Ultra Short, Ultra Fast LED Flash

[Zero999]

Some oscillograms.

A low current test: 3A, but the LXM2-PH01-0070 is only rated to 700mA continuous. The signal generator waveform isn't great because I used it to drive the gate drive transformer, without a MOSFET driver IC. The photodiode was about 100mm from the LED.


A high current test: 60A through the CMA3090, which is only rated to 3.6A continuously. The photodiode distance was increased to 200mm, otherwise it saturated. I managed to increase the voltage to 210V, giving a 120A pulses, before the LED failed open circuit.

[Siwastaja]

I do not see inductor as better solution than simple resistor. I would just make 3-channel system with individually programmable voltages and timing for each color, obviously with (inrush) current limiting resistors to drop like 3..5% of voltage on each string. Not for god's sake 30%. It is not needed to increase voltage specs of whole system just because someone here in the forum said that ballast resistor shall drop 30% of the voltage which for high power LED's shall be considered as insanity.

I said: at least 20-30% for good current regulation (against Vf variations through temperature and unit variations). This may or may not be important. For a low-duty special effect equipment, a 20-30% efficiency drop might not be a problem at all. For a general purpose LED lighting fixture, it would be an environmental disaster IMHO, so I understand where you are coming from.

Yeah, don't follow my advice blindly, do your own analysis. 20-30% drop was based on my own work on time-of-flight imaging where exposure consistency does matter.

Of course, if current regulation is unimportant, a no-resistor-at-all, limited-by-bondwires approach is often just fine. Or as you say, a few percent of the supply voltage. If you just need light for some photographic trick photo and have enough dynamic range in the imager, and are not expecting predictable exposure within a few percent, and your duty cycle is low and you are confident you are not killing the LEDs on unpredictable overcurrent, by all means reduce the waste power in the series resistor.

You may even be able to choose the FET so that its Rds(on) temp coeff compensates (at least partially) for the LED Vf temp coeff, regulating the current! This can be a neat trick.

You may even want the current to increase with increasing temperature (within reason; kind of controlled/anticipated "thermal runaway"), because LEDs produce fewer photons per coulomb at high temperatures, so you may want to increase the current with temperature to keep the light output constant. If you want this, then use a resistor "too small".

Of course, all of this babble is just about finetuning a simplistic, easy approach which, by definition, will be inaccurate in light output. I'm assuming this is good enough for the OP.


As a general note, I have seen that the very short pulse handling capabilities of LEDs often tend to be better than specified, but the efficiency does go down, so increasing current a lot has diminishing returns.

[ogden]

You may even be able to choose the FET so that its Rds(on) temp coeff compensates (at least partially) for the LED Vf temp coeff, regulating the current! This can be a neat trick.

Good idea indeed, but you don't have any control over results *and* you shall thermally connect FET to LED. "Standard" and more controllable approach would be to measure LED temperature, set voltage accordingly (assuming that LED V/I curve is known, measured in advance).

[ajb]

Many switching constant current LED drivers support PWM dimming by using a transistor to short the LED(s), so you could do the reverse here.  Establish your target current in the inductor, then open the switch across the LEDs for your desired duration, then close it again.  If the pulse is long enough and the response of the switching converter is fast enough the converter will pick up the load to maintain the LED current, otherwise you can just use a big enough inductor to sustain your required pulse.

Been there, done that. This is the best way. Use Ti parts like LM3409 or TPS92641, TPS92515, etc, all support shunt PWM dimming. You can get 300-500ns pulses like that if you control the circuit inductance carefully.

Why, why do you people suggest such  ?

First one, LM3409. Datasheet states that buck regulator frequency above 1MHz is hard to achieve. This means that one buck regulator pulse alone is longer than required, not to mention that first pulse most likely will not reach nominal current. Following waveform from LM3409 DS clearly shows problem why buck regulator and inductor-based approach as such is not the best choice (to say it politely) for this application:

You've missed the boat.  The point is not to turn the switching regulator on at the start if the pulse and off at the end, that wouldn't work well at all, as you've mentioned.  The point is to turn the converter on ahead of time, while the LEDs are shorted by the PWM transistor.  This will establish the desired current in the inductor, then when you want a pulse of light, you turn the PWM transistor off, which will allow the inductor current to flow through the LEDs instead.  This way the current is directly controlled, rather than relying on a regulated voltage and a resistor (which becomes subject to all sorts of other variables that will affect the actual current through the LEDs). 

