RGB LEDs for color enlargers

[brbo]

Yes, I briefly mentioned the 2000 in my blog post. I never inspected one, but AFAIK they used dichroic filters and a single (?) light source with a mirror system to create the R, G and B components. I've never used one, so I can't comment on how well they work. The company did have a reputation for decent quality products, at least back then.

PCS130 and PCS2000 use 3 separate 35W bulbs with R/G/B dichroic filters fixed directly in front of them. The controller unit is then used to dim individual halogen bulbs thus providing desired filtration. With PCS130 you can also completely turn off individual bulbs which you can't on PCS2000 where individual bulbs will go down to pretty dim but not completely off.

 

[DREW WILEY]

Bernard - the Phillips was a simple rheostatic device, and really suitable only up to 6x6 film and small prints. But at least is was simultaneous exposure instead of the more laborious sequential RGB method. Later Beseler made a xenon flashtube model with its own idiosyncrasies and lack of sufficient power for serious commercial use; but a few of these are still in operation. Then they marketed a much more muscular tricolor halogen model for their 4x5 chassis which suffered from too many electronics being crammed into too small a space relative to EMI - the Achilles heel of that particular design. But a number of people have either rebuilt these or figured them out well enough to keep the system going. Beseler itself basically walked away and orphaned any service needs, though for a price, there is a viable private option even now.

Durst, on the other hand, designed a seriously powerful head for use on their 10x10 commercial enlargers. It never reached the market; they outright shut down their commercial division right around then. But six colorheads did go into the use at an NSA classified facility where they worked with 9 inch aerial color film. The problem with the Durst concept is that it ran so hot that a special tech had to replace the filters every six months, with an overall maintenance expense of over $30,000 per year per enlarger, if you factor travel expenses and the very specialized nature of replacements. Additive filters of that type are a lot more dense than subtractive, and involved a lot more light and heat to use in large applications. I've seen all the key components or colorhead guts in person, though certainly not at the classified facility itself.

Won't go into my own 8x10 design here, except to state that it alleviated that terrible heat issue, and uses pairs of dichroic "trimmer filters" on each side of the respective nm transmission points rather than super-dense one-color filters.

 

[DREW WILEY]

Koraks - slightly different combinations of color separation filters during either shooting or when making separations from color chromes might be selected relative to either the idiosyncrasies of the exact film in use, or the nature of the specific dyes chosen to make the print, since much of this was argued in relation to dye transfer printing at one time. The 58 green, 25 red, and 47 blue set is less dense and can be exposed faster than the denser set of 61 green, 29 red, and 47B blue. But with the less dense set, you're obviously letting a little contamination from other parts of the spectrum through in each case. None of these are of real practical use for simultaneous printing colorheads, though they can and have been used under the lens for the more laborious method of sequential additive printing. The whole topic can get complicated fast unless one has sufficient background.

 

[koraks]

@brbo and @DREW WILEY thanks for the info on the PCS enlargers. Sounds like they were a little less advanced than I had hoped them to be. Rheostats for dimming is really old-tech and I'm not sure if it's ever going to be consistent enough for RA4 printing. Then again, apparently it sort of worked.

Won't go into my own 8x10 design here

That's OK; it's a pity because I really am curious about it, but no pressure.

Thanks on sharing your thoughts on the filter kits. I see what you mean in the difference between both sets and the overlap of the 58/25/47 set compared to 61/29/47B.
I need to collect these combinations somewhere because I'd like to run some analysis on them. I find it somewhat surprising that there's apparently quite a few configurations that are/were used for consecutive RA4 printing and it makes me curious as to how they compare spectrally. Sounds like a nice little puzzle for a Sunday morning.

they can and have been used under the lens for the more laborious method of sequential additive printing.

Theoretically they could also used in the same way as in the PCS enlargers, I suppose, as long as 3 separate light sources are used and integrated in a suitable mixing chamber. The sources would also need to be dimmed with sufficient resolution. It's an avenue that might be fun to explore next to the approach utilizing R, G and B LEDs. Both would have their strengths and weaknesses - both sound like they might have potential.

The whole topic can get complicated fast unless one has sufficient background.

Oh, it can, and in ways you generally don't anticipate. I know; I found out the hard way and learn every day.

