Update: I've been doing some hard design work on the final version. I've finished my first draft (the largest schematic I've ever drawn on computer, 1310 x 1138...ack) and have moved on to the more useful board-oriented schematics with a few improvements, below. Barring obvious errors I may've totally glazed over, it should be in a working form. I'll be getting this reviewed by some other minds to make sure it's up to snuff, then test each seperate segment for operation before connecting everything together and running it to a few kilowatts, then finally full power. Hum, I hope someone has a 240V isolation transformer I can borrow.
Clarity is why I've broken this intimidating circuit into five interconnected schematics. Not only have I labeled the parts (something I don't do much anymore in schematics), but I also labeled them by hundreds to be able to refer to a specific component in a specific circuit segment with no ambiguity. All drawings copyright Tim Williams, 11-1-2005.
This schematic has three segments to it. First, all three relays are turned off and only T101, R110 and R111 have power. The resistors charge the high voltage supply capacitors before the main circuit switches on, reducing power-on surge. Since T102 is off, the circuit is off and there is no high voltage load. Q101, D101, D102, R101 and R102 constitute a constant current source of low current (approximately 190μA) which charges C102 with a constant slope of 1.9V/s. When it reaches 1/3 of the supply voltage (11.5V/3 ~= 3.8V; the time is thus about 2 seconds), the bottom comparator switches and RL1 and thus RL2, the main contactor, is engaged, ready to handle full load current. After another 2 seconds, the top comparator flips and RL3 passes power to T102, which supplies power to the low-level circuits by FWB2, IC102, IC103 and, for the floating high-side driver, a cute resonant inverter consisting of C112, C113, R112 (which supplies base current to start oscillation) and T103, plus a supply bypass of L103 and C111. Since this oscillator always runs to saturation when operating properly, the peak-to-peak voltage on T103 is always equal to twice the supply voltage. It should operate over a wide range thanks to Zetex's wonderful transistors with scary low saturation and massive hFE at high Ic (said ZTX651 is rated for β = 100 at Ic = 2A!). (T103's secondary should have slighty more turns on the secondary to account for losses. The base drive winding should be about a tenth of the primary, if even that; a few turns are usually sufficent for circuits like this.) After RL3 is engaged, the circuit turns on and significant current is drawn through FWB3, C103, L101, L102 and C105. C104 and C106 are provided, in addition to others at the inverter itself, for HF filtering, transient protection and peak current supply for the inverter during switching.
The oscillator is centered squarely around IC201 (yeah, so it's a third to the right on the schematic, so sue me!). The SG3524, and other similar switchmode chips -- MB3759, KA7500, TL494, and others in the SG series -- are free-running oscillators with pulse-width modulation and a push-pull output. Most come with uncommited output transistors, meaning you can ground the emitters and use open collectors, or run the collectors to +V and use the emitters to directly drive a small output transistor (or a large darlington or MOSFET), or whatever else your heart desires. Here I use the emitters for their low impedance characteristic, one to drive an isolation transformer for the high side and an equivalent circuit for the low side for timing reasons. (On the high side I have a current-limiting resistor, R207. You'll see why below.)
The left half of this schematic is dedicated to the overload latch, which clamps duty cycle control in event of a desaturation event (that is, IGBT collector voltage rising excessively while gate voltage is high) or other overload situation. (I will put a DC current limit on this circuit later, and it will add another SCR.) Though the SG3524 has a SHUTDOWN pin provided, I've grounded it because it is active-high; instead I have opted to shunt the control voltage externally (an action comparable to the internal function of the SHUTDOWN pin). SW201 and the RC components R208, R209, C205 constitute a turn-off network which gives a negative current pulse through the SCRs to turn them off, with D201 clamping any excess charge. It should reset the circuit within 20μs, fast enough that it can re-latch again if a fault condition is still encountered.
The oscillator is the heart of the beast; the control circuit is the brain (as such). This monitors the vitals: inverter current (read at the ground return of the AC-coupled output network), inverter voltage (the IGBT square wave output) and tank voltage, and provides feedback to keep in check whichever has priority. The two voltages are clipped by diodes D301-D304 and turned into happy square waves by IC301. The phase is detected by an XOR gate (composed of IC302), filtered by R308 and C303 and compared to a variable voltage to a maximum of +V/2, corresponding to Θ = 90°. Minus the phase detector, the other two properties -- voltage and current -- undergo similar comparisons. Note that all three comparators are actually op-amps wired for a gain of ~45. I did this to "soften" the response of the circuit. No, it won't be able to keep the voltage or current or phase exactly some value against varying conditions, but I don't want this thing bouncing like the Tacoma Narrows bridge as it tries to settle on some frequency! Oh, and speaking of frequency, since this varies a local oscillator's frequency in comparison to a tuned circuit (which has a constant resonance frequency with respect to the circuit), it is a PLL (Phase Locked Loop). At least...when you have Θ limiting it to a constant (locked) phase. Yeah.
