LED lighting for macro illumination

[Bob^3]

The topic of using LEDs for macro illumination has come up in a number of previous threads. AndrewC posted some of his test observations here:

http://www.photomacrography.net/forum/v ... 6946#66946

So I thought it might be interesting to continue the topic in a new thread.

LEDS: even knocking a 6 amp pulse through 3 emitters I can't drop to 1/1000" pulses which is what I want for "flash" illumination.

Humm, that seems pretty low for a minimum pulse duration. Were these Cree XP-E LEDs? Wired in series or parallel? I’m pretty sure I could get pulse durations shorter than 1/2000 sec using a Cree MC-E LEDs, which is a quad unit (essentially four XP-E type emitters in one lamp). I drove them in series up to about 2 amps (each chip is rated at 700 mA max, continuous) for 0.5 ms and looked at the output using a fast avalanche photodiode connected to the o’scope. Based on the rise/fall times, I think it would have been possible to go shorter. I know that most single wavelength LEDs can be switched at below 1 uS, limited by the ESR and chip capacitance. But I’ve read that some of the high power white LEDs are limited by an “after glow” effect from the phosphor coating used over the base blue or UV emitter. I’m sure you noticed there is not much info on the net about this stuff. So it’s pretty much, try it and see.

In terms of achieving proper subject exposure, I think your goal of 1/1000 sec is doable (but borderline), depending on the number and peak power of the LEDs and the optical coupling efficiency. One huge advantage of flash over LEDs is that even small flash tubes typically produce thousands of times more peak light output than the current lot of high power LEDs. Here’s a link to an older post on CandlePowerForums regarding the issues of using LEDs for high speed photo flash (they do not discuss macro use and newer LEDs now have a few times higher power outputs, but the relative magnitudes of LED vs flash power output are still valid):

http://www.candlepowerforums.com/vb/sho ... p?t=121144

To determine what’s possible under optimized conditions, I’ve tested several LEDs in continuous mode using efficient parabolic reflectors that will focus most the light down to about a 10mm spot and made meter reading off of an 18% gray card using a macro lens at an effective aperture of f16. The Cree MC-E running at full rated power (700mA continuous per emitter) gave a correct exposure at 1/1500 sec. Add on layer of white Kleenex tissue over the reflector dropped the meter reading for correct exposure two stops to 1/375 sec. Given the expected coupling losses incurred using a FO cable (I haven’t tested this) the required exposure time will be lower---let’s say 1 stop, so 1/180 sec.

Therefore, it would require about 5.5 MC-E LEDs to allow a 1/1000 sec exposure. If you can pulse the LEDs at much high current to get let’s say 3x the light output, then you could possibly get to 1/1000 sec with two quad-emitter MC-E LEDs or 8 single XP-E LEDs.

I’m toying with an, AFAIK, unique idea (or more likely, I just haven’t found it in my searches) to improve the results, which I think has some potential (but not yet tested). Surround the macro subject with 10 to 20 LEDs in a spherical shape directed through something like Charles’ pingpong ball diffuser. A very large heatsink would be necessary to cool this thing to allow running all the LEDs at full power---maybe requiring water cooling or heat pipes. But that shouldn’t be necessary. For modeling, they could be run at a friction of the rated power and still provide plenty of light. Then, pulse them at high power for the exposure. Another interesting possibility would be to wire each LED (or groups of LEDs), separately, to individually adjustable current sources. This would allow for “custom” programmable illumination and shadow effects, in order to enhance structures on each specific subject!

Also, the previously mentioned issues with the non-blackbody color spectrum of many LEDs can be successfully dealt with by using high-CRI (Color Rendering Index) LEDs (CRI up to 93) and/or by using Adobe’s DNG Profile Editor to create custom color correction profiles that can be used in Adobe Camera Raw or Lightroom to normalize the spectrum.

If anyone sees flaws or issues in any of these concepts, I welcome critiques. That’s what I find so valuable about this forum.

More ideas, comments?

Bob

[g4lab]

Also, the previously mentioned issues with the non-blackbody color spectrum of many LEDs can be successfully dealt with by using high-CRI (Color Rendering Index) LEDs (CRI up to 93) and/or by using Adobe’s DNG Profile Editor to create custom color correction profiles that can be used in Adobe Camera Raw or Lightroom to normalize the spectrum.

