Though I'm definitely a fan of the Spellmans, I do some homebrew HV here. For the small stuff (10 watts or less) I like nothing better than the older, dumber CCFL inverters. With a fast diode or few and some small caps, they are ideal for things like detector HV. I have made these in closed and open loop designs, but really they are so "stiff" if done right, you can simply feedforward regulate the input to them to get a fixed output. To do that, you don't use the tiny series caps they intended for ballast, but something meatier, like a .01uf or larger. At the 50kHz they nominally run at, these tiny caps make good multipliers and good ripple filters. In a pinch, for something sensitive to ripple, I'll add a final RC section with maybe a .1uf or .47 uf final filter, and that gets the job done fine -- I keep my closed-loop adjustable one for tests of new things, but use the open loop ones in practice here -- I use an LM317 and a potentiometer to regulate the input to those, and to provide a bit of current limiting, it just works fine. 10kv is about the max with useful current output, but I've pushed a CW stack on one of these to 30kv taking heroic measures to cut corona losses. Much more often, I'll take a 12v input CCFL, run at about half that, with a volt doubler on it to get 1-3kv outputs. This is a compact and reasonably power efficient way to get that detector power. You'd probably do something to improve on that for battery operated things -- and you can get those nice little transformers as separate part numbers at DigiKey and elsewhere. (note to self, part numbers here)
This is my favorite one so far -- good linearity down to low input/output, and very stiff in/out impedance. The only trick is getting fast HV diodes for the output -- you cannot use normal diodes at these speeds, as their slow turnoff times kill them and the driver. As circuitry for CRT things goes by the wayside, it's only getting harder to find good HV parts -- get them while you can. The Digikey number is 285-1029 for that one.
Note to self -- diodes and the good caps need part numbers and pictures here. Schematic for the closed loop version?
//////////////////////////////////////////////////
For larger things, I decided to see just how simple a high power H bridge could be made, and the one shown here is about at that limit. A little too simple as it turns out, but with care these work really well. Spellman uses a more sophisticated driver, and a lot more sophisticated protection that for instance will allow more than one small arc before just shutting down in disgust than this, and the upshot is, this one is simple, but you may sometimes have to replace parts if you abuse it too hard. The magic chip is an IRS24353D and it works with "any FETs" if you adjust the gate drive series resistors to match the drive frequency and FET input capacity for the right off time (eg no shoot through, but not excessive off times). Using FETs, you find that the very low voltage high current ones are good, the medium voltage medium current ones are best, and the high voltage low current (15 amp still!) ones have a lot of on resistance and tend to need more heat-sinking. In other words, as volts go up, the FET on-resistance goes up quicker than is ideal for best efficiency, while at the bottom end, they have so much input capacity (gate plus miller capacity) they become hard to make go as fast as you might want. Switchers are not in general a thing where you can fix all problems via brute force techniques, you have to understand all the tradeoffs.
- Simple H bridge design
Of note in this picture are those transformers. The pretty ones are from a huge Glassman supply I stripped (the rest was mostly too weird to keep working). These are nice, and note the cool trick of using two primary windings to get the series inductance just so for good driving characteristics in series resonant mode. The other one is one I wound on the lathe, too small a series L, and due to the layered (vs universal pie) windings a very hard self resonance at 12 kHz with the greater turns ratio I used there to get a higher output voltage. That's too low a frequency for the size of capacitors I'd like to use in a CW multiplier -- stored energy in those is a danger to both the gear and the supply itself, so you want it low. I wasn't so aware of those tradeoffs as I am now, so this is the second batch of parts I got for stacks -- caps still far too large for safety.