 

[DREW WILEY]

Why I don't go into my own design is that, among other things, I had routine access to heavy shop machinery as well as massive discounts on certain supplies, and got a number of key elements for free, and as a key distributor of certain lines, was even outright given samples of a lot of precision fabrication gear. The fact I was doing really fussy work was a feather in their cap, so they actually got a lot in return in terms of my endorsements. But I still had to drop in 15K of my own to finish the project, and it had to be reasonably right the first time. The prototype was the final itself. But being strictly for personal use, rather than resale, I could freely cannibalize from other machines anything I wanted without fear or patent infringement. Any machine like this made for sale would end up costing the purchaser at least 75K, and probably a lot more. It took three years to finish.

This is also a really big heavy tall machine. The precision vacuum easel is so heavily made that I can actually stand atop it without deflecting the surface. All the business end above is precision micrometer-gear driven. That required some luck. The yaw correction was adapted from a WWII era machined bronze battleship artillery aiming device, which had to sufficiently hold an optical sight securely in place under repeated shock, and would probably cost fifteen grand itself to make today; but I got it free military surplus.

Then the electronics keeping the two hemispheres feedback coordinated involves some custom items no longer available; and the improved way of doing it, which was not available in any practical sense when I started the project, namely, sinewave multiple light control, is computer and software dependent, which means it goes obsolete every five years or so and periodically needs to be replaced at significant cost.

So in general, I think the future does lie in advances in LED technology instead. There will still be big commercial halogen enlargers from Durst and DeVere around for awhile, since they were made to last. Yeah, those were pretty darn expensive too, but periodically turn up very reasonably when some old lab retires and wants to liquidate its stuff. Most of that kind of opportunity occurred when many switched over to scanning and big laser printers, but there are still sometimes opportunities to acquire high end gear cheap or even free. It takes both luck and timing.

So yeah, some of you LED experimenters are in the position of the Wright Brothers, but without the funding or shop gear of an outfit like Heiland. A step at a time.
At my age, I mainly just want to print, and not try to re-invent the wheel again.

 

[Ethan Brossard]

This thread has inspired me to do a little more tinkering with my design. I was researching higher bitrate LED hardware for increasing my number of brightness values from 256, and came across the info that the DMX decoder I'm already using can do 16 bit dimming with a different mode, so I switched modes and did a little reprogramming to enable it. I actually am only using 15 bit dimming now, as the arduino I based the controller can deal with 16 bit integers, but only 15 bits in either the negative or positive direction, so while I could probably do some further reprogramming to access all 16 bits, from what you guys are saying it sounds like 15 bit will be fine.

Once I have it setup again, I might do some experimenting to figure out the proper settings to match Ilford's multigrade filters. The enlarger is currently disassembled since I'll be heading back to college soon, but theres a good. chance I'll be able to set it up in one of the the college darkrooms in a couple weeks.

For any of you interested, here's some images of the head as I was testing the 15 bit dimming:

(I'm not sure if you can see the file names, if you can the explenation is that a friend of mine told me the enlarger looked like Gort from The Day the Earth Stood Still, so that is now the enlarger's name.

I also have a couple videos about the whole enlarger system, they're not the best, and I'll likely write a new thread with new videos going more in depth about it once it's set up in a darkroom.

 

[koraks]

That kind of scene is a familiar sight
Looking good, have fun printing with that. It's nice and compact, too.

 

[Lachlan Young]

Durst, on the other hand, designed a seriously powerful head for use on their 10x10 commercial enlargers.

Probably only because their much more serious competition in the aerial/ mapping markets were doing 4.8kw dichroic heads (or more) - and without the failing filters (though even an ELH bulb will happily destroy a dichroic given half a chance). And Ilford (who ought to know a thing or two about filters) had effectively built the MG500 in the early 80s as an additive head using the correct B & G dichroic filters - and that variant has the same problem that #47/ #47B filtration or equivalent LEDs (which have even narrower bandpasses) have - i.e. they don't pass deep enough into the blue to get max contrast, while a magenta filter does (Kodak's data effectively agrees, though it suggests an MG400 (which is Y/M) is capable of delivering higher contrast than any other filtration system - though I've never been bored enough to subject mine to a step-wedge). Ilford found that the max contrast of the Heiland blue LED (whatever it is) was about the same as the blue filtered ELH bulb in the MG500 - and on the G5 setting was delivering about 2 stops more output than the dichroic (which is no slouch, believe you me). And there have been many attempts over the years to adapt the light source from cinema printers (which are RGB) onto enlargers without much in the way of commercial success - and that's before we contemplate that there was even an RGB cold cathode head that existed.

