamental and fatal flow: an inappropriate red spectrum.

You might ask, what’s the problem? Sadly, I didn’t save my test prints, but the problems were twofold (and of course interrelated):

  • Poor spectral purity of primary and secondary colors. Especially yellow suffered badly, with an RGB COB led being simply not capable of printing a bright yellow. It always shifted to orange/brown. Red also turned out to be poorly saturated, with it shifting towards a dried-blood shade of reddish brown.
  • Severe crossover, as witnessed in printing greyscales and color patches. You get one end of a greyscale right, and the other end shifts either to blue or to yellow. You try to get one primary color to print right, and the other ones fall into the abyss.

On a side note, I have seen examples of RGB LED enlarger light sources, both from hobbyist and commercial sources, that looked very nice at first glance. But on closer inspection, there were always two problems: only nice looking subject-matter prints were shown, often with a limited gamut, such as green foliage. Or examples were embedded into YouTube videos with probable color problems actually being visible if you know what to look for – but they don’t stand out to the casual observer.

Finally, there’s one more part to the story, and that’s the efficiency problem. Observe that the red sensitivity of RA4 paper is considerably lower than green and blue sensitivity. This means that in an absolute sense, we need to throw a lot more red light onto the paper to get a neutral color balance. If you add the efficiency of LEDs to this, further problems arise. These are the typical luminous efficiencies of red, green and blue 1-Watt power LEDs you might encounter on eBay et al.:

  • Red 620-625nm: 50-60 lumens
  • Green 520-525nm: 80-90 lumens
  • Blue 460-470nm: 35-45 lumens

You might think that this doesn’t look so bad; the blue ones appear quite inefficient, but that’s OK because the paper is very efficient on the blue end. And red doesn’t lag all that much behind, does it? Well, based on practical experimentation, it turns out that you want an electrical power distribution of around 4:2:1 for red:green:blue. So that would for instance be 40W of red, 20W of green and 10W of blue. There’s considerable leeway to settle around these numbers, but this is the general ballpark. If you use RGB LEDs, you’re always stuck with roughly a 1:1:1 power ratio, which means your blue and green LEDs don’t see much use, and the red ones are thundering away at full power all of the time.

If you’re also going to use your color RGB head to make B&W prints, then the power issue becomes an actual problem. Whereas RA4 paper is quite sensitive (it translates to something like 25ISO in film terms – yes, quite fast!), B&W paper is much slower, especially warm-tone paper, which is even around two stops slower than neutral-tone bromide paper. This was really the Achilles’ heel of the first successful version of my RGB color enlarger. I designed it around a 3:2:2 power ratio: about 30W of red and around 20W each of blue and green. The problem was that with slower (warm-tone) papers and enlargements from 35mm film, the light source lacked considerable ‘oomph’ in the green channel to give me comfortably short exposures. Blue wasn’t the problem – all photographic materials are so sensitive to blue that this is nearly never a concern.

By the way, there’s a rather silly reason why green is more problematic than it often seems. Take another look at those luminous efficacy numbers above. Green is the most efficient one of the bunch, giving the most lumens for your Watt! That’s right, but what are lumens, really? There’s a problem with lumens in this context, and that is the lumen scale being adjusted for the sensitivity of the human eye. And that sensitivity turns out to be a massive peak at green, and a steep drop-off towards UV on one side and IR on the other.

In other words: the luminous flux numbers of single-color LEDs are misleading, because they skew the results in favor of green. In reality, there is something called the ‘green gap’ that describes the lagging of luminous efficacy of green LEDs compared to red and blue ones. Add to this the fact that in a multigrade/variable contrast paper, green sensitivity is miles behind blue sensitivity by definition (Ilford tells us the same!), green sensitivity is far below blue sensitivity, it’s easy to see why you’d quickly run out of steam on the green channel when doing B&W work on a LED color enlarger. This is one more reason why I believe combined RGB LEDs are an unfortunate choice for building an enlarger light source.

So summarizing, there are several reasons why combined RGB LEDs are not a suitable choice for building an RGB color enlarger:

  • The spectral emission of the red LEDs precludes decent color reproduction.
  • The power balance between red, green and blue does not match the real-world requirements of color enlarging; the red is lacking.
  • The power balance is also extremely unfavorable for B&W reproduction when using variable contrast papers; the green is lacking.
  • Apart from the red and green being problematic, the blue seems to be off as well.

So in short, integrated RGB LEDs have three main problems that stand in the way of successful use: the red, the green, and the blue. It’s sad, but that’s what it is!

PS: I mentioned addressable WS2812b LEDs earlier. While they are nice enough technology for interior lighting and perhaps even artistic effects in photography, they are also only 8-bit devices. They can generate 255 light levels on each of the R, G and B channels, and for accurate filter adjustments when exposing RA4 paper, this turns out to be much too coarse. I tried it, in practice, and it just doesn’t work. There’s no way you can get an 8 bit LED head sufficiently linearized and keep sufficient filtering resolution to accommodate the real-world variance in color negatives. My estimate is that at the very least, 11 bit resolution (2047 steps) is necessary, but 12 bits (4095 steps) in practice is the lower limit of avoiding filtration problems in color work. But perhaps more about this at a later date, when I get to write about the color light source I’m currently using.