This is a fairly conventional topology for PWM dimming with constant current switching converters, and many have built-in support for it, like in the example below.  This shows the dimming transistor being controlled through the switching converter; this is often done to allow the converter to adapt its switching behavior or even apply power saving behaviors when the dimming transistor is active, but driving the dimming transistor externally may be better for this application.

[Zero999]

Many switching constant current LED drivers support PWM dimming by using a transistor to short the LED(s), so you could do the reverse here.  Establish your target current in the inductor, then open the switch across the LEDs for your desired duration, then close it again.  If the pulse is long enough and the response of the switching converter is fast enough the converter will pick up the load to maintain the LED current, otherwise you can just use a big enough inductor to sustain your required pulse.

Been there, done that. This is the best way. Use Ti parts like LM3409 or TPS92641, TPS92515, etc, all support shunt PWM dimming. You can get 300-500ns pulses like that if you control the circuit inductance carefully.

Why, why do you people suggest such  ?

First one, LM3409. Datasheet states that buck regulator frequency above 1MHz is hard to achieve. This means that one buck regulator pulse alone is longer than required, not to mention that first pulse most likely will not reach nominal current. Following waveform from LM3409 DS clearly shows problem why buck regulator and inductor-based approach as such is not the best choice (to say it politely) for this application:

You've missed the boat.  The point is not to turn the switching regulator on at the start if the pulse and off at the end, that wouldn't work well at all, as you've mentioned.  The point is to turn the converter on ahead of time, while the LEDs are shorted by the PWM transistor.  This will establish the desired current in the inductor, then when you want a pulse of light, you turn the PWM transistor off, which will allow the inductor current to flow through the LEDs instead.  This way the current is directly controlled, rather than relying on a regulated voltage and a resistor (which becomes subject to all sorts of other variables that will affect the actual current through the LEDs). 

Yes, that will work and I did hint on using it in posts 10 and 13. The main downside is the inductor needs to be charged up before the pulse, which takes time. It's not a problem if there's an advance warning before flashing the LED, but it's no good if it has to trigger as quickly as possible.

[Marco]

Lets do some numbers, lets say you want 10 kW for 1 us. That's 0.01 Joule, lets round that up to 0.1 Joule to keep the current approximately constant for the pulse ... also a relatively safe amount of energy.

Lets say you use 100V string of LEDs (LED filaments would work for that) so you need 100A of current, so you need 20 uH of inductance, of course with a SRF>>1 MHz. That's a spicy inductor. Compare that to storing 0.1 Joule at 100V in a capacitor, that's a 2 dollar electrolytic capacitor in singles.

[ogden]

You've missed the boat.

Yes, kind of. Did read only first datasheet.

This will establish the desired current in the inductor, then when you want a pulse of light, you turn the PWM transistor off, which will allow the inductor current to flow through the LEDs instead.

I am afraid that this is where you are missing the boat. When load is shorted, regulator provides let's say 0.1V @ 10A  = 1W power. When shunt is open then regulator shall suddenly jump to 20V @ 10A = 200W power (or whatever is operating voltage of the LED's). Voltage of inductor cannot jump instantly, it is against it's formula V = L(di/dt). Sure/agreed - stopping/starting converter is even slower. Inductor can't compete with capacitor no matter how you twist it. I did not find any real, measured LED current PWM shunt waveforms for this case. If you know any or can measure - please provide, would be nice to see.

[ajb]

You've missed the boat.

Yes, kind of. Did read only first datasheet.

This will establish the desired current in the inductor, then when you want a pulse of light, you turn the PWM transistor off, which will allow the inductor current to flow through the LEDs instead.

I am afraid that this is where you are missing the boat. When load is shorted, regulator provides let's say 0.1V @ 10A  = 1W power. When shunt is open then regulator shall suddenly jump to 20V @ 10A = 200W power (or whatever is operating voltage of the LED's). Voltage of inductor cannot jump instantly, it is against it's formula V = L(di/dt). Sure/agreed - stopping/starting converter is even slower. Inductor can't compete with capacitor no matter how you twist it. I did not find any real, measured LED current PWM shunt waveforms for this case. If you know any or can measure - please provide, would be nice to see.