 

[DREW WILEY]

Durst was apparently planning to convert all their high-output 8X10 / 10X10 colorheads to additive, but then suddenly decided to get out of the commercial enlarger business entirely. I once had an early version high-wattage Durst color mural head, and just the damn cooling fan itself required more wattage and caused a bigger utility bill than the entire rest of my property. Leave a transparency in there for a full minute, and it would be faded to about 25% of what is what it was when it went into the carrier. It would punch a 30X40 inch Cibachome print plus a .90 density mask in about 15 sec. A dual ELH bulb Dust CLS301 colorhead is like a spare flashlight by comparison. Thank goodness, RA4 style printing requires far less intensity of light.

But you're incorrect about a 47B being less effective than a deep magenta. I own a full set of the very deepest magenta and violet Wratten filters, which are quite a bit deeper than a no. 5 Multigrade filter. But a 47 or 47B blue filter will yield even higher VC paper contrast with those papers capable of going that far. The result is identical from what I obtain using the blue channel only on one of my narrow-band additive enlargers.

But some of this question could very well be related to the color temp of the originating light itself. Even my cold light (a 12X12 high-output Aristo V54) is blue-green, so very efficiently truncated one way or another using hard blue versus hard green selective filtration below the lens. Many old conventional subtractive colorheads were inefficient, and passed a fair amount of "white" light too (actually rather yellowish tungsten); and of course, bulb-only heads started out all basically yellowish. So that fact might explain certain alleged statements by Kodak, perhaps based on their own now defunct ventures into VC paper. But it's telling that Ilford did switch their own actual Multi-contrast head from MY to BG internal filtration for sake of better performance.

I don't think anyone took those cold cathode heads very seriously. There were some very powerful pulsed Xenon RGB Sequential heads sometimes used by the printing industry. Some of those heads required water jacket cooling. My own additive heads are pulsed halogen, and RGB simultaneous. Somebody is sure going to inherit some odd toys they neither want nor need.

 

[albada]

Cree XM-L RGBW LEDs

Referring back to the OP's article at https://tinker.koraks.nl/photography/why-rgb-leds-suck-for-a-color-ra4-enlarger/, I just ran across the datasheet for a new line of Cree RGBW LEDs called "XM-L" (don't they know that XML is a mark-up language?) at this link: https://assets.cree-led.com/a/ds/x/XLamp-XMLDCL.pdf
The red and green LEDs have the typical values mentioned in the article, but blue is a surprise: It's actually royal blue, with a peak at a slightly shorter wavelength than typical blue. The spectrums (spectra?) are on page 7 of that datasheet. I suspect these 4-in-1 LEDs would work well for B&W because of that royal blue. For color, the red wavelength is too short. The package's current-limit is 1750 mA, so that's almost 600 mA per LED. If you buy it, be sure to get the "high density" version and not the "high intensity" version which has a green bump in its blue spectrum. The package is 5x5 mm, so there's some hope of being able to hand-solder the thing.

 

[koraks]

The package's current-limit is 1750 mA, so that's almost 600 mA per LED.

A little less. There's also a white emitter. As you said, these will likely work quite well for B&W, albeit that it's likely a relatively expensive solution that offers less bang for your buck than discrete blue and green LEDs in terms of power density and economics. For RA4, there is indeed likely an issue with red, but it might be nice to experiment if someone is inclined to do this.

 

[albada]

Cree's new XE-G LEDs include a violet which should work well for B&W, and a "photo red" (660 nm) suitable for RA4. They can handle up to 3 amps! Cutter Electronics in Australia is selling XE-G LEDs mounted on copper stars suitable for DIY folks. Here's the violet version.

 

[koraks]

a violet which should work well for B&W

I don't think it's really necessary to go below 450nm for B&W. I'm not sure if it hurts, though. It *will* hurt color reproduction for RA4, so personally I wouldn't go there as there's no clear benefit but a distinct drawback to this.

In any case, those XE-G LEDs do look quite nice given their high power density. Mind you, thermal design will be a challenge. The PCB-mounted products you linked to from Cutter would make things a little easier, although this will also seriously limited power density of the final illuminator design, so it's kind of a win-loose situation. Personally, I'd try to get the bare LEDs and solder directly to an aluminium substrate PCB and then ensure proper cooling for the entire assembly. The latter it is noteworthy, BTW, given the difference in thermal flux curves of the R, G and B LEDs. Practically, you want these to run at a pretty stable temperature during an exposure. It's not super critical as you could adjust filtration for longer exposure times, but it will make the whole thing somewhat less intuitive to use.

