If you want to take a picture of something fast, and we mean really fast, you need to have a suitably rapid flash to illuminate it. A standard camera flash might be good enough to help capture kids running around the back yard at night, but it’s not going to do you much good if you’re trying to get a picture of a bullet shattering a piece of glass. For that you’ll need something that can produce microsecond flashes, allowing you to essentially “freeze” motion.
You can buy a flash that fast, but they aren’t common, and they certainly aren’t cheap. [td0g] thought he could improve on the situation by developing his own microsecond flash, and he was kind enough to not only share it with the world, but create a fantastically detailed write-up that takes us through the entire design and construction process. Even if you aren’t in the market for a hyper-fast flash for your camera, this is a fascinating look at how you can build an extremely specialized piece of gear out of relatively common hardware components.
So what goes into a fast LED flash? Rather unsurprisingly, the build starts with high-quality LEDs. After some research, [td0g] went with an even dozen CREE CXA2530 arrays at just shy of $7 USD each. Not exactly cheap, but luckily the rest of the hardware is pretty garden variety stuff, including a ATMega328P microcontroller, some MOSFETs, and a TC4452 driver. He did pack in some monstrous 400 V 10μf capacitors, but has since realized they were considerably overkill and says he would swap them out if doing it all over again.
To make development easier (and less costly, should anything go wrong), [td0g] designed the flash so that the LEDs are arranged in banks of three which can be easily removed or swapped in the 3D printed case. Each trio of LEDs is in a removable “sled” that also holds the corresponding capacitor and MOSFET. Then it was just a matter of getting the capacitors charged up and safely dumping their energy into the banks of LEDs without frying anything. Simple.
At this point, the astute reader is probably thinking that a high speed flash is worthless without an equally fast way of triggering it. You’d be right, but [td0g] already figured that part. A couple years back we covered his incredible ballistic chronometer which is being used as a sensor to fire off his new flash.
It is funny that in both articles (this and ballistic chronometer) are about fast things and Tom Nardi start by mentioning kids running :)
“A standard camera flash might be good enough to help capture kids running around the back yard at night…”
“It can be hard enough to take a good photograph of a running kid or pet, and if we’re being honest…”
[td0g] post is really well written. Each decision was well thought and experimented before commit to final version.
He has fast kids?
When my daughter was younger, when we saw her doing something “cute” and wanted to take a photo.
By the time the camera booted, she had moved on to something else…
Film (not movie, but analog) cameras for the win
Fast Times at Ridgemont High
You know, I realized that and had considered changing the intro. But since the chronometer post went out I’ve had another kid and now the challenge is greater, so the reference seemed more appropriate than ever.
Thanks for the kind words Fred!
Some kids are remarkably quick… ;)
A traditional flash can’t stay on continuously, LEDs can.. making a super fast flash with LEDs doesn’t make much sense. Just turn the LEDs on, shoot the glass, turn the LEDs off. It’s a 5 minute build.
That’s not the point. LED-Flashes are used because they are much quicker than the shutter of the camera.
High speed photography does not work by illuminating everything and then use the shutter to capture the event. You do it in a dark room, leave the shutter of the camera open for a long time and just use the flash when you want to take a photo, as the flash has much less delay between the trigger and the event and can reach much shorter exposure times (less motion blur)
That’s exactly right. The max speed of most camera shutters is 1/8000 seconds (125 microseconds), which is way too slow. Plus it’s a focal-plane shutter, which means the whole image would be screwy if when photographing a high-speed projectile. Finally, controlling the shutter timing is problematic – the camera has its own lag which seems to change a little from shot to shot. One millisecond of lag and a projectile won’t even be in the frame anymore!
Therefore, the easiest solution is a high-speed flash in a dark room.
The same set up can also be used for strobe photography by just changing the firmware.
Just turn on the LEDs.. makes no sense to make them into a flash.
The LEDs blink to only illuminate the object for a really short period of time – and so freeze the motion – the camera shutter is open for way longer than the period of the flash.