Can you furnish a link to the spectrum of the supposed CRI=93 LEDs

I think that color rendition will continue to be a problem for critical applications. LEDs have probably gotten to be better than fluorescent lamps but they aren't simiulating black body radiators just yet from what I have seen.

[Bob^3]

Can you furnish a link to the spectrum of the supposed CRI=93 LEDs

I think that color rendition will continue to be a problem for critical applications. LEDs have probably gotten to be better than fluorescent lamps but they aren't simiulating black body radiators just yet from what I have seen.

I didn’t mean to imply that LEDs are now capable of producing perfect blackbody spectral curves. Here are the specs of two devices with CRIs of 93 and 92:

http://www.seoulsemicon.com/en/product/ ... rLEDp4.asp



and:
http://www.nichia.co.jp/specification/e ... T-H1-E.pdf

I have samples of the first one, the Seoul Semiconductor N42180 (lower plot). Obviously, there are still some anomalies in the spectrum compared to blackbody, especially the spike in the blue at 450nm (bleed through from the blue base emitter). But they are much improved compared with the earlier generations (top plot). As you say, these offer somewhat better CRI than the high-CRI fluorescent lamps (e.g. OttLite) often used for photography. I agree that for critical color applications such as documentary photography used to identify certain minerals, insects and grade gem stones, special light sources are necessary. Even electronic flash tubes don’t radiate as a perfect blackbody and some incandescents can also be problematic.

The question is for general photography of macro subjects for hobbyist use or for photo-art to for sale, is the color appearance due to variation in the LED spectrum significant enough to be an issue? My guess is that other factors like improper white balance in an image are often a bigger issue. Also, as I mentioned, Adobe’s DNG Profile Editor can correct much of the issue:

http://labs.adobe.com/wiki/index.php/DN ... les:Editor

Photograph a MacBeth Color Checker chart under the LED light source, convert it to DNG format and import it into the editor to create a custom color profile for use in ACR or Lightroom to correct other images taken under the same light. I’ve compared this method using LED illumination compared with images of the same subject taken in direct sunlight and the results are very similar, at least to my eyes with normal color vision. Even images shot under the lower CRI (80) LEDs can be corrected to a large degree, at least for non-color critical applications.

[elf]

I'd try one of these but I've never attempted to solder surface mount components

[Bob^3]

Elf, I can't do that either! Fortunately, most are also available on ceramic star boards that have larger solder pads. Don't know about those, though...but they look interesting.

[AndrewC]

LEDS: even knocking a 6 amp pulse through 3 emitters I can't drop to 1/1000" pulses which is what I want for "flash" illumination.

Humm, that seems pretty low for a minimum pulse duration.

You got the wrong end of the stick ! I can pump a 6A current regulated pulse of 50usecs through them. The point is that that doesn't put out enough photons. I've got a target exposure time of 1/1000" The minimum usable exposure time I've been getting is somewhat longer than that. You can keep stacking up more and more leds but that defeats my object of a compact light source. You might ask how is it that people use leds for camera flash - they don't go for a short pulse - just dump a super capacitor through the led, usually with some sort of current clamp and have a "flash" that is relatively long duration.

There is no need to mess about with fancy reflectors - most good power led resellers sell custom designed optics which can capture 90% of the output and throw it forward in a 10deg spread beam. Either with a conventional mirrored "bowl" reflector or using moulded plastic with internal reflection and an integral lens. Here's an example from Carclo:

http://www.carclo-optics.com/opticselec ... _optics=58

[AndrewC]

I’m toying with an, AFAIK, unique idea (or more likely, I just haven’t found it in my searches) to improve the results, which I think has some potential (but not yet tested). Surround the macro subject with 10 to 20 LEDs in a spherical shape directed through something like Charles’ pingpong ball diffuser. A very large heatsink would be necessary to cool this thing to allow running all the LEDs at full power---maybe requiring water cooling or heat pipes. But that shouldn’t be necessary. For modeling, they could be run at a friction of the rated power and still provide plenty of light. Then, pulse them at high power for the exposure.

It's not particularly unique but I can't find any links just now.

