Now I've got the basic idea down, time to scale it up. Bipolars aren't easy to get in this size -- say, 400V 30A -- so I'm going to try MOSFETs. (Yeah, I know, a few HOT's like 2SD1887 and its cousins, paralleled, would make a skookum BJT for this. But I had already ordered 10 x MJE18008 from Digi-Key before I realized I could find those cheap from replacement suppliers..)
So anyways, here's the spread: 100V power transformer (two MOT primary windings, one is smaller apparently, hence the non 120V output), FWB and cap bank, as well as two strings each of 10 x 0.1 and 10 x 0.47μF capacitors, one set for power supply bypass and the other for resonating directly on the coil. And of course, the 10 x STW11NB80's (the other six are in the baggie) which I ordered from Mouser.
Since MOSFETs switch so nicely, it would be a mistake to try a negative resistance situation like a self-excited oscillator with their fragile gates. So I tried shooting for class D instead, using my signal generator to drive them. Two in parallel for starters.
BIIIIG mistake. I hadn't realized that MOSFETs don't much like fighting against 6 microfarads of drain load and the switching harmonics essentially exploded them. Instantaneously, what happened is: signal, gate rises. MOSFET turns on. Drain current goes through the roof. Drain voltage starts to move. Slowly. Drain current still high. Voltage slightly lower. Can't....hold....much....longer.....*pow*. We're talking instantaneous dissipation of around 500W, or more. But...the part that really pisses me off is, they felt like failing by shorting exactly three leads together. Which means I got +100V on the (unisolated) input of my signal generator. Shit. To add insult to injury, I didn't notice what was making that smell until the damaged parts had smouldered away a bit inside the generator, fully preventing any hope of replacing the toasted resistors and transistors.
So screwed out of a signal source, I'll have to go back to self-excited. Or a breadboarded oscillator, like uh, a 555 or something stupid. Not very PWM enabled, so it's back to the bipolars. (You can do self-excited class C with BJT's because collector current depends on base current, and the drive winding is easier constant current due to the weak coupling. It also means power is consumed, which makes the drive winding feel useful.) Well, I've got these MJE18008's and the power supply, but no heatsink for them. Guess I'll have to cast one. Yeah...like that'll happen! Eventually I scratched up 6 x 2SC2625s, which are popular in AT size computer supplies. Then, I have absolutely no idea how, I managed to squentially blow each and every one, in order. I'm guessing parasitics, but we'll never know. I had them paralleled with 0.15 ohm Re's, so it can't be that bad. That never got any heat output, though I got oscillations for about a second before the next transistor failed shorted.
Enter the NON resonant system. I finally decided that I'll have to bite the bullet and eat reactive current through the switching devices and power supply. On the other hand, no capacitive reactance means the system can respond in nanoseconds (50 or so), practically eliminating commutation loss. But you can't just feed PWM to a common half bridge circuit such as this:
Because the FETs take time to turn off, but nearly none to turn on, you will get feedthrough currents momentarily shorting the power supply, and costing dissipation to boot. Hence the dead time illustrated in the input waveform. (Not shown on the above: gate slow-down resistors, possibly gate-clamping zeners and the intrinsic diodes in the MOSFETs.) So I need to generate that somehow. It just so happens I have an SG3524 on hand, which used to be a popular switchmode PWM generator chip. With a little extra help, it's basically perfect for the application. Here's the latest breadboarding exercise:
The chip is wired almost like the test circuit, except the current-limiting amp (pins 4 and 5) is disconnected. The open collector outputs, pins 12 and 13, pull current from a pair of PNP's which then control the complementary follower. (Yes, I know if I grounded the chip to the negative rail instead, I could remove two transistors.) The G.P. transistors need only to handle the voltage and moderate current; I used 2SA970 and 2SC2362. The TIP31/32C follower is overkill, but for some reason it dissipates a lot of power so I even have little heatsinks on them. Go figure. The transformer is on a toroidial black powdered iron core, which seems to perform quite nicely. Coupling and inductance are very good; with the turns and voltage shown, saturation doesn't occur until 10kHz at 49% duty cycle.
The circuit appears to operate as it should, at least with a resistive load. An inductor just screws it up though, and I have no idea why. At the very least, it appears to saturate at +/-10V, meaning the transistors are somehow dropping 7V *in both directions* (turned ON, then reverse-biased as the intrinsic diode handles the flyback). That has me stump-diddly-umped...