Fine work.
Turning them on for a very limited time means you can over-drive them (a lot), and not need to worry about heat.
Melting the bond wires (if the LED has them) and electromigration set the limits of current you can use. Minimum flash duration with white LEDs is determined by the phosphor that takes time to light up and remains lit briefly after the actual light pulse from the LED(s) has ended. You can probably get more efficiency and shorter flash durations using LEDs without phosphors.
Interesting idea. I’m going to look into this some more!
When you want to photograph something much faster than your shutter, this is the only way to do it, otherwise you have motion blur. Also when you turn them on for only a microsecond, you can squeeze much more power for that instant, even 10x more.
It’s not just faster than your shutter but when having moving objects that highlight rolling shutter artefacts
I believe most cameras start performing this when you capture something with an exposure less than 5ms long (200th of a second). So even if your camera claims to be able to capture an event in 8000th of a second, the resultant image is not a reflection of reality at a single time instance.
I believe good photo cameras in long exposure mode don’t have rolling shutter, but collect charges on whole matrix and then serially read them, but collection is from one instant or whole exposure. With bad cameras in film mode, this is visible. Also with rolling shutter and very fast flash, you would have bright rows collected during flash and rest of frame would be dark.
Is it possible for a camera sensor to stop light from charging up the pixels after the electronically-decided frame time ends? The only designs I’ve seen need to physically close the shutter to end the shot. That means you’re always going to have a rolling shutter type of effect at 1/8000th (for a consumer camera without a specially-designed superfast shutter, at least).
Yes, electronic exposure control is possible. Many cameras these days do not even have a mechanical shutter. The best option for this is a special sensor, I think it’s called frame transfer CCD, which shifts the whole picture into storage to end the exposure.
But it’s possible, that CMOS sensors offer comparable mechanisms now.
My analog-era flash unit claims a duration of 1/10,000 second at it’s lowest-power setting. Would that do?
It cost ~AUD$300 at the time – mid 80s – and offered a range of settings from 100%, through 1/2, 1/4, 1/8, down to 1/16th – that’s where the claim of a 1/10,000 second exposure comes in.
> For that you’ll need something that can produce microsecond flashes, allowing you to essentially “freeze” motion.
100 microseconds is not good enough. In that time, bullet travelling at relatively slow 400m/s will travel 4cm, which is much more than length of that bullet. It will look like a streak. At 1 microsecond it will travel 0.4mm, a little blurry but manageable.
Perhaps if you read the article you’d get your answer… Given that the author showed a comparison of this setup vs. an SB-900 at minimum power (shortest possible duration)
You certainly can have short pulses using standard speedlights – you just need to use low power settings. A cheap Yongnuo YN-560 can get 1/23,000 (around 40 microseconds) on lowest power, which is not bad at all.
Still about 2cm of travel of a bullet.
“is not bad at all” to describe something that was shown, in the linked article with photographic evidence, to be woefully insufficient?
Exactly. The Nikon SB-900 at 28 microseconds ‘isn’t bad’, but it sure doesn’t make sharp images either!
It seems those are white LEDs working with a phosphor to create a spectrum – those phosphors need time to full brightness, and have an ‘afterglow’. To fully use the fast-bright-fast-dark potential of LEDs you’d need to use RGB-LEDS (or IR, if you are not interested in color but want to use the most efficient wavelengths for both LED and sensor)
But I think the rise and fall times are in the 1ns range. So no problem here
I agree it looks strange that td0g has better timings with white LED (which is blue LED + phosphor) than xenon tube flashes.
I linked to an application note from Cree, those LED’s take nanoseconds for illumination to rise/fall.
I’ve looked for Cree’s decay time specs many times. Where did you find them?
I find it odd that Cree specifically omits specifications for pulsed operation on this entire class of LEDs, but it is a common spec on other LEDs. (but it’s ironic that their published specs are only on pulsed measurements, because that’s how they are tested)
https://www.cree.com/led-components/media/documents/XLampPulsedCurrent.pdf
Page 10. Now if I’m misunderstanding this, please let me know! Several commenters have shown concern that the rise/fall times are too slow.
