Threading the maze.
It is quite understandable that given the diversity of circuit topologies and device options, particular design instances may elicit equally particular attention to distortion mechanisms, like the infamous PIM among others. Travelling this maze may suggest to the unaware that amplifier design is like a wild forest journey, full of varied and challenging threats.
While this is not very far from the truth, it should not obscure the fact that there exist abstractions of a higher level which allow to design and predict reliably the final outcome, as long as the particular details are reasonably well handled. That is, a basic forward chain devoid of gross design flaws, will perform as good as one wants to, no matter the actual selected topology, in a global negative feedback arrangement.
The kind of abstraction that lets one to assert this, is simply the tried and true NFB equation. Being A open loop forward gain and B feedback attenuation, system gain G is given by the well known expression G = A/(1+AB).
Sensitivity to deviations of system gain with respect to forward gain A (plant in terms of Control Theory). is given as ratio of relative errors as:
(dG/G)/(dA/A)=(dG/dA).(A/G)=1/(1+AB) or approximately 1/AB.
This holds no matter the nature of the deviations dA with respect to ideal (dA should be zero), so it includes both amplitude, frequency and phase anomalies.
1/AB is the available correction factor, meaning:
1. If the starting amplifier A is very bad, it can be corrected only so far.
2. AB is frequency dependent, meaning different corrections will be available for different frequencies.
3. AB cannot be made arbitrarily large, being limited by practical device capabilities and closed loop stability.
4. This does not include signal amplitude dependent limitations like slew rate, which wreak havoc on the calculations if significant.
Global negative feedback works, this is beyond discussion and current electronics should not exist without it.
Rodolfo
Bob Cordell said:
It is quite understandable that given the diversity of circuit topologies and device options, particular design instances may elicit equally particular attention to distortion mechanisms, like the infamous PIM among others. Travelling this maze may suggest to the unaware that amplifier design is like a wild forest journey, full of varied and challenging threats.
While this is not very far from the truth, it should not obscure the fact that there exist abstractions of a higher level which allow to design and predict reliably the final outcome, as long as the particular details are reasonably well handled. That is, a basic forward chain devoid of gross design flaws, will perform as good as one wants to, no matter the actual selected topology, in a global negative feedback arrangement.
The kind of abstraction that lets one to assert this, is simply the tried and true NFB equation. Being A open loop forward gain and B feedback attenuation, system gain G is given by the well known expression G = A/(1+AB).
Sensitivity to deviations of system gain with respect to forward gain A (plant in terms of Control Theory). is given as ratio of relative errors as:
(dG/G)/(dA/A)=(dG/dA).(A/G)=1/(1+AB) or approximately 1/AB.
This holds no matter the nature of the deviations dA with respect to ideal (dA should be zero), so it includes both amplitude, frequency and phase anomalies.
1/AB is the available correction factor, meaning:
1. If the starting amplifier A is very bad, it can be corrected only so far.
2. AB is frequency dependent, meaning different corrections will be available for different frequencies.
3. AB cannot be made arbitrarily large, being limited by practical device capabilities and closed loop stability.
4. This does not include signal amplitude dependent limitations like slew rate, which wreak havoc on the calculations if significant.
Global negative feedback works, this is beyond discussion and current electronics should not exist without it.
Rodolfo
MOSFET power amp closed loop BW
Adam,
You are correct: The 3 dB bandwidth was about 2 MHz.
Bob Cordell
darkfenriz said:
Adam,
You are correct: The 3 dB bandwidth was about 2 MHz.
Bob Cordell
Re: Re: PIM
Yes, I believe that my views on PIM (so-called f.m. distortion) are in disagreement with those of Charles and John. As I've stated earlier, the phenomenon is real, and is measurable, but in practice the phenomenon appears to be very small and, most importantly, is not exacerbated by negative feedback, and exists in equally large, if not larger, amounts in amplifiers without negative feedback.
In fairness to John and Charles, maybe they have measured it on otherwise-identical amplifiers with and without negative feedback and gotten different results than I have. I would be very interested to know if they have measured PIM, how they measured it, and what results they have gotten.
Bob Cordell
PMA said:
Yes, I believe that my views on PIM (so-called f.m. distortion) are in disagreement with those of Charles and John. As I've stated earlier, the phenomenon is real, and is measurable, but in practice the phenomenon appears to be very small and, most importantly, is not exacerbated by negative feedback, and exists in equally large, if not larger, amounts in amplifiers without negative feedback.