You're neglecting the energy stored in the inductor, and that the defining characteristic of an inductor is that it resists changes in current, which is the whole point.  We're literally talking about a boost converter minus the output capacitance and with a larger inductor.  Boost converters clearly work, including constant (ish) current LED boost drivers, so there's no reason that this won't too.  Of course there's plenty of room left to argue about the practicalities, costs, and other relative design tradeoffs of the inductor vs capacitor based solutions, and a lot of that comes down to how closely you need/want to regulate the LED current, and how much energy is involved.  The OP said 75W * 500ns, while for some reason Marco has decided to make the problem much harder by assumed 10kW for 1us, which is 266 times as much energy.  (I have no idea how much energy is actually required to get the results the OP is after, but chances are it's somewhere between 75W and 10kW...)

[ogden]

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.

[Marco]

In retrospect I decided that 75W isn't really a replacement for an open air spark gap ... it's not going to be enough to expose a normal camera.

He's going to need some serious power, keeping the energy to supply that power circulating in an inductor is awkward. For a couple ms it's not so bad, but when you want to do it continuously to be ready for a trigger you need so much copper to keep the losses down. Meanwhile a capacitor can just sit there minding its own business doing nothing.

[StillTrying]

LED or xenon, 10s of kW is needed for a small close-up 1us exposure.

[Marco]

Unless it's some kind of special high pressure Xenon lamp it will have lousy fall off.

[JAndrew]

75 Watts is a starting point, it may not be enough light.

If I can get a 500ns 75w flash then scaling up as needed is probably doable.

Xenon Lamps are too slow on the fall off and will create motion blur.

Historically units like these were used.

https://www.ietlabs.com/genrad-1538-a-stroboscope.html

I can pick one up on EBay for ~$300 dollars but the light isn’t as bright as a air-gap. Maybe it’s worth it to try and pick up a couple of these old units rather then to try and cobble together my own.

Sent from my iPhone using Tapatalk

[Marco]

75 Watts is a starting point, it may not be enough light.

In that case I'd try to drive one of those 12V 10W COBs with 24V, that should just about get you there electrically (lil higher). Use a MCP14A0302 with a simple ~500p/1K high pass filter on the input to create your pulse from a 5V step (unless the Arduino can generate a 500ns pulse on its own). Something like BUK7M45-40E to drive the COB.

[Siwastaja]

You may even be able to choose the FET so that its Rds(on) temp coeff compensates (at least partially) for the LED Vf temp coeff, regulating the current! This can be a neat trick.

Good idea indeed, but you don't have any control over results *and* you shall thermally connect FET to LED. "Standard" and more controllable approach would be to measure LED temperature, set voltage accordingly (assuming that LED V/I curve is known, measured in advance).

They tend to be somewhat thermally coupled, because the inductance needs to be minimized as well, so they are close. Of course it isn't exact.

[Wolfram]

It probably makes sense to get an idea of exactly how much light you need, this is one of the most important design specifications and it affects a lot of design choices. A few exposures using the planned camera system, and a charged electrolytic dumped into an LED through a resistor, will already tell you if your planned 75 W for 500 ns is a realistic starting point.

[JAndrew]

When I approached the project. I broke it down into two main problems. The trigger, and the flash.

The trigger is fairly easy to solve, there are various optical sensors I can use to trigger the flash as the bullet passes buy. Even put in a variable time delay.


The flash and the flash duration has been the harder problem for me. Again my thinking was using power MOSFETs to control a bank of LED’s. If I needed more light I would add more banks of LED but the circuitry would all be the same. Regardless how I go, my thought would be as long as I have one circuit that works setting up several in parallel or looking at ways to overdrive the LED’s for a brief moment wouldn’t be a challenge.

I tried to use an Arduino as a pulse generator. I figure that would be a pretty easy way to go. I could specify the length of the flash, time delay, and even do several flashes. However I don’t think I could ever get the program to run fast enough to switch the I/O pin that quickly. I remember playing with it, and looking at things like the Beagle Board or Raspberry Pi to see if I could do with with a faster processor. Even bought my first O-scope for this project.

