Ultra Short, Ultra Fast LED Flash

[Zero999]

I've done this before. The requirement was for a 1µs. Phosphor LEDs are variable. Some are fast and others are slow.

Attached is a schematic of the test set-up I used to drive LEDs with 1µs pulses of hundreds of amps, hence the low value current limiting resistor. It's possible to overdrive LEDs by a factor of 10, for short pulse lengths. Note transformers are used to drive the MOSFET and measure the current, to avoid large currents passing through the grounds of the test equipment.

I'm working on digesting all the idea's that have been suggested here. Zero999, I'm just making sure my flea brain is understanding the diagram,  you use an inductor to drive the LEDs rather then using a Flash Capacitor.

No, C1 stores all the energy. None of the inductors are used for energy storage. T2 is a pulse transformer which drives the MOSFET. T1 is a toroidal current transformer, which measures the current through the LED. The current in the secondary is divided by the turns ratio. In the above schematic, the current through the secondary and R3 and R6 (the termination resistor at the oscilloscope end to match the 50Ω cable) is ¹/₁₀₀ of that in the single turn primary. I used several different transformers, depending on the current. Firstly a small 4mm ferrite ring with a 10 turn secondary and a 0.1Ohm resistor for R3, which was fine for up to 5A or so, then the  Murata 54100C (the one in the above schematic with both primary windings in parallel (datasheet linked below) and eventually a 10mm ferrite ring with 21 turns and 2R2 for R3. The reason for 21 turns was R1||R6 = 50||2Ω2 = 21, giving 0.1V per Amp in the primary.

https://docs-emea.rs-online.com/webdocs/0eb1/0900766b80eb1ad9.pdf
https://www.grainger.com/content/supplylink-what-is-a-current-transformer

I appreciate the lead on the LED you were testing. Vela mentions they spent a lot of time trying to find the right LED's for the job. I suspect this gets back to the issue of how long the phosphor stays excited. I'm fairly confident they are using white LED's, the images of the unit show the LED's and the yellowish tint characteristic of white LED's is clearly visible.  This would simplify the circuit, as we wouldn't need to worry about white balancing the RGB LED's. Since LED's MFG's aren't typically concerned about the switching time of a White LED, I can image there is a lot of variation, possibly even from lot to lot.

I have used a Vela strobe before for another test a few years ago. Unfortunately it failed to meet our requirements. It was bright enough, the problem was there was a huge seemingly random delay 8µs to 24µs, if I remember rightly, between the trigger pulse and the strobe firing. It used lots of LEDs in series and a microcontroller-based control system, which I believe was responsible for its downfall. If the variable triggering delay isn't an issue for you, it's a perfectly good strobe, otherwise steer clear of it.

So we would want the MC to trigger on the pulse, with a programmable delay (just a few microseconds) before sending the trigger for the flash.  By having a short delay, we can dial in the position of the bullet in the frame without physically moving the set up. I know I've tried to use an Ardiuno to create a short 500ns square wave but I seem to remember that it struggled. I don't think the code was fast enough to switch the I/O pin that quickly.  It will be another problem I'll need to work through.

I would not recommend using an MCU to control the triggering, certainly not directly, without any additional protection. If there's a problem with the firmware, it could blow up all the LEDs. A CPLD or even discrete glue logic gates are ideal. Whatever you decide, another layer of hardware to protect against too longer pulses, at too higher repetition rate, is a good idea. A high pass filter can be used to limit the maximum pulse length and a monostable multivibrator to control the minimum delay between pulses. Connect both to an AND function, before the MOSFET driver: MCP14A0301 looks good because it has an enable function, so the AND isn't required. I'd go for the 74HC123 for the monostable. It needs to be negative edge triggered with an inverting output, so the MOSFET driver is enabled, when the input goes high, then disabled, after the input goes low and triggers the monostable.

[Giaime]

You're neglecting the energy stored in the inductor

Oh, suddenly it's me who neglects energy stored in the inductor. 
Look, unless you show real-world shunt-PWM waveform that disproves what I say, we do not have any further progress in this discussion.

Here you can see a measurement on one of the systems I work with. This is exactly what we're discussing: a DC/DC already running from some time in short circuit, suddenly the shunt FET is turned off (you can see it's VDS - green - rising). This is a 1200ns pulse (digital1 is uC PWM) but you can clearly see it can be cut down to 500ns.
The limit here is the risetime of the current-voltage on the output node, largely due to the interaction of DC/DC current and node capacitance. Here for EMC reasons there was also 1nF additional capacitance at the output, if you skip this you can go faster. Also higher DC/DC currents helps a lot, here it's only 400mA.

I have a lot of experience with this kind of systems, if you or the OP needs some more info I'm here.

[ogden]

Here you can see a measurement on one of the systems I work with.

One scope screen is worth thousand words Thanx! I am really surprised that it works so well. Until now regarded shunt PWM as snake oil, especially when saw na'ive simulated waveforms in the TI datasheet. What is specs of inductor and LED used, what is frequency of the converter while shunt is short/open?

[StillTrying]

I can't see how this inductor storage works for 1 very big triggered LED flash.
The single ~1us pulse applied to the LEDs needs to be around 30A to 100A at 35V to 120V.

[ogden]

I can't see how this inductor storage works for 1 very big triggered LED flash.