PPS: Still here? If so, you might also like the series of posts I wrote later about the journey towards a functional color enlarger light source and controller. It’s a four-part series:

13 thoughts on “Why RGB LEDs suck for a color RA4 enlarger”

  1. Wow! good work!
    thanks for sharing this with any of us who dare to bring some contemporary electronics to the analog RA-4 world of color.
    I will try the 4:2:1 ratio with my smartphone prints on RA-4 paper. I hope there’s magic in the numbers!

  2. Well, you have gone an admirably long way in this LED area already. I deal with this idea in my head already several years (and on my table undone experimental heatsink populated with LED beads at least one year). My initial idea was also to just put there those all-in-one RGB LEDs, but I rejected it soon after some rethinking and deeper theoretical self-study. I am glad to see that your practical results match with my theoretical output and that my direction is about right.
    Besides sensitivity peaks match we must also watch out for the absorption curves of the negative film. The lights, the absorption peaks of dyes in the negative and the sensitivity peaks of the paper should match. But in real world none of them match.
    The yellow dye absorption peak (ca. 455 nm for Kodak Gold) will not match with blue-sensitive peak (480 nm Fuji CA DP II) in paper. Other absorption peaks vs. sensitivity peaks match. I am not sure why this discrepancy (maybe it has something to do with action effect of the mask or maybe lately they changed the spectral sensitivity for better compatibility with laser beams in digital photo phinishers, or maybe part of some anti-darkroom printing conspiracy).
    So for blue light is “royal blue” the best choice. Green is a big trouble. 550nm would be ideal, but 550nm power LED is not available. I found some 560nm 3W beads on ebay, and I bought several dozens of them. But they have quite broad spectrum, reaching to the red area. But they contain almost no blue. They must be driven at maximum current, because at lower currents they shift to amber/red even more. Even if they contain orange and bit of red light, there is still a chance they will not cause any significant “cross-talk” into red-sensitive layer, because of quite a big gap between green- and red-sensitivity. Moreover the red sensitivity grows quite slowly towards its peak.
    I also tried to apply 550nm dichroic filters on 525 nm LED, which work fine, but only for axial (90°) beams. Beams with off-axis angle will be filtered at different wavelength. So this is no go, unless you ensure to use only axial beams. Or does exist some non-dichroic narrow band pass filter out there? Another solution would be lower current in 525nm LEDs. The lower the current, the longer wavelength they glow. But to achieve 550 nm they would glow so weak that you would need hundred of them. Or to use hundred of “classic green” (greenish-yellowish) SMD LEDs or plastic THT LEDs.
    And 660nm LED as red light is definite choice, there should be no troubles.

    1. Ah, yes, the LED issue. Well, concerning the absorption peaks, what I think is essential is that you hit the peaks of the paper without crossover to other curves. So you need a blue that only exposes the yellow-forming layer without hitting the magenta-forming layer, etc. That the peaks of the film and the paper don’t overlap exactly is apparently not a problem, since the combination works fine when using regular dichroic filters or RGB lasers in a production environment. At first, I worried about this a lot, but in the end given the fact that things apparently work OK with RGB lasers, I stopped worrying about it. It’s interesting stuff, but easy to break your brain over it. Btw, I never found any indications of a conspiracy here, haha!
      You’re right about green being a tricky color when it comes to LEDs. Here as well I figured that if solid-state green lasers suffice, then green LEDs will too, since they rely on the same semiconductor bandgaps. As I write somewhere, it would be interesting to test with a theoretically more optimal green color, but we’ll have to wait a bit for those to become available. I have no doubt that they will in a few years. Big strides have been made in recent years particularly in this area!
      > Another solution would be lower current in 525nm LEDs. The lower the current, the longer wavelength they glow.
      My experience is the opposite. Take a regular THT green LED, the old-fashioned 5mm kind, and run it at 50mA or so (so far beyond its typically rated current of 20mA). You’ll see it’ll glow yellow or even orange, instead of green.

