Thanks for that.
I assume Pspice uses steps to do the Tian probe.
Does Pspice not support nested steps?
Otherwise you should be able to nest your probe steps inside your resistor steps.
This is not hard in LTspice but I haven't done it quite as neatly as I would like.
I just have a Tian1 function for the first value, Tian2 for the second etc.
Best wishes
David
I assume Pspice uses steps to do the Tian probe.
Does Pspice not support nested steps?
Otherwise you should be able to nest your probe steps inside your resistor steps.
This is not hard in LTspice but I haven't done it quite as neatly as I would like.
I just have a Tian1 function for the first value, Tian2 for the second etc.
Best wishes
David
Hi David,
Yes, it uses the stepped approach. It would seem there is no support in pspice for nested steps I'm afraid. More info on the loopgain probe I use here http://www.analog-innovations.com/subcircuits.html in the loopgain.zip file.
Best wishes.
Ian
Yes, it uses the stepped approach. It would seem there is no support in pspice for nested steps I'm afraid. More info on the loopgain probe I use here http://www.analog-innovations.com/subcircuits.html in the loopgain.zip file.
Best wishes.
Ian
Last edited:
I have worked a little more with the design and fitted a JFET frontend, It seems a little more critical in terms of phase margin but this version seems stable.
The cherry compensation provides good distortion figures under large signals and high frequencies so at +-30V@20 KHZ distortion is really low with 0.05%. Something is in a way odd as with smaller signals the 3. harmonic is dominant, but as signals and frequency increases the dominant distortion changes and 2. harmonic becomes the largest (by far).
The cherry compensation provides good distortion figures under large signals and high frequencies so at +-30V@20 KHZ distortion is really low with 0.05%. Something is in a way odd as with smaller signals the 3. harmonic is dominant, but as signals and frequency increases the dominant distortion changes and 2. harmonic becomes the largest (by far).
Last edited:
Hello Mike.
In order for Miller compensation to work effectively, it is a requirement that the output current from the input stage flows through the miller compensation capacitor. Indeed, the classical textbook analysis assumes this fact. (Vout = Iout from input stage x compensation cap reactance).
In several posts above, I produced plots where the VAS base current became significant when compared with the magnitude of current produced by the input stage and the compensation capacitor current. Increasing the VAS degeneration reduced the magnitude of the current disappearing into the VAS. Recall that input resistance of the VAS (neglecting the comp cap) is ((Beta+1) x (re + Re)). Clearly, Re (the VAS degeneration resistor) will influence how much of the input stage current will flow in the comp cap.
Thanks,
Ian
In order for Miller compensation to work effectively, it is a requirement that the output current from the input stage flows through the miller compensation capacitor. Indeed, the classical textbook analysis assumes this fact. (Vout = Iout from input stage x compensation cap reactance).
In several posts above, I produced plots where the VAS base current became significant when compared with the magnitude of current produced by the input stage and the compensation capacitor current. Increasing the VAS degeneration reduced the magnitude of the current disappearing into the VAS. Recall that input resistance of the VAS (neglecting the comp cap) is ((Beta+1) x (re + Re)). Clearly, Re (the VAS degeneration resistor) will influence how much of the input stage current will flow in the comp cap.
Thanks,
Ian
Last edited:
Depending on what you mean by "effective" and a particular circuit topology, this may or may not be true.
It is much more productive to think in terms of the Miller theorem, which states that the compensation impedance is equivalent with impedances at the input and the output, to the ground, with the values depending on the ratio of the output to input voltages, aka the "voltage gain".
As you see, the currents have no direct relevance here. Now, it is true that if the VAS/TIS input impedance is low, the Miller compensation cap value will need to be larger, for the same crossover frequency. But, except for pathological cases, this is not a real problem.
The application of the Miller theorem is what made the "Miller compensation" name and I believe it is also largely responsible for calling the second gain stage a "VAS".
I am sorry, I haven't followed this thread closely, so I don't really know what "pure Cherry compensation" stands for.
Just what Self calls "Output-Inclusive Miller Compensation" and discusses on p. 217, what Bob Cordell calls "inclusive Miller compensation and pictures and discusses on p. 182.
Forcefully recommended by Edward Cherry in several JAES articles.
Best wishes
David
I've read Cherry's articles, the "pure" word is confusing to me. What would then be a "not so pure" Cherry compensation? I also recall Cherry recommending taking the compensation at the output node, from the mid Zobel cell, is this "pure" or not?
I pointed out to Richard Lee that "Cherry" is merely a subset of TMC, he considers TMC to be a unnecessarily complicated (impure) version of "Cherry"
The term is just his florid style, independent of the Zobel take-off point.
