VSSA with Diamond input and DC servo.
Proof of concept

The Vssa power amplifier by Andrej Lackner is a fantastic piece of audio design. With a rare combination of clever topology choices, extreem simplicity and fine tuning witch are both good for detailled reproduction. And it is highly regarded by listeners, probably thanks to the CFA topology and minimal components in the signal path.
But it suffers from a main problem: The two big electronic caps in the feedback path. They are expensive, they bring coloration to the sound and, so, need to be carefully chosen, they increase distortion at low frequencies due to the phase shift they introduce in the feedback signal before comparison, even at 100Hz. And they need burning time to give their best while they will begin to wear immediatly after ;-).

Let's be rude, and compare 40Hz distortions between original VSSA and Diamond input modified one:

Diamond input

So, let see if we can improve-it, without adding any pole in the loop.
Schematics (quick'n dirty diamond version, not optimized) are here:

The idea is to add a diamond stage in the input, in order to can have both the signal and the feedback inputs both at 0V DC.
The base/emitter offset of the NPNs compensate the base emitter drop of the PNPs on both side of the symmetrical input stages.
It avoid the double path feedback networks, witch add distortion if not perfectly matched.
And they add extra current gain and isolation.
It has been reported by L.C. that diamond can bring some kind of "nervous" charateristic to the sound. This is because it can be near instability. We had chosen to dump this input with a RC filter between the two stages and that's cured the problem. Note the the first stage of the diamond, despite it is in the DC arrangement of the amp, is not in the feedback loop, thus don't ad any pole. On the contrary, with the added current gain it provides, it increases both bandwidth and slew rate, and reduce distortion. As-it, the amp is DC coupled.
let see the results:

FFT is nice too;

Well, it is talkative enough. Slew rate is two time faster, square waves edges are perfectly linear, distortion is two time better at 10K (mainly less high order harmonics) and 5 time better at 100Hz.
And this (more feedback at low frequencies) is a promise for better damping factor where it matters. All this just by replacing two expensive electrolytic caps by one more resistance, one more little film cap and two cheap litle transistors.

DC Servo

But we are now facing a problem. The amp is now full DC coupled with full gain at DC. So offset can be a problem.
Solution is simple and not expensive: Let-us add a DC servo. It is not expensive (can use some TL072), we will get rid of the fragile offset adjustment pots as well as time consuming & painful tuning. Only will stay a pot to adjust the quiescent current in the output stage witch can be replaced by a resistance, once the value is fixed.
If the servo is made with SMD, it will not take more space on the board than the two removed caps, i presume. It appears too that distortion is yet better with the servo, due to the total absence of offset.
(Notice high frequencies bandwidth of the Diamond version had been adjusted now to feat the original one, adjusting the low pass filter between the two diamond stages, and protecting further the amp against EMI and TIM).

Offset in non measurable in the simulation, and will even cure any from preamplifier when DC coupled. (pV).
Not to forget it helps to set the imput impedance highter (47K ?) with no impact on the offset.
Hoping it will inpire some of you (including Andrej for his future designs;-)


[Update]Optimized to fit exactly the VSSA currents in each stage, the Diamond simul, now, and for 80% (-2dB) of his max power, a distortion of;
0.0005% at 100Hz (servo included)
0.0005% at 1000Hz
0.0009% at 10KHz (0.0006% of first 4 harmonics)!!!

Article is Here