This scheme is not as fool-proof as the transformer scheme.  If there were, for some reason, a mismatch in switch timing (or if a bridge leg fails to switch) then the capacitor impedance to the very fast switching voltage is pretty much zero and you will see "shoot through" currents between bridge legs.  What i mean is, imagine 1 bridge output switched hi, but the other "parallel" one was still low, you are now applying 2 different voltages to the same node in your circuit, and if the impedance between those voltages is small (remember, capacitor impedance approaches 0 at very high freq) then currents are big.

This does concern me for the coil I am building - I have done some simulations and the results aren't pretty if one half-bridge fails for some reason.
I plan to try and do some sort of sensing to shut everything off if there is a fault like this (potentially by looking at current or voltage difference between half-bridges - I have a separate CTs for each so have a few options for how I do it).
I also am looking at how to limit fault current from any failure so that I'm not dumping several kJ from the main storage caps into a failed IGBT - am currently looking at a simple DC rated fuse, or potentially using a low-VCE(sat)/Rds(on) IGBT/MOSFET as a fast switch for reacting to a fault situation (in this case with more flexibility than a simple fuse).
Steve (and others) - any suggestions/ideas on either of these issues would be welcome!

The other question I have for Steve is regarding the impedance of his transformer coupled coil - I'm interested in what the primary tank values are and what the transformer ratio is to see what the tank circuit impedance looks like to each half-bridge (and to compare it to what I have, which has no transformer but a relatively low tank impedance for a QCW coil).

A quick check looks OK, but haven't got the time now to go through thoroughly sorry. UVLO pin needs to be moved to a P12_x location I believe, but didn't see anything else wrong with the PSoC pinout, but as your said it should be test-compiled with the IDE to be sure.

The 2877 CTs have a bandwidth of something like 200MHz, so will not be the limiting factor. I was proposing to use a 35MHz CT for the base current measurement and a 2MHz CT for primary current (determined by what i have available - 2x2877, a 4100 and a 7800). I would be surprised if the CTs filter anything of interest out with a lower -3dB of 300Hz. I assume that you agree the best way to use 2 extra channels is to measure current?

Sorry to hear it blew up, but good to hear the 1/2 second was OK! I would recommend trying polypropylene capacitors rather than polyester next time - polypropylene is much better for this use.

I still think it will have close to zero effect - the black ring will be <1uH which is tiny in comparison to the reactance of the capacitors at TC frequencies. For example at 500kHz (a reasonable guess, if anything on the high side which makes the lead wire effect bigger:

14*0.1uF in series = 7nF, reactance at 500kHz = ~45 ohms per string of caps

With an estimate of 1uH for the black wire(generously large - it's likely a quarter of this or less from the geometry) gives reactance of ~3 ohms @ 500kHz, or less than 10% of the capacitor reactance.

While there will be some effect from different connection points, I don't think it will be enough to change much at all - certainly won't cause enough difference in current to have an effect on current sharing/heating.

For my 160mm (~6") coil I used PVC pipe end caps with M12 (~1/2") nylon bolts. I obtained them from a components distributor who specialised in plastic hardware, and were as big as I could buy without paying crazy money. Wish I'd bought a few more for future projects!

So I got the coil running outdoors, unfortunately with wind and in the middle of the day so no worthwhile video, but I have attached a pic of the setup taken from the "control room" (upstairs window where the variac, scope and PC were):


Control was via serial over the orange fibre optic lead, and I was able to look at bridge output current/voltage using a pearson CT and a couple of differential probes. The heater that can be seen is acting as ballast for my poor 2A Variac so it won't blow it's brushes to bits trying to charge the caps after each burst

The test was a success in that nothing went pop at 360VDC on the bus (variac cranked up to 11 on 240VAC supply), but the coil didn't behave that well otherwise. The spikes seen when hard switching were rather extreme, at up to ~750V (that may be a little high - I know that the probes used will ring themselves due to parasitics of the leads they use, but having checked with other standard 100x probes, I know the spikes are still there and large). This seems to cause the controller to lose ZCS synch, and causes waveforms like the following throughout most of the burst:


I'll be looking at the code to see if it can be made more resilient to noise, but to really fix things I think I'll need to get a revised version of the coil running with better layout etc (and higher current capability with 4x the number of IGBTS - was hitting my 225A limit a lot!). It has been really useful building up part of the bridge as a small scale prototype though - I've learnt a lot about what I should do for the final coil and will probably re-use some of the spare parts to make a smaller coil in the future.

Anything that works over a <5MBps fibre link should be fine. Not sure how you'd do CAN, but the RS485/RS232 based protocols will work fine with fibre as the physical layer rather than differential twised pair copper, as they're basically asynchronous serial at heart.
Absent something more standardised I had actually thought of re-purposing a RS485 protocol we use for CCTV control at work (mainly because there's well-debugged code that I could re-use, though not release), but using something standard and open-source would be best.

I'm just putting together a Mouser order for UD3 parts, and the 80MHz speed grade part (CY8C5888AXI-LP096) is barely more than the 67MHz part, so I'll build a board up with one of those and have a go at increasing the clock frequency (work on the clock speed will likely be delayed by a long vacation though).

Lastly, I'll probably have a go at running my coil outside at full mains voltage in the next few days - will try and get video of the success/destruction.

This is exactly what I discussed with Mads on IRC - it's definitely possible to lay out a board using a PSoC dev board + headers as a "through hole" part, and the dev board is _cheaper_ than the chip by itself. I'm probably not going to do such a layout myself (gonna stick to SMD), but am happy to help someone who is. BTW I disagree that soldering a QFP counts as "exotic", but the cheaper argument remains, and the dev board also comes with a USB connector inbuilt and even a whole extra PSoC chip (the programmer chip could be re-used as another mini dev board if you don't need many pins!)

The only reason I haven't posted more detail is because my setup is such a hack, and while all pins are routed on the PSoC, many are un-used. I just have the gate drive pins (going to another PCB with the actual drive circuitry on it - the green board in the top of the attached pic), UVLO shorted to +5V, CT current-sense and ZCS pins used. I've attached a close-up of the breadboard setup (note that while there is a crystal on there I couldn't find the right load capacitors in my parts box so I'm using the internal oscillator, which is really good enough anyway).

The biggest thing to watch for is to make sure that all pins can be internally routed to the required places on the PSoC chip - there are some restrictions as to what pins can go where on the chip, and limited routing resources (especially for analogue), so I had to shuffle some pins around the chip until the PSoC creator software could find a way to route everything. The other thing that should be done (but doesn't have to) is to adjust the speed that the PLL is running at - the PSoC dev board has a faster version of the chip than the UD3 design, and can run up to 80MHz rather than 67MHz (actually set to 64).

Yeah I was wondering if it was pulsed DC or not - the pulsing action probably helps a lot with stopping the gap from just pulling a big arc directly from the power supply. The other way to do it is with DC resonant charging (see http://www.richieburnett.co.uk/dcreschg.html#resonant ), but that requires the extra choke and diodes.