GZPA Class D 4000W with IRFP264

[Pafi]

In fact, sad truth is that you may manage to make it work with IRFB4229 by further increasing gate resistors to 150 or 220 to reduce di/dt to the levels that the PCB can tolerate without EMI becoming self-destructive

If this is really the root of the problem, then IMHO it would be better to improve PCB by adding some wires and capacitors. (Given the much better FETs the diode cross conduction problem has been eased, so lower stray inductance won't cause too high current peaks.)

[Eva]

Originally Posted by Pafi

If this is really the root of the problem, then IMHO it would be better to improve PCB by adding some wires and capacitors. (Given the much better FETs the diode cross conduction problem has been eased, so lower stray inductance won't cause too high current peaks.)

Adding capacitors results in ringing, then you have to add snubbers (and fit everything), then something else is no longer up to the job, then something else has to be improved too, and the loop goes on until you get really fed up.

I think the original poster just wanted to make it work again, clumsy as it was when it left the factory.

[Pafi]

However, have you ever repaired one of these? They are intended for LF output only, >250Hz, switching frequency is usually in the 50khz range, and the discrete gate drivers they use are ugly.

I haven't met with these, but a switching event lasts for some microseconds sounds to be a nightmare for me. And with the original MOSFETs it must have been much worse. With 4*50 nC miller charge the switching transient must have been more then 10 us long. This would be almost fully linear operation, not ClassD!

[Pafi]

Originally Posted by Eva

Adding capacitors results in ringing, then you have to add snubbers (and fit everything), then something else is no longer up to the job, then something else has to be improved too, and the loop goes on until you get really fed up.

Why would a supply decoupling capacitor made it worse?

I think the original poster just wanted to make it work again, clumsy as it was when it left the factory.

Yes, but I suppose it was ClassD! I try to figure out what could the charging resistance could really be. If it is 100 ohms per gate, then OK, it is usable, but more then 220 ohms for all of the paralelled devices is unreasonable.

[B.I.G]

as mgm2000ro said the clamps that hold the fets to the heatsink might be the problem ... i hate these things .. in the next test i will use an external +/-90V power source and external heatsink for the irfp4229 and see what happens

i can not afford blowing another 20xIRF3205 it isn`t cost effective ...

Eva you are right i just want to make it work but if it can be improved ... i would like that because it is my personal amplifier ...

[mgm2000ro]

what can change to improve this schematic?

http://img854.imageshack.us/img854/1314/biguv.jpg

[B.I.G]

thanks mgm2000ro for the schematic ... so what can be done to optimize this circuit ?

should i use IRFP4227 instead ? but +/-91 is close to 200V ...

[darkfenriz]

Turn-on current is the range of 80 to 100mA, while total gate charge is around 500nC. Simple calculation gives a total turn-on time of around 5-7us for around 20us period.
Ergo: the wavaforms look perfectly like the 'designer' wanted them to.
Good news is mosfets are pretty forgiving here, the 99% of transition takes probably more like 1.5 to 2us, the rest is sub-threshold charge and post-miller charge. The first makes no harm at all, the latter results in just a bit higher effective Rdson.

[Eva]

Originally Posted by Pafi

I haven't met with these, but a switching event lasts for some microseconds sounds to be a nightmare for me. And with the original MOSFETs it must have been much worse. With 4*50 nC miller charge the switching transient must have been more then 10 us long. This would be almost fully linear operation, not ClassD!

Well, you get the idea now, but switching events don't last that long, a few hundred nanoseconds would be more accurate. Yes, charging the gate to the full drive voltage can take a few microseconds, but IRFP4229 is already fully on at 6.5V and IRFP264 is similar, so it's not that bad.

Consider that the PCB is good only up to 50A/us per MOSFET or so, so building up the current has to take some time (achieved just with 100 ohm gate resistors and relying on finite transconductance). Not much to do with our >500A/us switching.

[Andrew Eckhardt]

edit:
With R192=220 ohms the gate waveform is understandable, but what kind of idiot place there such a high resistor there?[/QUOTE]

Cheap cross conduction limiting? Though with that much resistance there's really no need to double the NPN drivers. Is that a verified resistance value?

You might be able to get away with things at 55kHz that just wont go at 550kHz, if you're used to looking at full bandwidth switching amps.