- Parts with my cat girlfriend on guard
I had (and still do) troubles finding diodes both fast enough, and high enough voltage to do this, and also troubles getting enough arc free voltage out of my transformers to support the volts/stage I wanted, so I made the caps large to support a lot of stages without too much loss. I was able to find some fast 1.5kv @ 6 amp diodes in TO-220 so I designed around that -- and found that in a 24 stage thing I needed to heatsink the first few stages as current at stack bottom is pretty fierce. On the theory that info on "good efforts that don't work" is also valuable, this is what I tried that wasn't ideal. The driver boards grew some current sensing, and this shows two of them mounted on a thick Al rack panel as heatsink. One is tuned about 2khz for big XRT's, and the other at about 70khz for my big homebrew transformers. They share a big bulk rectifier/cap arrangement for main DC input, and a small 15v supply to run the driver chips. That part of things works pretty well here and is very efficient. The downside of the simple driver is fixed frequency and 50% duty cycle square waves -- the only way to control the voltage out is to control the DC input (variac and big transformer). If I do this again, I'll take a page from Spellman's book, and use something more like the LM3525 as the main drive chip. Takes more parts, but has built in current sensing and a lot of safety type stuff, a lot more controllable, and has variable PWM outputs. IR also sells standalone H bridge gate drivers after all, and they are basically the same thing as integrated into the "simple" H bridge driver. I note that in the ones I have, Spellman gets a lot fancier than that, and uses some fairly hefty pulse transformers for gate drives, and runs them hard enough to warm them up some. Whether that is design inertia (good 600v solid state bridge drivers are a sort of new thing) or for some reason I don't know -- I don't know.
I have since found some much better material for cores, and better ways to wind them. Since I'm going for volts/turn here but also efficiency, this is way overkill on core material for the power levels (a mere kilowatt) but it keeps the self resonance very high -- I can drive at 1/3 of that with square waves and not have excess current draw on the 3rd harmonic energy.
This runs at about 1 watt core loss -- for reasons I'll get into later, I should run that much-much hotter.
- 12 stage bigger stack, I used two of these in series at about 3kv/stage. See simple transformer.
When you arc a CW stack to ground (hey, it happens), it tries to blow the full, multiplied voltage into the transformer secondary. In this kludge I was running a 24 stage stack. Having the transformer core saturate can protect the drivers from insane overvoltage from this, but I didn't do that and found out the hard way why I should have. I suppose that a full wave centertapped transformer/stack would help with this effect/problem, and may be why you see them in some designs. In this simple driver, the output devices are always on (50% duty for each) but even that isn't enough to protect them all the time, and I can only imagine this being worse with PWM and lower duty (all drivers off sometimes).
Here is the thing as deployed here for awhile.
- Deployed homebrew supply
This was our main supply for a couple months, and put out some pretty good power in the 40kv range. As you can see in the picture, I made yet another driver board. This was efficient enough, even at kW levels, as not to need big heatsinks on the FETs. This ran with 0-30v or so on the main rails for the H bridge. But at this point, I needed higher voltage, was having the occasional diode failure due to overheating or my ballast arcing over, and realized I'd spent a few months on this - suddenly the Spellman looked like a great deal, that stack wasn't easy to tear down for repairs!
To sum up -- next time, if there is one, I will use much more volts/stage, and put together a nicer coil winder than my lathe in threading mode to get a better transformer secondary. I will use smaller capacitors to keep the stored energy lower, and I will run that transformer core a lot hotter and accept that loss to protect the drivers better. I will also use a smarter driver scheme with PWM and series resonant drive like the pros do. That's if I use the CW stack at all, right now my mind is looking at a "summer" design that would have lots of secondaries on the transformer, each one multiplied less, then all the DC hooked in series -- this would keep the self resonance up high where I like it, allow the use of much smaller caps, and so on. Easier to isolate DC than AC when you are sweating parasitic capacity as well.
Until that time, I have the Spellmans, and I did save the stack and half the transformers off a huge Glassman supply I scrounged, which is good for +125kv at multi kW levels with that simple driver using only two of the transformers (it was originally meant to run off 3 phase 440v input power -- a real hog, and just too large for my lab).