I suspect you’re right a 75w target is not enough. Perhaps even the number of LED’s needed to get to the right level of light are impractical. So it could be I’m going about this wrong. Perhaps I should look into what makes a “General Radio GR GenRad 1538-A Strobotac” tick. (I’m thinking aloud)

I haven’t looked spec sheets for all of the different chips that have been suggested or studied everyone’s response. I’ve been tied up at work. I’m going it over this weekend so I suspect I’ll have a few questions.

Thanks for all the input.




Sent from my iPhone using Tapatalk

[Marco]

I suspect than LEDs are 10-100x more efficient than sparkgaps though, so the difference might not be that large.

[StillTrying]

Perhaps I should look into what makes a “General Radio GR GenRad 1538-A Strobotac” tick. (I’m thinking aloud)

It's a smallish 800v tube, 0.5us to 5us, I've not found a matching data sheet.
Triggering is easy compared with getting the light level up!

[Wolfram]

I suspect than LEDs are 10-100x more efficient than sparkgaps though, so the difference might not be that large.

It's hard to find good sources for the luminous efficacy of an air spark, but it seems to be in the range of a lumen-second per joule. A non-overdriven LED will therefore be roughly hundred times more efficient, but 75 W is also about five orders of magnitude below the peak power of a normal air gap flash. The total light output for an LED driven at 75 W for 500 us will therefore be about a thousandth of that of an air-gap flash like the EG&G Microflash 549.

[Marco]

Looking around the web I'd guestimate a LED spotlight overdriven to 100W will do about 100K Candela. So to achieve the same intensity as the Microflash you'd need about 50kW of overdriven power.

[JAndrew]

Looking around the web I'd guestimate a LED spotlight overdriven to 100W will do about 100K Candela. So to achieve the same intensity as the Microflash you'd need about 50kW of overdriven power.

Hmmmm, my first impression is that that’s beginning to touch on the realm of  impractical.


Sent from my iPhone using Tapatalk

[dmendesf]

You need to use an Arduino timer to drive the pulse (probably in pwm mode, but I'm not sure... I know pics better than avr). This  way you will have very fine control of the pulse width (with steps of 1/f, so about a hundred ns).

Enviado de meu moto g(6) plus usando o Tapatalk

[NiHaoMike]

Some modern xenon flashes use IGBTs to allow switching off the tube at an arbitrary time. 500ns is definitely on the fast side for IGBTs, but maybe it can be done with MOSFETs? Apparently, using a higher voltage but lower capacitance also helps for shortening the duration of xenon flashes.

Where LEDs excel is for video lighting in order to reduce motion blur, where the operating frequency is much too high for xenon. Unfortunately, few consumer/prosumer cameras have a sync output, but maybe it could be possible to use the vsync from the HDMI output, then use an adjustable delay circuit to sync the LEDs to exactly the right time?

[JAndrew]

This is the unit we used to take the pictures.

http://prismscience.com/spot.php

Taken Directly from the specification page.

SPOT: Specifications
Housed in a 6.7" H 6.7 H 8" enclosure (Fig 1), and weighing in at 10 lbs, SPOT is a compact submicrosecond flash system operating from 120 VAC power. Three trigger inputs are included:
· Manual — push button
· Switch closure — �" phone jack
· Pulse (e.g. TTL) — BNC
The stored energy of approximately 8 J is released in a short flash of less than 500 ns. Fig. 2 illustrates FWHM and FW1/3M times for full visible spectrum and for 520 nm line filter. The light collected by a 6" fresnel lens forms a relatively uniform "spot, " Fig. 3. The peak intensity of 69 million lumens delivers a pulse having 28.5 beam candlepower-seconds. At 40" from SPOT, the beam diameter is nominally 20" to 24"; however, beam spreads from 14" to 40" can selected during manufacture. A typical camera setting is f/5.6 to f/8 at 200 ASA for objects illuminated from 3 feet.

The design can be adapted to create different output characteristics for specific applications, such as a 2.7 J pulse with 200 ns FWHM for extremely fast motion, or point sources for shadowgraphs or schlieren images.

I'm referencing this as a comparison for what I'd like to try an emulate.  The one point I see that I hadnt looked at before was the lumens 69Mil. Trying to drive that sort of output from LED's even if it is for a brief period doesn't seem feasible.