I recently read on the Dreamingrobots Forum (http://www.dreamingrobots.com/forums/viewtopic.php?f=8&t=2237) about a very fast LED-based flash gun called the Vela 1 (http://www.vela.io/). This flash gun is amazingly speedy, delivering significant amounts of light within 500ns. That’s incredibly quick even by comparison to xenon flash guns that when run at low power, can generate pulses with durations in the region of 5 – 30us. Even spark-based flash guns can’t reach the speeds of the Vela 1. Also, the flash from an LED follows the pulse used to drive it with very low lag, and none of the afterglow associated with a xenon flash tube. Essentially, the light output is the same shape as the pulse used to generate it. It is clear that the Vela 1 is set to transform very high-speed photography.
The basic idea of an LED-based flash gun appears fairly simple; high power LEDs are driven for very short periods at much greater powers than they can tolerate when driven continuously. Using very short pulses means that the chip doesn’t over-heat and the wires leading to it do not fuse. Naturally, being of a curious nature, I wondered what I could achieve using some spares bin components. I thought I would blog my experiments in the hope that others would try and improve upon them, and that we might thus develop a DIY LED flash. In particular, I want to make something that is fast enough for the sort of macro photography I do, and that, by means of having some ancillary optics, puts the light in a small area. Here I describe the beginnings of the development of a flash using small 3W (3.5V) white LEDs. The circuit is designed to be readily modified to experiment with 10W COB LEDs that are normally driven at 10-35V. My reasoning was to start small with cheap parts, and if it worked, then order the more expensive larger LEDs and if necessary, higher voltage, lower ESR, capacitors.
My feeling was that rather like with the high-speed electromagnetic shutter described in an earlier post, the best way to “overdrive” an LED would be to use a MOSFET switch and a capacitor bank. Research revealed a paper written in 2010 that described the pulsed operation of high power LEDs for imaging flow velocimetry (Willert, C., Stasicki, B., Klinner, J. and Moessner, S. In Measurement Science Technology 21, 075402 (11pp)). These authors had succeeded in driving a variety of 3 to 77W LEDs at between 30 and 220A, many times their continuous current ratings of between 1.5 and 36A, without destroying them, by using current pulses in the 1 to 20us range. The circuit they used to do this looked rather like I had imagined, a MOSFET switch triggered by a TTL pulse that allows the current from a low ESR capacitor bank to flow through the LED thus producing a brief, bright, flash. They describe some interesting effects using coloured LEDs where with very high current pulses the LED’s normal emission spectrum is shifted towards the blue end of the spectrum.
Clearly, here was the starting point for some experiments. I wrote a very simple programme to generate microsecond pulses using an Arduino and constructed a circuit very like that shown by the authors of the paper cited above. I found that an Arduino isn’t capable of generating pulses much less than a few us without coding in assembler. Assembler scares me so I have stuck for now to investigating pulses in the range 5 to 500us. I had to modify the circuit to work with the components I had – I used a nice MOSFET that I had in the spares box, an IRF1407. This will carry the same huge currents that the IRFB3206G used by Willert et al., it is very robust, fast, and has a very low on resistance. Also, Willert used a MOSFET driver, the UCC3722P, that I couldn’t get hold of, so instead I used a TC4420. This will take a 5V pulse from the Arduino and turn it into a pulse of up to its supply voltage – I used a 12V supply for the MOSFET driver, plenty enough to gate the MOSFET. I had a pack of 3W white star LEDs and it was these that I decided to begin to systematically destroy!
Here is my slightly modified version of the circuit from Willert et al., 2010. In my case the capacitor bank consists of 3 x 1000uF 35V low ESR capacitors in parallel with a 1uF polypropylene capacitor. The extra resistors are necessary to prevent the gate of the MOSFET and the driver from floating. My first version of this circuit was built up on a bread board – probably not a good starting point had I wanted to start with high speed pulses, but good enough to get the circuit up and running with millisecond pulses and modest overdrive currents. Note that the Willert et alia’s circuit has provision for monitoring the overdrive current by virtue of placing 20 milliohm resistor between the MOSFET source and ground. While it is possible to buy very low resistance resistors, the simplest and cheapest thing to do is to make one. I did this by looking up the resistance of the 27AWG enamelled copper wire (0.169 ohms per metre) I happened to have and then cutting the right length for 0.02 ohms, folding it in the middle, twisting the two lengths together, and then spiral wrapping them around the body of a resistor, scraping the enamel off the ends of the wire, and soldering them to the resistors leads. You can find a description of this process and its whys and wherefores here: http://www.cappels.org/dproj/DIY_Copper_Wire_Resitors/DIY_Copper_Wire_Resitors.html.
The bread-boarded circuit worked but it was clear that a PCB would be required to take things further. I produced a PCB and this too worked but now I could push the pulse duration down into the low us range. Below is a picture of the PCB hooked up for measuring the drive current and light flash as measured using a photdiode. Below that are details of the PCB – this is absolutely not a final design – the traces will need to be thickened and shortened for a higher voltage version which will probably use a single low ESR 100 or 200V capacitor. Capacitors described as ‘photo flash’ may turn out to be the best to use in the higher voltage versions – they are available on eBAY.
I started to investigate how far I could push the 3W daylight white LEDs in terms of the drive current. I noted that Willert et al. had been able to push LXHL-PD09 3W red LEDs to 20A using 1KHz 1us pulses. I opted for a lower frequency and a higher duration because I couldn’t get less than 5us out of my Arduino program. I was able to get to 10A without damaging the LEDs (see IV diagram). The light output is pretty damned bright but I doubt that a single 3W LED has much application. However, I do think that in principle, since it is possible with a 3W LED, it should be fairly straight forward to build a more powerful flash with 1/200,000th second duration. How easy it is to get down to say 0.5us remains to be seen.
I will add to this post when I have taken things a bit further. I am in the process of working out how to measure the light output in some kind of meaningful unit………
Update…the 3W white LEDs g0 west just north of 19.6A…the voltage across the 0.02 Ohm resistor started to have a peak at the front end and then the light output ceased.