The LM1875 for best

[kokoriantz]

Too slow. I will return to AD8620 for all and redesign to make as good.

[kokoriantz]

ELIPTIC SINGLELMAD.JPG
I went back to the beginning, now that I understood what it should result. I was comparing the output, reduced to input level with directly with the input. I don't have intention to correct its high frequency behavior, so I pass the input through a low pass of same frequency or lower than the amplifier, and ask the amp to imitate it. To reach so, as I had to increase the gain, I increased the input by 10%, instead of decreasing as much on the output feedback, I decreased it 20% so that the error is not working at null but with slight signal so that it resembles the output.
500ns AD.JPG
The amp for now has offset problems due to AD826 and the erroneous simulated gain of AD8610. The frequency response is back to original, The initial transient is that of open loop, I am now satisfied.

[kokoriantz]

This is the single supply version. This ends the first round.
Now back to the beginning to check for the best of output stage.

[kokoriantz]

I review the output stage. I delay the input stage to match the inverter by 47ohm/68pF. I reduced the Q of the anti resonator and increased the resonant frequency. The feedback made including the current multiplier without inductance. Although I had 1.4mv offset, I balanced to few uV.
The result is very smooth transient see 20ns,200ns,500ns scale for 10khz 10v start.
20ns ellip.JPG200ns ellip.JPG500ns ellip.JPG
The square wave 10khz 10v
20us sq.JPG 1us sq.JPG
where it goes wrong is at saturation when the amp is biased 200ma but ok without.
sat.JPG satur with 200ma.JPG
The frequency response. It is possible to eliminate the notch, may be it is security to avoid oscillation that occur at high currents.

[kokoriantz]

I have forgotten the 0.22 ohm resistor on the multiplier. Now biased the saturation is ok. All other measurements are identical.

[kokoriantz]

This is the final of second round. The error signal now follows the output up to 1Mhz instead of 30khz on first trial.
error phase.jpg

I have now 22p compensation instead of 150p which had about 4 times distortion reduction at 10khz. The Inductor L3 can be eliminatef by adding a capacitor on R34. unfortunatly my simulator stops working even it is 0pF. To say that it takes several minutes to resolve such complicated circuit. I will switch to LTspice may be I get faster resolution.
I will start now the power supply .

[kokoriantz]

I am testing for clipping and start/stop plops. For saturation, ad my input voltage is high a pair of zener 8.7v diodes on C5, is sufficient to clip just before the amp at 8 ohms load. See at 10khz.
saturation.JPG
I can see the plop, The problem is that the supply voltages of the opamps +/-12v must be very fast following the 42v main. Not easy but possible.

[kokoriantz]

This is the output Zobel network for each side.


This is square wave 10khz response with only 4 ohms load.
without.JPG
And this, with 100nF along 4 ohms load.
with.JPG
I used the snubber after the LR because this how I understand the function of this network. The other way is to follow thumb rules. Most Japanese brands and McIntosh use this way.

[kokoriantz]

plop with.JPG
This what the simulator shows when I gradually decrease to zreo and then increase the supply. The signal first starts saturating then at 2.3V total supply strange things happen. Can someone confirm if this is the case in reality?

[kokoriantz]

This is regulated supply of 24v always centered in the single supply from34v to 45v. To avoid plop at start up and shutdown the 33v zener conducts at about 34v to conduct the supply switch transistor, a hysteresis resistor R11 provokes frank switching to avoid linearity and parasite.
The output impedances are bellow 0.5 ohm on simulator, but in real world I always got 0.2 ohms with BD139 at 100ma.
Vv.jpg +V.jpg

+V to -V -----------------------------+V To ground
psu imp -V.jpg

-V to ground