OK. I had seen that app note before. I originally interpreted their ” is limited by the turn-on time of the device” as meaning the driver device, but it could mean the lamp. In any case, they do say it’s only the turn ON time. They still don’t mention the phosphor decay time, which I mention below I measured to be 2 microseconds (on similar but not-Cree LEDs).
If you look carefully at high speed projectile images you’ll see the asymmetric colour in the leading-vs-trailing edge blur. In the airgap flash it’s present, and it’s reasonable to assume it’s just the blackbody radiation cooling off. Seeing it in LED flash images surprised me, which is why I looked for the data, couldn’t find it, and then finally just measured it.
That’s interesting, thanks for pointing this out. Given the application note and the fact that the Vela One uses similar LED’s, I made an assumption that may have turned out to be wrong.
Vela describes some of their testing here and claims that their flash illuminates for 3us when set to 3us: http://www.vela.io/posts/how-fast-is-your-flash . I don’t know whether to believe it or not.
A lot of the light from the LED is the blue component, which goes away within nanoseconds. It’s the remaining yellow phosphor component that takes a while to fade away. Depending on where they set the threshold (like at the 50% level), it’s easy to “see” a light duration of 3 us from a 3us current pulse. What’s telling is the so-called “t0.1” time — the time it takes for the light curve to fall to 10% of the peak intensity. At that point it’s entirely yellow (phosphor) light, and it is that component that has the characteristic 2us decay time. If you squint right you might believe you can see it in Vela’s ‘scope plot on that page, but it’s obscured by some kind of lowpass filter on their detector, which itself has about a 1us response time.
BTW: the photodiode-50ohm detector they describe is an excellent way to measure fast optical waveforms. I have a very similar design that I measured (with a 7 nanosecond laser pulse and 1 GHz 5 GS/s scope) to have a 25 nanosecond response time. Use a ‘scope with a 50 ohm terminator to measure it though, not an arduino.
Inspired by the Vela One, I built a high speed 10 kilowatt LED flash in 2014, capable of 1 microsecond pulses. It used 10 “10 watt” cheap white LED modules (not Cree), grossly overdriven.
I measured the phosphor decay time with different colour filters.
My electrical waveform had a 100 nanosecond rise & fall time, and the risetime of the light output was that fast.
The output light fall time was 100 nanoseconds in the blue (the source LED), but the phosphor had a much longer decay: 2 microseconds.
That sounds awesome. Any build details somewhere?
Not really. After my experience with Maurice Ribble’s Airgap Flash (see https://www.youtube.com/watch?v=kc3fFIetbeI and https://www.youtube.com/watch?v=L5CvJuQA7X4
), I wanted a safer (and quieter!) equivalent, so looked to the Vela One for inspiration.
I followed pretty much the same design steps as [td0g], but didn’t proceed to a finished device because, frankly, the performance sucked. At 10 kilowatts (100x overdriven) the luminous efficacy of the LEDs is a putrid 15 lumens per watt — the same as the airgap flash. But the airgap flash is about 600 times more powerful (6 megawatts).
I came to the conclusion that, to match the performance of the airgap flash with LEDs, I’d need $10,000 in parts (in 2014 — maybe cheaper now). So I abandoned the project.
If you want to build your own, [td0g]’s design is a fine place to start. Or live on the edge and build a lethally-dangerous and deafening airgap flash…
Looking at this a half decade later. [tf0g] uses LEDs 7x more powerful than mine were (for about the same cost!), with a better luminous efficacy, so that factor of 600x in performance for an airgap flash is more like 50x now. Coupled with the much better/cheaper modern image sensors, LEDs are much nearer to being a viable replacement for an airgap flash.