These two very similar circuits drive the IGBTs. I could've gone with just a follower, but it needs biasing, not something easily passed through an AC-only transformer. (The goal is positive 10-15V "ON"-state gate voltage, negative 5-10V "OFF"-state.) I also can't go with just transistors, because I was suggested two things: a UVLO (Under-Voltage Lock Out) and a desat(-uration) detector. Starting with T201, the slightly-greater-than-supply-voltage signal passed from the oscillator is clamped by diodes D401 and D402, with current limited by R207 (on the oscillator schematic). This is divided to about 6.4V peak (assuming +V = 12V and -V = -6V) and compared to the zener reference voltage, 5V. Since 5 < 6.4, the comparator goes low, pulling Q401 on, which yanks Q404 and Q405 away from Q402 (which forms a current mirror with Q403), turning the IGBT "ON". If supply voltage drops below 14V total (i.e., +V - (-V); as shown, total is 18V), the comparator will never pull low.
Drive is made possible by a complementary pair of Zetex transistors, whose high speed and β allow a mere 10-20mA current source to turn off a pair of chunky International Rectifier IGBTs. Peak current through the driver should be around an ampere, for maybe half a microsecond total. This high rate of change will make a lot of trouble so I have indicated bypass capacitors from IGBT to supply. I will also tack one across the ZTX's collectors to be sure they stay nice and stiff.
Ground notes: see that the top and bottom circuits do NOT share the same ground! If you manage to forget and connect them anyway, you'll end up shorting the output. The high side ground is local to just the high side driver circuit. Another ground note: the low side IGBT emitter is already grounded (to main circuit ground, same as in above schematics), so the mark indicated on this drawing is redundant. I added it here just for emphasis, in case you forget that the emitter is grounded to the same ground shared by the +12V and -6V supplies.
A short description of the desat detectors before I move on. R407 (R507) couples the collector voltage to the other comparator half IC401b (IC501b), whose input is clamped within safe limits by diodes D404, D405 (D502, D503). When collector voltage is greater than the voltage set on the trimpot R408 (R508), the comparator switches high, unloading D406 (D504). Now, this happens every half cycle no biggie, so we need to logical-AND this with the gate voltage, hence D407 (D505). If, at any time, gate AND collector voltage happen to be high, Q406 is allowed to rise and current flows through the optoisolator. On the low side, a transistor and opto aren't necessary so it leads back with a single wire. In both cases, a TRUE output from the desat detector for more than 5-20μs will trip the respective SCR on the oscillator circuit, shutting it down without damage to the IGBT junction (hopefully!).
Simple enough: picking up where the drivers left off, the inverter handles the amps and volts of the device. R601-604 are recommended for some reason, parasitics I suppose. They limit peak current to about one ampere total, assuming a perfect risetime from the driver stages. Note the connection points for collector and emitter; with 50A per transistor switching off in 1μs or faster, the resistance and, more importantly, inductance of just a few inches can generate significant error voltages. To reduce this as much as possible, I am going to use flat copper to connect the transistors to bypass capacitors (a few of which are shown here), then to the power supply with stiff wire. I may also use snubbers and commutation capacitors at each transistor to improve efficiency (i.e., keeping Vc low while Ic switches off).
After the transistors, C603 blocks DC voltage, since I don't want to use a +/- (bipolar) supply for the circuit. (This choice simplifies measuring current in the circuit, but makes for an unhappy situation: not only do I have 120VAC from circuit to ground, I have 50A behind it! A good reason for plastic-handled screwdrivers, ladies and gentlemen.) L601 is a rather beefy piece of copper and ferrite, required to couple the fast-changing, efficient squarewave output of the inverter to the round sine wave of the resonant tank. Lmatch has to be rated about 10A at 250μH, on up to 50A or so at 50μH. The former isn't so hard to do (I've wound 18AWG bifilar on an old flyback transformer core for it, for testing), but I've yet to see a piece of ferrite large enough to do the 50-50 coil. From there, ah yes, the tank itself. Fill in the blank here: as long as the coil and cap don't melt, you can hook it up and give it some volts!
A great thanks to Terry Given, "The Phantom" and all the others on sci.electronics.design for helping me get things to this point.