      1. Film mask helps a lot in this issue, that’s for sure, especially with problematic magenta dye, which partly blocks also lots of blue and red light. But your choice of 450nm LED is wise and (besides paper cross-talk) helps to keep some stray absorption of magenta layer on minimum. Otherwise our worries would be much worse…
        Lasers are different story, because given by their nature they are almost ideally monochromatic (they shine in one wavelength), so with them it’s easier not to expose adjacent layer, even if their wavelength is not straight on the sensitivity peak. Moreover with them the image is directly drawn on the paper, they don’t project some dye image, so there is no “transfer loss” of colour purity.
        To prove my assumption that modern RA-4 papers have been adapted (but did not broke film compatibility) to laser exposure (conspiracy theory was a joke of course), let’s mention that Noritsu laser unit uses 690nm laser for red, 532 nm for green, 473nm for blue. Blue peak was for sure shifted towards longer wavelength, because old print materials had around 450 nm peak for blue, to match absorption spectra of yellow dye. Green was not shifted and remained the same, maybe because it would get too close to blue peak and it would corrupt film compatibility. 550nm green laser is recently manufactured, but expensive, so 532nm laser must suffice and it does. 690nm laser is perfect match of course and it matches film too.
        LEDs have broader spectrum than lasers, so they can cause paper-crosstalk much easily.
        UV, blue and so-called “real green” (520-530nm) LEDs shorten their peak wavelengths with higher currents. They are InGaN based. Old-style green-yellow ones have different chemistry, so maybe that’s why they glow towards red if you fry them.

        1. Well, this is really interesting, Ivan! I never really looked into semiconductor laser properties and always assumed that their wavelength distribution would in the end be similar to that of LEDs (but with much better collimation, evidently) since a semicon laser and an LED of the same wavelength would exploit the same bandgap. But your comment makes me think a little more on this! What I also really appreciate is your mention of wavelengths apparently used by Noritsu – I never knew this! Do you have a source or additional information about this? I’m surprised at the relatively long wavelength they use for blue, but apparently you’re right and it’s not a problem because of the narrow peak. And I find it fascinating that they use a very deep red 690nm one, which aligns very well with my observation that a 620nm red LED doesn’t work well for RA4, but 660nm does.
          Thanks also for setting me straight on the wavelength shift issue.

          1. Lasers are fascinating thing really. I don’t know exact working of those semiconductor laser, I have only high school knowledge about how laser light should behave.
            (Semiconductor lasers were novelty back then and in physics lesson they demostrated us a true laser gun – wonderful NeHe gas laser from 1970s – at least 1.5 meter long glass tube in wooden holder with many wires along and huge power supply box with minuscule output power. A truly fascinating technology!)
            But I think the diode in semicon laser works there as a photon source and there must be again some optical resonance chamber like in any other laser, which amplifies the photons of only one frequency and phase…
            Because one guy complained me that RA-4 darkroom prints don’t come out as good as they used to, I made research of modern RA-4 papers, and also put down wavelengths of Noritsu digital lab lasers in my notes, but I don’t remember exact source. I wanted to see, if there’s some relation between “digitisation” of RA-4 photo labs and supposed change in RA-4 paper properties.
            I think the reason for change of peak sensitivity of blue-sensitive layer was limited availability of other laser wavelengths back when digital laser photo labs started. The first blue diodes were also quite dim, so it was crucial to match the sensitivity peak with them. Nowadays the situation is different and 450 nm lasers are not a problem of course.

          2. But I cannot tell if change in blue peak makes noticeable difference in darkroom print from film negative. He complained about grass green rendering. Green is composed by cyan and yellow. Yellow is detected by blue-sensitive layer, where blue light is modulated by yellow layer in the film. But there is stray absorption of magenta layer too, which absorps part of blue light too. I looked in my notes and found that stray absorption of magenta layer is about 20% higher at 480nm than at 450nm. So there must be some loss of yellow purity in analog prints, it they are printed in broad spectrum light.
            If you print with 620nm red LED, your red sensitive layer on paper gets about 60% cyan information and 40% information from magenta layer! Green sensitive layer exposure would be minimal, but your cyan image in the print would be unusable. In 660nm light you get 80% cyan and 20% magenta, which is much better.
            Here is some graph, where I put spectral curves of RA-4 papers and spectral absorption curves of a negative. Sorry for czech description, I made it already before. Dashed lines are negative layers, full lines are RA-4 paper. You can see easily what you get onto paper with lights of chosen wavelengths.

          3. Sadly I didn’t keep the test prints (or at least I didn’t file them systematically) when I was testing with the 620nm red and 460nm blue LEDs. What I do recall is that particularly the yellows were very ‘dirty’ and unsaturated, reds were also quite muted. Green is always a bit of a tricky color I suppose, not in the least because it’s just hard to get a good green when you only have yellow and cyan to work with. So even under optimal conditions, saturated greens will be a weak spot of RA4 prints – as will violets, and reds.

  3. Through the 90s I had a custom printing business, using the 45a, which has flash tubes and RGB filter’s. Really nice setup, but I found digital was the way to go, around 2002. Sold my $15,000 setup for $150 then. Found your page with a search. Very interesting! Lots of work. I’m a colorist in the film industry these days.

    1. Cool thanks for reaching out! Yes, digital is great, and objectively speaking, it’s just the most convenient, effective and reliable way to get the job done.
      Your job sounds very interesting indeed! As you’ve gathered, color is a fascination of mine – but at a mere amateur level.

Leave a Reply

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