A clear disadvantage of this route of course is unit cost; these LEDs don't come cheap and they're an order of magnitude more expensive than the 3W components I used. Evidently, you can use fewer of them for the same net power, but it would still be a somewhat pricey affair. Let's say you aim at something like 20 LEDs per color, that's $ 300 worth of LEDs, not counting other components such as drivers etc.

Drivers for let's say 2500mA (with a little safety margin) are also IMO a bit less convenient to work with than drivers in the sub-1A range due to the size of the inductors, dissipation issues in sense resistors as well as in the MOSFETs themselves as well as the MOSFET drivers. This is assuming a typical buck mode LED driver topology. All this will add bulk, cost and increase already present EMI concerns and thermal. For instance, the driver boards for my 350W-range LED head don't require any active cooling and can be built on normal FR4 PCB material; for higher power density drivers, this may not be valid anymore.

Up to this point, it was quite desirable to try and push the power level of individual LED units as high as you can, because overall higher power density for such a light source was simply a necessity. But I think at around the 3W mark for individual LED beads, there's sort of a peak beyond which other issues become more of a hassle; see some mentioned above. Of course, there is an attractiveness to trying to build a quasi-point source out of these ca. 10W units; this would be their primary benefit, I think.

 

[albada]

I don't think it's really necessary to go below 450nm for B&W. I'm not sure if it hurts, though. It *will* hurt color reproduction for RA4, so personally I wouldn't go there as there's no clear benefit but a distinct drawback to this.

I would try this as an experiment to see whether contrast increases at all.

Let's say you aim at something like 20 LEDs per color, that's $ 300 worth of LEDs, not counting other components such as drivers etc.

20 LEDs/color? Wow! I'm using only 5 LEDs/color, having copied Mal Paso's design. And these are 3W LEDs that I'm running at 2W (0.7A) because I dislike running electronics at their rated max. That's 10 watts total per color (at 100% PWM). So LED-cost, electronics-cost, density, and cooling were not problematic for me. I just checked my records for the latest 8x10 print I made: exposure was 24 seconds at f/8 (green was at 100% PWM). That was 35mm; 6x6 would be quicker. A 16x20 (from 35mm) at f/5.6 would need 48 seconds. Thus, I see no reason to have much more than 10 watts/color up there.

20 LEDs/color at 3W each is 60 watts/color. Are you passing the light through a filter? Or do you print very large sizes? Or does RA4 need much more light than B&W?

EDIT: I just thought of a good reason for you to need much more power. My Beseler 45 MCX enlarger goes up to 4x5. If yours is 8x10, then you must illuminate a diffuser having 4x my area, thus needing 40 watts/color to get my intensity out of your diffuser. So 60 watts/color would be reasonable.

But I think at around the 3W mark for individual LED beads, there's sort of a peak beyond which other issues become more of a hassle; see some mentioned above.

Agreed. If an XE-G is run at its max of 9W, then about 4.5W of heat must dissipate through its 1.8 mm^2 thermal pad. I'm not a mechanical engineer, but even I can see that much heat-transfer while keeping temperature-difference acceptable will be a challenge. My plan was to run these XE-G LEDs at 0.7A with my existing drivers, thus dissipating only 1W.

 

[koraks]

Are you passing the light through a filter? Or do you print very large sizes? Or does RA4 need much more light than B&W?

No, it's just a source that covers 4x5 and also needs to cover 35mm and everything in between. It has a diffusor at a decent distance of the light source and the original Durst condensors, albeit in a non-standard configuration. RA4 is faster than b&w btw, but I also like to do warmtone b&w from time to time. Wordt case scenario is 35mm enlarged to larger than 8x10 on warmtone paper; that takes a decent amount of light.
My current light source is over 350W total BTW, with around 175W for just red and the remainder divided into green and blue. It's a little faster than strictly necessary for my purposes, but I'd rather have a little too much than not enough power.
I did some prints yesterday at about 9x11" from 4x5 negatives; dense negatives took around 6 seconds at f/32. Color 35mm to 5x7" is around 2 seconds at f/22 through a 75mm lens. That's more than fast enough evidently, but the day inevitably comes where I need to enlarge 35mm to 20x24" and then I'll still have somewhat decent exposure times.

