VSSA with Diamond input and DC servo.
Optimisation of a Diamond mod of the VSSA

For the fun, let-s push this DVSSA to the limit.

Schematic:

Models are here:



At 35V, the amp is a 50W/8 Ohms unit. At 45V ~90W/8Ohms .

First, let set the Miller for flat bandwidth, (removing C11 & C12 ) and see what we can expect (before the output coil) :







Now, let-us optimize the diamond low pass filter (c11, c12) for no overshoot on little signal square waves(values on the schematics) The square waves at 100KHz now look like this:




Ok, let-see how it behave when clipping. Note there is no risk of stressing the VAS because it run out of current when it clips:




It gives-up this final bandwidth after the output coil:




And, now, the distortions:







Distortion measured @ 80% (1V input) of his max excursion (30W 8Ohms). It gives 0.00042% 1000Hz 50W
At +-45V, it gives 0.000433% @ 85W, 8Ohms.
Nb: It has been reported my models for the OPS were not accurate. With Cordell models and +-45V, the distortion, at 1V input, goes up to: 0.0006% Not a big deal.

Conclusion:


The Diamond topology is a nice way to get rid of the base emtiiter offset of the input stage in a typical CFA. (offset between base & emitter = signal input & feedback input)
It isolate the source, add more current gain while it don't add a stage (pole) in the feedback loop, preserving the stability and high bandwidth.
But it suffer from slight instabilitly. The very simple idea was to set the input filter between the two stages of the diamond. This simple filter brings the following advantages:
Dump the Diamond from any instability
Set an input filter, idependant from the source impedance, in order to protect the amp from excessive slew-rates that it should be unable to follow (TIM).
As a consequence, remove any risk of overshoot with fast little signals.
Of course, you are free to add, right at the input, an other low pass filter to remove further RFI and have a better speed margin (find the good value with carefull listening).

Contrary as some writers pretend, and once again in accordance with Bob Cordell, i don't understand why some would like to continue to use BJTs for output devices,except for economical reasons.

Lateral mosfets don't need buffers to drive them (here the result with 5mA in a single transistor VAS-driver).
Lateral mosfets are damn fast, despite their gate capacitances, and able to bring very high slew rates.
Laterals are secure, because they are immune to secondary breakdown.
Laterals have a flat or negative temperature coefficient of current as a function of gate voltage. So they don't need any compensation stage, with the stability evils these compensation stages can bring.
Lateral love to be run hot (at 100 to 150mA) and this is good for getting a large Class A behavior at listening levels.
The simplicity they allow in the design is a good way for reproducing audio little details. Simple is beautiful.

They have two disadvantages:
First, they need more Gate/Source voltage, so, you loose some 5V on the power rails. It can bring advantage as they will never saturate, and you can feed them with a separate highter voltage if you want max power.
Second, they have lower transconductance than BJTs. This lead to higher output impedance (GNFB take care of this) and higher crossover distortion (Higher quiescent current take care of this).
This simulation want to demonstrate all this, plus the benefit of damped Diamond input stage with CFA, as well as some advantages of CFA topology.
Slew-rate >1000V/Ás, extended flat bandwidth (-3dB at 0.1Hz & 800KHz), distortion < 4-6ppm with only 3 stages in the FB loop.

That's all folks ;-)