It does not. We already discussed here that capacitor is much better "tank" for LED flash energy storage. I was arguing that shunt PWM cannot achieve speeds required for application, yet here we have proof that I was wrong. Yes, "shunt PWM" is not suitable for LED Flash, yet I (personally) do not see any problem to discuss it here.

[mzzj]

Pretty inexpensive  for 10,000LM, but if I drive these by 1000% and get 10 of them. Then I could match the 1mil Lum.

10K lm? - Scam. With such specs he can put OSRAM out of business, yet they are doing well You may indeed user those to start with, but do not expect to get even close to specified light output. Not to mention CRI

10k Lm out of 100W COB is hardly any sort of state-of-the art nowadays so its certainly possible even for low cost chinese COB.
Probably the chinese are bit "optimistic" and you get only 8kLm and lifetime is certainly a mystery vs. more expensive brand name product.

[Siwastaja]

Here you can see a measurement on one of the systems I work with.

One scope screen is worth thousand words Thanx! I am really surprised that it works so well. Until now regarded shunt PWM as snake oil, especially when saw na'ive simulated waveforms in the TI datasheet.

Shunt switching, driven from a constant current supply, is widely used in 3D Time-of-flight designs, with pulses down to around 10ns. I designed mine using a voltage source and series resistor instead, encouraged by our chip supplier's reference design with such a simple topology, but OTOH I'm only modulating at 20MHz (hence, 25 ns pulses with 25ns offtime). Works well enough for us, although we do see a drop in differential "payload" signal amplitude at 20MHz, but I'm almost certain this is limited by the LEDs, more than by the switching topology.

[Zero999]

Pretty inexpensive  for 10,000LM, but if I drive these by 1000% and get 10 of them. Then I could match the 1mil Lum.

10K lm? - Scam. With such specs he can put OSRAM out of business, yet they are doing well You may indeed user those to start with, but do not expect to get even close to specified light output. Not to mention CRI

10k Lm out of 100W COB is hardly any sort of state-of-the art nowadays so its certainly possible even for low cost chinese COB.
Probably the chinese are bit "optimistic" and you get only 8kLm and lifetime is certainly a mystery vs. more expensive brand name product.

Yes, the CMA3090 I tested is rated for 12k Lm at the maximum rated drive current of 3.6A. Over-driving does increase the intensity, but it's not linear, especially at higher currents, which result in diminishing returns. According to the results of my experiment, with an LED drive of 35A, the photodiode current was 5.35 times, that when the LED was driven at the rated current, which seems a reasonable and is way below the 120A which was destructive. You could have an array of 16 LEDs, each driven at 36A to give 1M Lm.

[Marco]

Shunt switching, driven from a constant current supply, is widely used in 3D Time-of-flight designs, with pulses down to around 10ns.

With much lower pulse energy. Inductive energy storage has multiple problems as you try to scale it up in energy. Bandwidth, inductor cost and standby power consumption. Bandwidth isn't really the problem here, inductor cost and standby power consumption are nearly insurmountable problems. You can either have a couple bucks worth of electrolytics which can run on AA batteries, or you can have a couple 100 bucks worth of inductors which require a mains connection and active cooling.

Get back to this when we get room temperature superconductors.

[David Hess]

With much lower pulse energy. Inductive energy storage has multiple problems as you try to scale it up in energy. Bandwidth, inductor cost and standby power consumption. Bandwidth isn't really the problem here, inductor cost and standby power consumption are nearly insurmountable problems. You can either have a couple bucks worth of electrolytics which can run on AA batteries, or you can have a couple 100 bucks worth of inductors which require a mains connection and active cooling.

If a fixed pulse width is acceptable, then a charge line or for longer pulse widths, a lumped delay line made from inductors and capacitors can be used instead of an inductor.

The way to do it in this application though is a switched current source or switched voltage source and ballast resistor.

[Marco]

With a switched current source with ballast you retain the problem of the amount of circulating current.

If you go the low voltage COB way, you could easily be talking about 1000A of current. That current isn't started a ms or so before the trigger either. Much of the time in these photography setups the events are kicked off by slow humans, so count on that current to be circulating for minutes rather than milliseconds. Even if you went for the very complex solution of circulating the current through a regenerative power supply so you don't have to burn the full Vs*I, keeping a 1000A running through some fast MOSFETs is going to burn way too much power to run this on a couple of AA batteries.

It's not worth it, he's not looking for single digit ns rise/fall times or 1% precision regulation of the current. Just dump some voltage on and take it off again with a MOSFET, good enough.

[David Hess]

With a switched current source with ballast you retain the problem of the amount of circulating current.

Sorry, in this case I meant driving a current into a cascode transistor (emitter/source switching) which then handles the voltage compliance.  An actual switched current source could be much faster but would also dissipate the peak power continuously.

The advantage is that the switching transistor can be much smaller for lower capacitance and higher speed and the fixed voltage across the resistor yields a more accurate and regulated output current.

[Marco]

I guess a little overvoltage with active current regulation makes sense given the likely high parasitic inductance of a COB (anyone measured it?). I don't really see the need for a cascode though, a simple source resistor with a BJT to pull down the gate should be enough to get decent  regulation. Or just a source resistor period for that matter.

[Zero999]

If a triangle wave will do, an inductor can be used to limit the current. I did a quick simulation in LTSpice and a 50nH series inductor with a freewheeling diode could theoretically limit the current to 90A into the Luminous devices PT-121-B, which is similar to one of the LEDs I've tested. 50nH is small and theoretically could be an air core.