A lot depends on mixing chamber design, diffusor materials and condensor setup if present. That's not the strongest point of the Frankenstein contraption I've made. It does cover evenly across everything I do with it, which was the main issue to get right.

For 8x10 of course this would need to be scaled up.

My previous head was a little less than 100W for R, G and B combined and that was lacking in power especially when printing b&w warmtone at lower grades. I need this ebecause I sometimes like to print alt process negatives which are kind of steep to begin with.

10W per color sounds on the low end of anything to me even for 35mm to be honest. Maybe if you don't go beyond 8x10" and abstain from warmtone papers. Btw, I prefer to print small most of the time; nothing wrong with that!

 

[bernard_L]

@ all DIY-ers and tinkerers. I see on the datasheets of power LEDs that the light output has a temperature dependence, and not the same for each color. How would that impact consistency of exposure? Problematic when the thermal time constant of the LED-heatsink system is comparable with the exposure time. Worse when time-additive test strips are made, and the result converted into a single continuous exposure. Or maybe that is absorbed in one or more whole-image test prints?

 

[koraks]

@bernard_L, it's a valid concern. I briefly went into it in #63 above.

Problematic when the thermal time constant of the LED-heatsink system is comparable with the exposure time.

Indeed. So the design goal for a cooling system is not so much to keep within the maximum ratings of the device, but to keep heating during exposure to a minimum. You'll notice that as the light source heats up, the difference in downrating between the channels becomes greater, so keeping the temperature rise limited, also limits this effect. It's one of the reasons why I like short exposures, especially for color. The head doesn't heat up much in 4 seconds or so.

Worse when time-additive test strips are made, and the result converted into a single continuous exposure.

Sort of; there's two things at play:
1: The light source may heat up during one partial exposure, stay warm, then heat up further during the second exposure, etc. So there would be a compound effect.
2: The heating up during each exposure creates a non-linear effect in increased exposures; in other words, exposing twice as long will not give twice the density. It 'feels' a bit like reciprocity failure, but of course the mechanism is totally different.

Finally, there is of course the question of how important absolute accuracy is. Is it essential that for instance longer exposures have a nicely linear relationship with higher density, or is it acceptable that there is some non-linearity here as long as print-to-print consistency does not suffer? Hence, there can be some room for compromise depending on how strict you are in your specifications. I myself am not that strict as long as I can prevent gross non-linearities that make the system behave in a non-intuitive way, or (worse) that would create visible differences between for instance a test strip and a final print.

In a previous version of my system, I actually had an issue like that with the red channel. It turned out that for some reason, there was an effect of using red (along with blue and green) for focusing on the color balance (especially cyan) in the final print. My workaround for this was to disable red when focusing for color prints; i.e. I focused with just blue & green. While it worked, it was not optimal. The current version of my print head does not need this workaround; although the LEDs are without a doubt also prone to these kinds of effects, I gave more attention to thermal management (with some degree of success, although there's much room for improvement) as well as to much better current limiting/LED driver circuitry. I suspect it was especially the latter that was the problem in the previous version, as it had *very* crude current limiting circuitry.

All of which goes to show that your concern is very much justified, and it's even more complex and broad in scope than you might have realized, as it also includes ancillary electronics and not just the LED source itself.

 

[albada]

@ all DIY-ers and tinkerers. I see on the datasheets of power LEDs that the light output has a temperature dependence, and not the same for each color. How would that impact consistency of exposure? Problematic when the thermal time constant of the LED-heatsink system is comparable with the exposure time. Worse when time-additive test strips are made, and the result converted into a single continuous exposure. Or maybe that is absorbed in one or more whole-image test prints?

Assuming the cooling hardware yields a 50 degree C temperature-rise at the junction, the drop of green and blue are about 10%, which would have little effect. Red is the exception: In Cree LEDs, it drops like a rock with rising temperature. I have measured that drop on my enlarger using the easel-meter sold by Darkroom Automation (the link) (Nicholas Lindan). After a few seconds, I noticed that red dropped by perhaps 0.03 stops, which is insignificant. I didn't notice any drop on green or blue.

Each of my LEDs is mounted on a disk-shaped aluminum "star" sold by ledsupply.com, which in turn is attached to a sheet-aluminum heatsink with thermal paste. I'm running them at 0.7 amp. As seen in prior postings, I'm only running at about 10 watts/color max, so my heatsink warms only a little. In fact, I can only feel the warming with my finger if I've been focusing/composing a long time. So I believe the LED's junction isn't heating much, and thus output declines insignificantly.