Best wishes
David
impure Cherry?
eg I don't susbcribe to his taking his FB from the middle of the Zobel or his evil Ccb at the drivers
I explain (what IMveryHO is) the real essence of Cherry in #163
My little mod (which Cherry spurns) is what allows 1ppm THD20k @ 50W 8R with very simple circuits and low order residuals .. at least in SPICE world
I have secretly admitted to various people that my 'pure Cherry' is actually impure as I don't use ALL of Prof. Cherry's dodges
eg I don't susbcribe to his taking his FB from the middle of the Zobel or his evil Ccb at the drivers
I explain (what IMveryHO is) the real essence of Cherry in #163
My little mod (which Cherry spurns
) is what allows 1ppm THD20k @ 50W 8R with very simple circuits and low order residuals .. at least in SPICE world 
...My little mod (which Cherry spurns
I don't think Cherry spurns it.
In the JAES article "Feedback, Sensitivity and Stability of Audio Power Amplifiers" he shows a picture of an EF2 OPS and one transistor VAS but does the analysis with a Darlington VAS and comments that the one transistor "conclusion is the same".
is what allows 1ppm THD20k @ 50W 8R with very simple circuits and low order residuals .. at least in SPICE world
Actually I think it has a lot to do with the ~20 MHz ULGF around the OPS.
Of course the extra loop gain is needed.
Thanks for those.
Best wishes
David
Last edited:
Yah/Nay stability
tian
But the my fudge is the 1n or so decoupling the VAS emitter resistor.
______________
I now have to confess mea maxima culpa.
I have in the past derided Guru Zan's assertion that his Tian probe on the output before the feedback or Cherry take-off is sufficient to assess stability.
I claimed it is possible for this 'inner' loop to be stable when the 'outer' loop is unstable. I made this claim cos usually making the 'inner' loop more stable with smaller Cherry caps makes the 'outer' loop less stable.
But if you take this to the limit to make 'outer' loop unstable instead of just less stable, you find the 'inner' loop goes unstable too. So Guru Zan is correct in that his 'inner' loop is a linear Yah/Nah measure of stability.
But there's more.
If you slowly increase decrease the Cherry cap and look at both 'inner' & 'outer' loops AND the Close Loop, you'll see very interesting behaviour including the triple Closed Loop peaking that I posted on this thread IIRC.
So if you are interested in more than Yah/Nah stability, eg peaking and overshoot, the practical thing to do is the same:
Look at 'inner', 'outer' and Closed Loop responses with different models and loads too
tian
I'd forgotten that.
But the my fudge is the 1n or so decoupling the VAS emitter resistor.
______________
I now have to confess mea maxima culpa.
I have in the past derided Guru Zan's assertion that his Tian probe on the output before the feedback or Cherry take-off is sufficient to assess stability.
I claimed it is possible for this 'inner' loop to be stable when the 'outer' loop is unstable. I made this claim cos usually making the 'inner' loop more stable with smaller Cherry caps makes the 'outer' loop less stable.
But if you take this to the limit to make 'outer' loop unstable instead of just less stable, you find the 'inner' loop goes unstable too. So Guru Zan is correct in that his 'inner' loop is a linear Yah/Nah measure of stability.
But there's more.
If you slowly increase decrease the Cherry cap and look at both 'inner' & 'outer' loops AND the Close Loop, you'll see very interesting behaviour including the triple Closed Loop peaking that I posted on this thread IIRC.
So if you are interested in more than Yah/Nah stability, eg peaking and overshoot, the practical thing to do is the same:
Look at 'inner', 'outer' and Closed Loop responses with different models and loads too
... and don't forget .TRANS with different models & loads as #2306 of discrete-opamp-open-design
I now have to confess mea maxima culpa.
That's another drink you owe me.
This still seems to confuse the "inner loop" with what my Tian probe placement looks at, which is the "inner return ratio" that includes the contribution of both the inner loop and the outer loop.
The inner RR is not just a Yes/No stability check.
It actually solves the debate about what is the measure of stability of an amplifier with multiple loops.
The correct measure is not the outer loop stability, as you stated. It is not the minimum of the different loops, as I first proposed.
It is exactly the inner RR in the case where that RR cuts all loops.
Certainly still useful to look at the separate responses, especially to decide where to improve.
And the Tian probe does not check feed-forward so the Closed Loop could still theoretically be free to cause trouble.
Middlebrook has at least one example of this, but it's RF. I suspect this is not a practical concern for an audio amp but I am of your opinion, can't hurt to check.
Absolutely.
Best wishes
David
Actualy there s a local compensation from the OS output
to VAS input , so no TMC at all.
The global loop is effectively sampled from the mid point of a zobel network but you could as well use a separate RC network such that the global NFB ratio is gradualy reduced at high frequencies , i.e , the feedback loop
bandwith is limited to increase phase and gain margin, actualy without this limitation the Cherry concept would be unstable and indeed void of sense.
I am not really sure what you mean.
What you have plotted is not the Tian probe response, it looks like the closed loop.
A Tian probe should never make a difference to the stability of a circuit.
I can't read the circuit clearly, if you can post an ASC it would help.
Best wishes
David
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.