There is no need to for enormous heatsinks and adjustable power - you set a current source to the maximum you want to use and then use a PWM to reduce the power to a low duty cycle or generate a pulse (very, very low duty cycle !).

Unfortunately with LEDs you can't just keep turning the power up and getting a linear increase in output. You can usually give them a current pulse of 2 to 3x their rated continuous limit but after that there is often little increase in output. For instance Cree XP-G emitters are rated at 1.5A, you can pump 3A through them with no problem, 6A they are still quite happy - I left one strobing at 1msec per second (0.1% duty cycle) overnight with a tiny vaned heat sink with no problems - but no real gain in output compared to 3A.

Here are some results from MC-E emitters:

http://www.photomacrography.net/forum/v ... ed&start=0

Though these were run at 700mA and I haven't (yet !) gone back and tried hitting them with much higher current pulses.

Andrew

[g4lab]

I merely mentioned the spike in color spectrum because according to Rik
it cannot be"Lightroomed" away.

[Bob^3]

I merely mentioned the spike in color spectrum because according to Rik
it cannot be"Lightroomed" away.

True, not by the traditional tools in Lightroom or Photoshop. Setting simple white balance won't solve it and attempting to "fix" color tonal flaws in affected areas in the image would be an exercise in futility. But with the DNG Profile Editor, it is possible to almost fully correct the general shape of the spectral response and at least partially mitigate the blue spike. It is not perfect because it only samples certain colors based on the 24 patch ColorChecker chart. But since the chart does include some patches contaning blue (at 450nm), it does help. If the editor could create a profile based on the the larger. 140 patch. ColorChecker chart, the profiles would likely be even better.

[DQE]

I've had seemingly good success just exposing a photographic gray card to my cheap 80 lumen (3 watt, single LED) LED flashlight, and then taking photos via this light source. Then I use Adobe Camera Raw and its white balance "eyedropper" to finalize the color temperature. This *seemed* to work qualitatively for flowers and bugs.

The Adobe calibration process you mention should work better.

In retrospect, it seems that my LED flashlight photography shouldn't have worked with no obvious flaws for generic macro photographic purposes (bugs against flowers, etc). The color temp used was 5200F.

What am I missing? Perhaps I just need to try taking a photo of a color checker and see how it turns out?

[Bob^3]

Personally, I don't think you're missing anything! For non-color critical applications, simple white balance will produce neutral whites and gray tones. The only potential issue would be for subjects that happen to reflect strongly in the aberrant parts of the LED spectrum, like at the high blue spike wavelength of the CRI 60-80 LEDs (or at the dip in the cyan and lack of reds). So these colors will appear relatively more or less luminous than an image of that subject under a "full specrum" source. But chances are you're only going to notice this if you compare the two images. Fluorescent bulbs are the worst offenders due to the high, narrow spikes in the spectrum. And the DNG Profile Editor can't do a very good job normalizing this issue. The ColorChecker will allow you to visually determine how much the light source effects the way you perceive different pure colors and common hues, like skin tones.

[AndrewC]

....

Here are some results from MC-E emitters:

http://www.photomacrography.net/forum/v ... ed&start=0

Though these were run at 700mA and I haven't (yet !) gone back and tried hitting them with much higher current pulses.

Andrew

Gave them (the MC-E emitters) some serious abuse tonight - they are quite happy strobing away at about 9A for 10msec per sec. However, not a great deal of extra illumination compared to normal drive currents.

Next on my list might be some SST-90 emitters - should be able to knock out about 2,500 lumens at 9A, but they are expensive and hard to find on STARs with high CRI and high flux bins.

[Bob^3]

You got the wrong end of the stick !

Yep…seems to be the story of my life…or maybe that's the short end?!

I can pump a 6A current regulated pulse of 50usecs through them. The point is that that doesn't put out enough photons.