 

[koraks]

0.03 stops, which is insignificant.

I sure hope so. Color is very finicky.

feel the warming with my finger if I've been focusing/composing a long time. So I believe the LED's junction isn't heating much

You can't feel the junction, and the surroundings you can feel don't give much insight into actual junction temperature. Look for instance at computer CPU's which have temperature sensors embedded into their dies - at least fairly close to the heat generating junctions. Observe the temperature difference between those sensors and let's say a sensor mounted on the motherboard in the vicinity of the heat sink, or even on the cou heat sink. You can easily observe a 25 to 30 degree Celsius delta between those zones. LEDs will be no different except that heatsinking is usually much worse than in computers!

 

[albada]

RA4 was designed for tungsten systems. And I notice that, at the wavelengths used for RA4 papers, tungsten light has roughly twice as much green-power than blue, and twice as much red than green. Thus, to avoid wasting LED-based hardware, we should build the red-green-blue circuits to output that 4-2-1 ratio at their max.
Koraks said that my 10 watts/color seems low, so I'm thinking of doubling green-power. It's always green that hits my max power when printing. Blue always has plenty of margin, and I think that 4-2-1 ratio is why. Koraks put much more red than other colors in his latest head. Perhaps we should do that with green/blue as well.

Warming:
> the surroundings you can feel don't give much insight into actual junction temperature
Quite true. In addition, after a few minutes, warming will give us an idea of how well heat is being removed.
But I trust my meter more than my finger. I think I'll conduct some more time-brightness measurements with it...

 

[albada]

I measured red again: After 35 seconds, brightness dropped by 0.01 stops. Insignificant.
However, the meter takes 2 seconds to adjust to a large brightness-change, so if red dropped some at the very beginning, I would have missed it.
I could connect a sensor up to my oscilloscope, but that's more work.

 

[koraks]

It's always green that hits my max power when printing. Blue always has plenty of margin

That's exactly my obsevation as well.
Ultimately for reasons of evenness I went for equal amounts of green and blue, but this was based on a dramatic increase of green power in my latest light source version. So I have a pretty gross excess of blue that's pretty much only there so that the diffusor can be somewhat more sloppy.

Btw your .01 stop red shift obsevation sure is quite comforting, I would say. I had expected far worse.

 

[albada]

@ all DIY-ers and tinkerers. I see on the datasheets of power LEDs that the light output has a temperature dependence, and not the same for each color. How would that impact consistency of exposure?

I was still wondering what happens soon after turn-on, so I built a circuit around a phototransistor and captured results on the oscilloscope. The result: Red had a flat line with no decline over the first second. Green and blue also had flat lines. Actually, the lines were noisy due to 100 kHz ripple from the LED driver, and I was disappointed at how large that ripple is. These drivers are Mean Well LDD-700L. Their datasheet says a capacitor up to 2.2 uF can be placed across the output. Has anybody tried that?

 

[Ethan Brossard]

I was still wondering what happens soon after turn-on, so I built a circuit around a phototransistor and captured results on the oscilloscope. The result: Red had a flat line with no decline over the first second. Green and blue also had flat lines. Actually, the lines were noisy due to 100 kHz ripple from the LED driver, and I was disappointed at how large that ripple is. These drivers are Mean Well LDD-700L. Their datasheet says a capacitor up to 2.2 uF can be placed across the output. Has anybody tried that?

If you’re using pwm dimming, a capacitor would cause trouble on low levels at a minimum I think

[edit: also, I’d also guess that variables like minor shifts in ambient temperature would cause more variances in exposure than a 100 kHz ripple, so finding a fix for it would be a lot of trouble for not much benefit]

 

[albada]

If you’re using pwm dimming, a capacitor would cause trouble on low levels at a minimum I think
[edit: also, I’d also guess that variables like minor shifts in ambient temperature would cause more variances in exposure than a 100 kHz ripple, so finding a fix for it would be a lot of trouble for not much benefit]

Hmm, the question of the effect on low-duty-cycle PWM is a good one. Here's what I'm seeing for 50% PWM with no capacitor.

The PWM period is 256 Hz, and 50% is showing a slow rise/fall. But even 1.56% PWM (1/64th; 6 stops below 100%) averages very close to what's expected according to my easel-meter. In this scope-capture, you can see fuzz around the line -- that's 100 kHz ripple that I'd like to remove.