Loop stability clarification . . .

[catalin]

I am measuring the loop stability ,in the loop, by inserting an resistance between the source and inverting input because in this node we have the positive feedback at high freq:

[magnoman]

Originally Posted by kenpeter

I seriously don't understand the analysis method that Kean and most are using.

Still throwing a huge simulated inductor at my feedback path, to break the loop
for AC, without permitting DC to swerve into lala-land. From there, its easy to
compare input to output, and Bode plot the open loop gain and phase shift.

Between these methods, not really sure whats supposed to be the difference?

Kenpeter

Years ago I spent several nights studying Middlebrooks method, what I can easily remember is that it includes the effects of loading in both directions.

Attached a quick comparison (using a typical bjt amp) of 3 methods 1) Huge Inductor (red) 2) Vac source in series loop (green) 3) Tian probe within loop (magenta).

The circuit and models are the same, so regardless of how accurate the models are, the techniques should give the same results.
Differences are not apparent until higher frequencies (where one can be model dubivous anyway).
Pretty easy to see if the common simple Vac source is relatively accurate by just replacing it with a parallel current source.
Nice thing about the Tian probe is that it can be inserted anywhere (not confined to a low impedance or high impedance node) so you dont have to do much thinking of the loading effects.

Thanks
-Antonio

[Dave Zan]

Originally Posted by keantoken

Dave, phase linearity is affected by any poles in the amplifier compensation design. Compensation loading determines differential voltage at high frequencies, and so if compensation loading is phase linear, the amplifier's response will be. This is why simple miller compensation is usually pretty successful at small-signal phase linearity. It is easy to goof up phase linearity by applying feedback and compensation tricks. I'm guessing I understood your question?

I sure don't understand your answer.
I checked and confirmed that a critically damped (no overshoot) response is not equal to a pi/2 PM. I learned it's usually a little less - about 1.34 radians (or 76 of those American units - how do I do the little circle symbol?)
Thanks to make me learn.

Best wishes
David

[keantoken]

Dave, you know a lot more than me, and I've had no formal education so you should take everything I say with a grain of salt. I may be catastrophically off my rocker.

Put more clearly (I hope), the load on the input stage of an amplifier causes an input differential voltage that subtracts from amplifier response; hence the BW limit. The input stage is loaded mostly by the compensation for conventional amps at the frequencies in question, and so if the compensation AC current is phase-linear, the AC response of the amplifier will be as well. For two-pole compensation, the pole in the middle of the amplifier response will split the AC response into two phase-linear segments (IE the entire amp BW will not be phase-linear). This will not be readily apparent if the second pole is placed close to the BW limit.

My understanding is that any simple 6db/oct LP filter is phase-linear. In my experience multitudes of BW tweaks tend to result in erratic phase behavior. Again, in terms of phase linearity, simple compensation schemes tend to win out.

You can plot group delay in LTSpice with the Tg() function, or right-click on the phase scale and select "group delay". I use the function usually because it uses a logarithmic scale so I can see the entire trace on-screen.

[Dave Zan]

I am not sure but I think there is some confusion between the amplifier's phase response and the loop phase response, that is, the feedback loop inside the amp.
The amplifier's phase response is essentially determined by it's frequency response (unless it is so ill conceived as to include an all pass filter).
Flat frequency response ensures a linear phase response so it is not a problem for any reasonable amp.
The loop response is purely an internal issue for the amp. We don't want too much overshoot because of potential saturation on transients and we need some allowance for stability but otherwise it doesn't affect the amplifier response that the user experiences. The non-linear loop phase response of two pole schemes is just fine, it all works out internally. So use more advanced compensation schemes with a clear conscience.
Thanks for the Spice ideas. The post may seem to imply that I am not worried by phase response, but at the extremes it needs to be checked carefully and I appreciate recommendations.

Best wishes
David

[keantoken]

Flat frequency response (as far as "flat" is a visual assessment) does not ensure constant group delay (phase linearity). At the least you would have to zoom in to see the AC response variations which signal larger changes in group delay. AC response can seem flat, but group delay will indicate phase anomalies near the corner, which will show up as a spiky or non-periodic undulating square wave edge. This could be written off as transient behavior when it is actually group delay nonlinearity. As for two-pole compensation, you would actually see a small abrupt gain change if you zoomed in on the top of the square wave due to the internal pole in the middle of the AC response. These anomalies are mitigated by feedback, but AC response is dependent on differential voltage so any internal phase distortion will add to the output, even if negligibly.

Of course, this only matters to those who want good inherent step response and signal integrity at ultrasonics and higher. If these don't matter for audio, you may as well design for the lowest THD and care about step response just enough to ensure there is no dramatic amplification of RF leading to problems. If you are comfortable with the input filter defining step response then you could actually get away with very bad phase linearity, for better or for worse.

It might not be important, but I find it fascinating nonetheless.