Amplifier with nested Miller compensation

[Dave Zan]

..The latter is probed both with Tian and Middlebrook probes...

Happy holidays!

I would like to clarify some nomenclature.
The term "Middlebrook probe" is a bit overloaded.
Middelbrook wrote a paper on Return Ratio calculations and instrumentation in 1975.
He analysed and discussed several different ways to calculate and measure Return Ratio.
The V(b)/V(c) method was one that he showed was inaccurate, before he proposed a better method.
So to call the inaccurate version the Middlebrook probe is a poor choice of name.
It's not useful for your application anyway.
Even his better method is somewhat limited, comparable to the Tian probe, acceptable for "normal" amplifiers but perhaps not for extreme cases like yours.
He later improved the technique with the so-called GFT Middlebrook method.
He claims this is the final, corrected method.
I have checked his results with Bode's book and they appear to match.
So the Middlebrook GFT probe is what you need to be quite sure.
Frank Wiedmann has an excellent introduction >Here<.
This is what I referred to in my earlier thread.

Best wishes
David

The phase in your second Tian plot still looks ... "non-physical", to put it politely

 

[matze]

Hi Daid,

Happy holidays!

Thanks, and the same to you!

... So to call the inaccurate version the Middlebrook probe is a poor choice of name.

OK.

So the Middlebrook GFT probe is what you need to be quite sure.
Frank Wiedmann has an excellent introduction >Here<.
This is what I referred to in my earlier thread.

Currently, I really think that the Tian probe shows the right picture. Although a part of NFB bypasses the OPS via TMC resistor (below the "knee"), distortion is still further reduced, since the VAS drives into the output load, too. Have built a very simple simulation example that supports this reasoning.
But I will check the more abstract arguments and the GFT probe next year. My aim was to first solve the slew-rate problem.

In the GFT probe thread, Hurst's 1995 paper is mentioned, e.g. in this post. If I understand Waly and you in a previous post right, then the paper provides justification for the simplified loop stability analysis in post #83.


Kind regards,
Matthias

The phase in your second Tian plot still looks ... "non-physical", to put it politely

Mike Engelhardt and Frank Wiedmann are responsible.

 

[Dave Zan]

...Currently, I really think that the Tian probe shows the right picture....

The Tian probe is fine in most amps and probably even in most loops in your amp.
So I think the Tian probe shows the correct picture - if the phase is reasonable.
But when the phase is non-physical then I am suspicious.
I expect that if the conditions are such that it can not calculate phase correctly then the assumptions used in the calculation are incorrect and the result is unreliable.

Mike E... and Frank W... are responsible.

Frank could reply that he has written a correct GFT probe for you.
Not his responsibility that you use his Tian probe instead
I still don't fully understand the phase anomaly so I look forward to your experience with the GFT.


Best wishes
David

 

[matze]

Hi David,

following your repeated advice, I had a look a the Middelbrook GFT probe, namely Middlebrook's IEEE Microwave Magazine paper (2006) and the "GFT Template User's Manual".
The goal was to evaluate the total loop gain around the output stage. Since it does not possess an explicit error summing point, I hoped that the example for emitter follower analysis in the second reference above, also available as example "manual3a.asc" in Frank Wiedmann's zip archive, could help. But as I do not want to model the "VAS" output impedance as in these examples, I currently do not see how to set up a useful analysis.

Instead, I further "played around". Connecting the Miller capacitor C1 directly to the VAS Q14 collector and placing a loop gain probe between Q14 collector and Vbe multiplier, the probe "feels like in a normal amp". Both Tian and simplified Middlebrook probes yield the knee at 50kHz in Figure 'bode-ops-tian' of post #100, together with the phase behaviour that seems "unphysical" to you. In both cases, the knee vanishs with isolation buffer between amp output and TMC resistor R6; result compares to figure 'bode-ops-middlebrook'.
So I'm the more convinced that my reasonings from post #91 and post #102 go in the right direction, and that the Tian probe shows the right picture when placed between amp output and junction of all feedback branches. At this point, signal flows in both direction (TMC resistor driving into load at low frequencies), violating assumptions of simplified Middlebrook probe.

Do you know how to apply the Middlebrook GFT probe for a sub-circuit without explicit summing point, and without resorting to the way in the examples mentioned above?

A second question: Why does the phase behaviour worry you that much? The whole circuit definitely includes parallel branches of signal propagation, together with bidirectional transmission. So one should not expect minimum-phase behaviour. As I understand, this means that one cannot unambiguously relate magnitude and phase of loop gain.

Even if the GFT trial should turn out to be unsuccessful, it was a real joy to see the abstract approach taken in the IEEE paper. So thank you for that already here.

Kind regards,
Matthias