and now for something completely different part 3

[JoaMat]

Perhaps you are right, but that’s nothing I can verify.

It’s correct that pos input is dancing. Now I think that the neg input is a perfect dancing partner which results in a working servo. With the servo the amplifier behaves similar to the ones without servo except that the bias doesn’t drift.

And cross talk? Not a problem - I’m married and used to cross talk.

[jokerman777]

my bad could've just read the previous page more carefully... I did a quick sim on Joamat's "new bias servo", 5Vpp on one side into 32ohm load, ofc not the same amp circuit, not opa445, and not assuming sim == real life behavior, but probably enough to convince me it's not gonna be free lunch. Below is before and after (also tried 300ohm load and difference is much smaller though).

Now there is also KG's optocoupler version but between adding another 8 x opa445 and more and just doing thermal coupling right in the first place, well... 😃

 

7 hours ago, kevin gilmore said:

i looked at the ltspice file for a bit.  best guess is that there is something wrong with the jfet model that causes issues. replace the fets with bipolar and see if it works right.

the reason all the darlington transistors are on the heatsink is so they can thermally track. effects thd.

another thing that really effects the thd is the power supply rails.  anything under +/-20v is too low and causes much higher distortion. +/-30v is the sweet spot.

and once you do that, you end up with an amplifier that is a lot more than 5 watts.

in your second schematic, you have taken out the low impedance load on the output of the current mirrors. lots more voltage gain resulting in required high amounts of feedback.  definitely not the original design goals.

if you want lots of open loop gain, probably best to pick a different design.

as far as the bias servo, would have to see thd with and without it.

And many thanks for the advice and detailed explanation! Yeah I'm aware of the generous amount of feedback used here, much like the original CFA2 right? where CFA3 is changed to open loop design with optional feedback config.

I didn't "choose" to go with this way on purpose and haven't built/listened to enough stuff to form a particular preference on the camps between no feedback/some feedback/a lot of feedback (which kinda has a bad reputation). I'm pretty much just randomly exploring options now, for possibly a smaller scale project than CFA3 for my need. Might still jump back on doing the original if I got enough messing with spice