My apologies Andrew, I didn’t mean to second guess you (seems to be a simple misunderstanding of terms). It appears your observations regarding the non-linear relationship between current and light output are the same as mine. In case you haven’t seen this useful thread:

http://www.candlepowerforums.com/vb/sho ... 07&page=13

This guy has tested and graphed the current vs lumen output of a number of high power LEDs, (incl the MC-E)---all the way to failure. His workbench is a lot better equipped than mine. He does the tests by applying continuous current (not pulsed). So he mounts the LEDs on massive heatsinks to wick away the heat. Near the upper limit of current, the resulting curves are clearly logarithmic in form, rather than linear or exponential, approaching an asymptote at some level where doubling the current only produces a bit more light. So both the light producing efficiency and the thermal efficiency drop off quickly at higher levels. My results using current pulses show the same relationship. The peak light output may be a bit higher using pulsed currents due to the reduced emitter temperature, but not much. And he also tested this by letting the LED cool down to ambient, then applying a high current and making a quick measurement of light output.

I've got a target exposure time of 1/1000" The minimum usable exposure time I've been getting is somewhat longer than that. You can keep stacking up more and more leds but that defeats my object of a compact light source. You might ask how is it that people use leds for camera flash - they don't go for a short pulse - just dump a super capacitor through the led, usually with some sort of current clamp and have a "flash" that is relatively long duration.

I fully agree on all points. As I stated earlier, LED “flash” units simply cannot yet compete with ancient Xenon flash technology for peak light output. If the small flash tube in a portable flash like the Vivitar 283 could be run continuously at full power, it would be blindingly (literally!) bright. Unfortunately, the only way I can see to provide the necessary peak lumen output with LEDs and allow correct exposure at short pulse durations, is to either use LEDs that provide higher peak outputs or add more LEDs to the system.

There is no need to mess about with fancy reflectors - most good power led resellers sell custom designed optics which can capture 90% of the output and throw it forward in a 10deg spread beam. Either with a conventional mirrored "bowl" reflector or using moulded plastic with internal reflection and an integral lens. Here's an example from Carclo:

http://www.carclo-optics.com/opticselec ... _optics=58

I’m also using those exact Carlo lenses for some applications. And they are designed to produce a very good quality “collimated” beam. The parabolic reflector I spoke of is not a “fancy reflector” (as in expensive). It is an off-the-shelf unit from one of the retailers you spoke of (I think it was Mouser). And it is also custom configured specifically to focus the beam of a Cree MC-E. The reason I choose it for the exposure test above instead of the Carlo, is that the LED can be moved slightly back in the mount, which will cause the beam to focus to a 10mm (i.e. brighter) spot size a few cm in front of the reflector(I wanted to measure exposure at about 4x at a known effective aperture). This allowed me to measure best case exposure time, with very little light loss. As I recall, the Carlo lenses are a bit more efficient at collecting the light from the LED. But not as good at focusing it to a small spot (in retrospect, perhaps a positive lens in front of the Carlo would be best?).

It's not particularly unique but I can't find any links just now.

If you happen to run across those, I’d be very interested.

There is no need to for enormous heatsinks and adjustable power - you set a current source to the maximum you want to use and then use a PWM to reduce the power to a low duty cycle or generate a pulse (very, very low duty cycle !).

True, but as I stated the large heatsink would only be needed if the intention was to run at or above full rated continuous current---or for that matter, the average current (which is proportional to duty cycle) if driving with PWM. In other words, there is no net advantage in terms of the optical or thermal efficiency of the LED when driving with PWM compared with constant continuous current (however, the driver circuitry can be much more efficient as well as simpler). In fact since LEDs become less efficient at higher drive currents as was discussed above, using a PWM drive at current amplitudes in the nonlinear region of the LED and reducing the duty cycle to produce a given average light output may well generate more heat in the LED than driving it with a continuous current source at a level adjusted to produce the same light output. In addition, since power (heat) increases as the square of the current (I^2R) and the voltage developed across the LED increases with current, thermal issues may be very problematic at high peak PWM currents.

Here are some results from MC-E emitters:

http://www.photomacrography.net/forum/v ... ed&start=0

Thanks for the link to your MC-E post. I did read that informative thread when you first posted it, but had forgotten you were the author.

Next on my list might be some SST-90 emitters - should be able to knock out about 2,500 lumens at 9A, but they are expensive and hard to find on STARs with high CRI and high flux bins.

That’s one of the big monsters that I haven’t played with yet. I’d be very interested in seeing your results!

[Bob^3]

I've had seemingly good success just exposing a photographic gray card to my cheap 80 lumen (3 watt, single LED) LED flashlight, and then taking photos via this light source. Then I use Adobe Camera Raw and its white balance "eyedropper" to finalize the color temperature. This *seemed* to work qualitatively for flowers and bugs.

