Simple way to decrease THD+N of opamp, what can go wrong?

[Kokoriantz]

Beside composite opamp, I need an error corrector to keep the signal entering directly in the opamp and get an error corrector do the cleaning.
Such method does exist in servo mechanics, so why not in opamps.
The principal is an inverter opamp. The gain formula as we know is valid only if there is 0v at the negative input and no current through it but there should be Verr juice across the inputs to be amplified and produce the output.
The idea in order to decrease the Verr but keep the same differential juice, is to displace a portion of the Verr to the positive input by inserting an inverting amplifier.
This amplifier works at very low level in millivolts. With 80db opamp gain at 1khz, the Verr is 1mv for 10v output. This why the amp required is rather very low noise than low distortion.
I simulated one using 2 paralleled 2sk170 with a gain of 36db where the main noise generator is the drain resistor, it looks to function.
Any suggestions?

[Terry Bites]

You cannot gild a turd my friend.
Buy a decent audio opamp.

[Kokoriantz]

This can be applied on power opamps too as LM3886, TDA7293.

[magic]

Interesting idea, but I wonder if it wouldn't end up being a proverbial power oscillator. Not sure what happens if the "correction amplifier" ends up having 90° phase within the bandwidth of the main amplifier's loop.

It should be relatively easy to try this.

[Kokoriantz]

R8, C5 limit the gain on high frequencies.
The Verr is reduced by G+1 of the added amp so is the THD reduction but the noise is less reduced. In total, 26db reduction for THD +N is possible.

[David Hess]

Either the linearity of the amplifier can be improved or the open loop gain can be increased allowing more negative feedback.  The attached application note discusses active feedback.

[Kokoriantz]

Thank you for the document.
The noise issue is not considered, THD neither.
The problem with decreasing the open loop gain it requires adjustment for each opamp as the loop gain has wide variation. It is also the case with Yewen composite.

[magic]

Noise should be as expected, reduced by the gain of the added input stage RMS summed with e_n of the input stage and Johnson noise of resistors, particularly the 10kΩ gain resistor which gives 13nV/rtHz.

Distortion should also go down.

I would worry more about stability. The original feedback connection serves as a feedfoward of sorts around your new stage, but for it to reliably drive the opamp, there must be no significant signal on IN+ at high frequencies. Or you would need to ensure very low phase delay in your circuit at high frequencies, so practically flat frequency response.

Another thing to watch out for is feedfoward through the Miller capacitor C5. If its conductance at some frequency of interest is greater than transconductance of the JFET, then drain signal (which you apply to IN+ of the IC) will be in phase with the input signal (taken from IN- of the IC) and will end up cancelling it.

One trick used in old opamps is capacitive feedforward around a noninverting input stage, in which case the feedforward has expected phase and contributes positively to stability. Second stage (the IC in your example) must be inverting, of course. See LM318 or LM301 datasheets. In those cases DC goes around through the input stage, but maybe you could AC couple the input stage and run a resistor from the feedback network straight to the chip (IN-).

[Kokoriantz]

I simulated the stability with lm3886 model. I reduced the lead C3 capacitor so that I get overshoot spike and adjusted the C5. It looks that the total loop with both inputs, the C5 is degrading the stability, without is better.