Apropos of LED flashes: the FastLane/EZPass (toll transponder system in the US) gantries use an LED flash to capture the plate numbers. I notice they’re constantly emitting a low level of visible light, and then flash as a vehicle is sensed. Don’t know if they’re visible, IR, or a combination.
I built an LED strobe to use as a microscope illuminator about 10 years ago. I used a single 5W LED and hit it with about 10X (or was it 15x?) rated current pulses from a capacitor. It was controlled by a PIC uC that I could use to vary the pulse duration a rate from 1-15 us and 1- 255 Hz. I used it to observe ciliated microorganisms and could adjust the strobe to freeze the motion of the cilia or render them in slo-mo. At the time I used a blue LED, but if I were going to do it again, I’d probably go with green- the eye is more sensitive to green and it’s less likely to damage your eye.
As I recall, 10-20 Hz was the sweet spot for most protists, which was very tiring to look at.
You know those wave patterns you see in the legs of centipedes as they run around? You can see the same sort of patterns in the cilia of protists. The interesting thing is that the protists have no nervous system, so it’s all powered by cascading chemical reactions with no central coordination.
How does it compare to using a high voltage MOSFET to switch off a Xenon flash well before it “tails off”?
Standard camera flashes including the one referenced in the article already do this (usually with IGBTs though). The IGBT is used as a series switch to turn off the current to the xenon flash tube. To get really fast pulses, more exotic pulse forming networks are required and as I recall the physics of xenon ionization limit the minimum optical pulse width. Take a look at air gap flashes for faster optical pulse width (and a good way to accidentally kill yourself if you don’t know what you are doing).
Low pressure xenon will have a long tail, it can’t cool fast enough. High pressure xenon is better, but those things are bombs waiting to go off. I’d feel safer with an air gap.
That’s a good question, why not use an air gap?
Safety, color temperature, size, cost
Air gap flashes are interesting and useful in a lab environment but difficult to use in less controlled environments.
Yeah. I’ve got two high pressure xenon flash systems for our pick and place. Original and replacement. There are warnings all over it about the explosion danger.
I wanted to replace them with LED, but figured I could only generate 3% of the light output best case following the best LED specs I could find. I figured I could overdrive them but I was thinking 2-3X, still too low. 10X… that’s starting to get into a range that might work. Hmm…
Why would pick and place need a (high power) flash? I’m curious.
Xenon flash tubes don’t have enough power to properly expose an image in a microsecond. Imagine a Nikon SB-900 at minimum power (28 microseconds), then divide that exposure by 28! That’s about 4.5 stops less light.
Air-gap flashes are similar to Xenon flashes, but can produce the power. Problem is that they require higher voltages (dangerous) and are difficult to control. LED’s are safer and much more controllable.
i’ve had several smartphones, and two kids, and only the google ones (nexus 4, nexus 5x) have had big enough sensors (pixels) to take a clear picture of kids running
I had to use old flashes triggered by an arduino and a few other components to get this water drop falling: https://hackaday.com/2014/03/01/stop-motion-water-droplets/ Being an non engineer person, I´m wondering how this could help us getting sharper images or any other improvements. Is there any enhancement compared to regular flashes? Flashing times? frequency?power? You hackaday people, please help here!
Flashing times and frequency. Power not so much, but it’s much easier to scale this if you need. However this setup can achieve more power density, that is more powerful but shorter flashes. When you drive leds for such short time, you can easily put 10x more power into them than specified.
Truthfully, a typical camera flash is fine for water droplets. They’re not moving quickly. If you drop it from 2 metres, it will only get to 6 m/s. I doubt most droplets reach that speed.
It’s when you start photographing projectiles traveling hundreds of metres per second that you benefit from a high-speed flash.
Watch out rotolight, youve now got competition
Hi all, I’m faschinated by the opportunity to read this project to build my own ballistic Flash but have not been able to locate any design description. Can someone, please, provide an updated and available link to get the design description?
Thanks in advance, Gianni.
The link in the article points to the design files. I just checked. It works.