That's exactly what I do to obtain neutral white balance. Many folks seem to confuse color temperature (CCT in LED speak) with color rendering (CRI). Your 3 watt flashlight probably has a CRI below 80. And the 5200K CCT may make it appear to the eye to have a blue or greenish tint due to the large blue spike its spectrum. But setting the white balance as you've done on a known neutral tone, corrects this in the image. But again, subjects that reflect at the "blue spike" wavelength will appear more luminous than they would in say direct sunlight.

I think the key question here is do we really need the 100 CRI of a blackbody source to capture completely normal appearing images of natural subjects, which match our expectation of how we would perceive or photograph the subject in nature? To answer that question, just consider how often photographs are take in outdoor settings under close to 100 CRI lighting---the only common one that comes to mind is direct sunlight. As a very common example, when you capture an image in shady conditions under a clear blue sky, what is the spectral curve of the illumination---big blue-cyan spike, perhaps? And probably a pretty low CRI as well. Yet it we set the camera white balance on “shade” or photograph a gray card and set the white balance in Photoshop, do the results not look “natural”? And what about the spectral illumination and CRI when you’re deep in the shade of some canyon with blue skies overhead and bright red canyon walls illuminated by direct sunlight bathing the scene you’re photographing? Nice smooth blackbody radiation curve…I don’t think so!

[AndrewC]

My apologies Andrew, I didn’t mean to second guess you (seems to be a simple misunderstanding of terms). It appears your observations regarding the non-linear relationship between current and light output are the same as mine. In case you haven’t seen this useful thread:

http://www.candlepowerforums.com/vb/sho ... 07&page=13

This guy has tested and graphed the current vs lumen output of a number of high power LEDs, (incl the MC-E)---all the way to failure. His workbench is a lot better equipped than mine. He does the tests by applying continuous current (not pulsed). So he mounts the LEDs on massive heatsinks to wick away the heat. Near the upper limit of current, the resulting curves are clearly logarithmic in form, rather than linear or exponential, approaching an asymptote at some level where doubling the current only produces a bit more light. So both the light producing efficiency and the thermal efficiency drop off quickly at higher levels. My results using current pulses show the same relationship. The peak light output may be a bit higher using pulsed currents due to the reduced emitter temperature, but not much. And he also tested this by letting the LED cool down to ambient, then applying a high current and making a quick measurement of light output.

I never really took the time to comment on this before but for anyone following this who is interested, there are two issues with trying to get a lot of light out of an led:

1) Heat: overload an led by running it at too high a current for too long and something will melt, literally. Usually the very fine wires used to connect to the emitter but the emitter itself can get damaged by excessive heat. You can address this by using short pulses with a low duty cycle and/or good heat management

2) Quantum efficiency: basically you make light by generating electron-hole pairs, when these recombine a photon is emited. So a particular material stack has a maximum number of hole-pairs it can generate per unit area. Doesn't matter if you pump more current in, you just can't generate any more hole-pairs. Ways around that are to make multilayer stacks (the h-p generation of interest tends to take place where you join two similar but not identical crystalline materials - all the InGaN, AlGan, etc stuff you hear about) or increase the area with a bigger emitter. Bit like squeezing juice out of a lemon - doesn't matter how hard you squeeze it, once the juice is out it's out. Only way to get more is a bigger lemon. That must be one of my worst ever similes

3) There's actually a 3rd issue as well - you need to get the photons out of the material. That's why you can't just make a stack of 100 junctions and get out 100x the light of a single junction.

What does this mean in real life ? Well an MC-E probably has a relatively low quantum efficiency but a high area. Increasing the current gives a bit more light but even if you keep the heat under control you quickly run into a brick wall. An XP-G has a higher quantum efficiency so you can feed more current in and get more photons out. The Luminus emitters are probably high quantum efficiency, high area and multilayer stacks and I want one to play with

There used to be an interesting report from an Australian firm making supercapacitors called Cap-XX comparing Xenon vs LED for camera phones, but they no longer seem to exist on the web. The link I had saved is http://www.cap-xx.com/resources/docs/CA ... ash_v3.pdf but no longer seems to work.