class-D amp so far

Looks pretty good. Why are you using a coupling capacitor again ? I don't think you should need it ? At least from the primary side. Maybe if a rectified load is connected it might be necessary ? That's easy to simulate.

At my last company, we made 3 kW 60 Hz inverters with 20 kHz switching and large-ish 60 Hz transformers. Worked well as long as you keep the high frequency switching off the transformer and core.

C4 might help limit damage propagation if the mosfets die but a slow dc offset removal servo will cost less on the BOM.

I usually add didoes across fets in spice for simulation purposes even though in real life I'd likely rely on the body diode.

Are you going to add simulated source inductance and resistance like you did on the previous model - can that be done without feedback?

piglet

If it were my circuit and the user can load it anyway they want, I would take that current monitor output and stop the show when either polarity exceeds a predertimed value for overlod protection

Also, is the 48V being supplied by a battery ? IF so, I would definitely add overcurrent protection. Battery current is what kills things.

And if either battery supply or not, does the output voltage need to be regulated ? Are you really going to use the LTC4444 controllers or are you using a micro ? A micro can do all the logic with added half bridge drivers as well as output voltage regulation feedback using a differential op-amp ciruit into its A/D converter. Overcurrent can be done that way too using a pin from your current monitor and comparator or even A/D converter input if it's fast enough.

But then again, maybe this is all overkill. I like desaturation protection drivers.

The toroidal transformer saturates with just a little DC voltage on the primary. I don't know if I could trust a servo circuit to keep the DC offset low enough.

The load is usually a switchmode shunt regulator followed by a 3-phase bridge and a filter cap, the customer's FADEC.

Laminated transformers are sorta tolerant of a little DC. This toroid isn't. Caps with low ESR and high AC current handling are a major issue here.

Yeah, the big caps are a nuisance. Counting on a servo alone is tempting but tricky. Too much risk maybe.

I've seen mosfet substrate diodes do ugly things that weren't simulated. I may as well add some diodes; they don't cost much.

Yes. The current sensor feeds back into the FPGA, and that lets us program the effective source impedance that the user sees. That's compatible with making this switcher open-loop. We'll add a bit of real inductance too.

It's a 1000 watt MeanWell power supply. It current limits at some value that I need to measure. Wild guess 30 amps maybe.

The existing board is analog in, analog out, and this is a higher-power drop-in. The shunt feeds an isolated delta-sigma ADC that goes back into our FPGA on the main board, so we know the current.

I want these fets to never come out of saturation!

The existing code has too many protections, so customer abuse is shutting us down. One existing protection is too much current for too much time, which we'll keep but really tone down. We need to tolerate a huge overload for a few seconds maybe. We won't fry a transformer in a few seconds.

The shunt current is used to program the effective impedance, so we can control the short-circuit current.

Ahhh... So much more of a reason to use desaturation protection.

Right now, I like the Rohm FET driver I think. Easy to use and fairly cheap as far as drivers with desat protection go. And 4 amps I think it is. You would need 4 of them. BM6104FV-C

We adjust this limit to something higher than our highest allowed surge current so that it can also help with simple over-current. Also as the FET gets hotter and its RdsOn rises, it will be more sensetive. For low ESR electrolytics, we like either UCC or Jianghai

Jianghai used to make a lot of the Panasonic capcitors in China which were the best for low ESR. Unfortunately, there are tariffs for any of these good caps right now. We'll see if those go away but I kind of doubt it because even the democrats like all these taxes I'm pretty sure.

I suppose you already have a toiroid or transformer picked out. A lower permeability toroid would be more lenient on DC offset. A servo works too but current limit is good for double added protection.

You don't servo the DC voltage to zero, you servo the DC current!

Jeroen Belleman

Wouldn't that turn them off? What else could it do?

The advantage of the LTC4444 is the LT Spice model. That's why we pay extra for LTC parts.

A big enough fet would just stay on, and the big power supply could current limit.

Reasonably sized radial caps seem to top out around 4 amps or so. I could parallel four or five. We have lots of air flow, which would help. The current limit is probably thermal.

Getting the custom transformers was a nuisance, and we have a lot of them now.

Oh my ! No, the transformer is nothing to worry about burning up !

It's those FETs you have to worry about... Unless it's a really shitty xformer. I doubt that is possible with your stuff. You know what you're doing here.

Yep ! One begets the other !

But volt-seconds balance on the core is what matters.

This can also be a problem on push-pull supplies if the duty cycle isn't 50%

There are different and fun kind of ways to fix this or keep it centered where it belongs.

Myself, I would rather not have to use a series capacitor but it probably doesn't matter too much as long as that ESR is really low and doesn't get too reverse biased.

I could locally demodulate the output of the delta-sigma ADC that monitors the current shunt; that just takes a lowpass filter. But a few big caps take less thinking.

A junky DC servo, on the differential bridge voltage, could protect the caps. It would only need to be accurate to a volt maybe, which would be easy.

Yeah, that's what it would do. It can turn off cycle by cycle (HF cycles) if you want it to though. Might or might not be noticeable depending on how often it trips. This protection also has an isolated feedback so your FPGA or micro can handle things smartly. For example, you could count the number of trips and if too many happen in a certain amount of time, shut off or reduce amplitude.

And this is exaxtly why LTspice is free and a great tool.

Yes, that certainly works too as long as the supply side capacitors can't supply too much blow-up-energy in addition.

Yep. Totally understan that !

BTW, this would be bidirectional from output back to the power supply.... Are the loads all going to be resistive ?

I just tested one. It's about 4 deg C/watt on the bench in still air, with maybe a 30 minute time constant. That's measuring actual copper temperature. It could absorb all the power I have available, about a kilowatt, for 30 seconds or so.

I'll just use fets that can short out more current than the power supply has. That's easy.

If I come off the two outputs, after the LC filters, I can go into a diffamp and then an integrator. With 1% parts, I can probably keep the offset below a volt or so. With 0.1% resistors, it would be correspondingly better, a fraction of a volt. Still too much for the transformer.

I tested some candidate 10mF 6.3 volt caps. At 1 or 1.5 volts reverse bias, leakage current is low and declining, maybe 40 uA after a while. At 2 volts, it's 200 uA and increasing. I think increasing is bad.

Posting about things like, and getting peoples opinions, helps me think.

Kinda crazy, 5 or so giant caps to keep DC out of my transformer, and a servo circuit to protect the caps.

We're doing all that, and when the customer does a certain thing, we shut down. I want to apply a huge amount of brute power to get over his surge. Maybe 10 seconds would do.

The load is a shorting shunt regulator, namely 3 diodes and 3 mosfets that feed an output cap. A really dumb regulator PWMs the fets as needed to sort of regulate the DC. Fets off, it's a 6-diode bridge. Fets on, it's a short.

It's usually driven by a PM alternator, which doesn't mind being shorted.

Ya know.... You can also combine those two separate output inductors (L3, L6) onto one core and use less turns. Then you would also only have to use one, instead of two C's (C6, C8) across, going to the transformer since it's all isolated from there.