In fairness to John and Charles, maybe they have measured it on otherwise-identical amplifiers with and without negative feedback and gotten different results than I have. I would be very interested to know if they have measured PIM, how they measured it, and what results they have gotten.
Bob Cordell
Re: Re: Re: PIM
It would be of great interest to know just how they could possibly remove and apply major loop feedback at will.
This is only practicable if the amplifier's forward path gain is made very low to begin with (20-30dB?), in which case the benefits of major loop feedback are significantly neutered; as such the value of such an experiment is negligible.
Bob Cordell said:
It would be of great interest to know just how they could possibly remove and apply major loop feedback at will.
This is only practicable if the amplifier's forward path gain is made very low to begin with (20-30dB?), in which case the benefits of major loop feedback are significantly neutered; as such the value of such an experiment is negligible.
PIM with and without negative feedback
Mike,
This is a good question. The basic idea is to build an amplifier with essentially the same topology, technology and cost, then operate it with and without negative feedback, making the minor changes to its circuit as necessary. What you have speculated is essentially correct, and this is pretty much what I did in my AES paper on PIM.
For details, read Section 3.3 of my PIM paper and look at Figure 13, where a schematic of the test amplifier for operation with and without NFB is shown. A much improved PDF scan of that paper is now available on my site at www.cordellaudio.com. The feedback version of that amplifier is quite straightforward and conventional. The no-feedback version is achieved by first using increased degeneration in the input differential pair, and secondly by adding load resistors to the VAS collectors. I believe this is the sort of thing you had in mind.
Even apart from doing such an experiment on a given amplifier with and without negative feedback, perhaps John and/or Charles have measured PIM on some of their amplifiers. If so, those numbers would be interesting to see.
Bob Cordell
mikeks said:
Mike,
This is a good question. The basic idea is to build an amplifier with essentially the same topology, technology and cost, then operate it with and without negative feedback, making the minor changes to its circuit as necessary. What you have speculated is essentially correct, and this is pretty much what I did in my AES paper on PIM.
For details, read Section 3.3 of my PIM paper and look at Figure 13, where a schematic of the test amplifier for operation with and without NFB is shown. A much improved PDF scan of that paper is now available on my site at www.cordellaudio.com. The feedback version of that amplifier is quite straightforward and conventional. The no-feedback version is achieved by first using increased degeneration in the input differential pair, and secondly by adding load resistors to the VAS collectors. I believe this is the sort of thing you had in mind.
Even apart from doing such an experiment on a given amplifier with and without negative feedback, perhaps John and/or Charles have measured PIM on some of their amplifiers. If so, those numbers would be interesting to see.
Bob Cordell
Bob,
I have a quick scan of your writings and I would like to say how I respect your systematic and quantitive approach.
On the subject of feedback I just wanted to add some othogonal food for thought. There is sometimes a theoretical assumption made that applying feedback to an audio circuit optimised for open loop performance ought to improve it. I don't see why this should necessarily be so. It doesn't surprise me at all that the opposite is often experienced. When designing a control system one may have to make circuit decisions that would be unhelpful for an open loop system but much better for a closed loop system.
I understand that the discussion is specifically about measuring PIM with and without a feedback loop and that's a valid experiment.
I have a quick scan of your writings and I would like to say how I respect your systematic and quantitive approach.
On the subject of feedback I just wanted to add some othogonal food for thought. There is sometimes a theoretical assumption made that applying feedback to an audio circuit optimised for open loop performance ought to improve it. I don't see why this should necessarily be so. It doesn't surprise me at all that the opposite is often experienced. When designing a control system one may have to make circuit decisions that would be unhelpful for an open loop system but much better for a closed loop system.
I understand that the discussion is specifically about measuring PIM with and without a feedback loop and that's a valid experiment.
Traderbam, in control systems we almost always assume significant phase shifts in the band. That's why I meant an analogy with flat VS round Earth.
traderbam said:
You are an Electrical Engineer, are not you? So you studied feedback theory, Bode, Nyquist, Black.
Industrial control systems usually work above slow actuators, sensors. In a current audio circuit, we have fast enough components to solve the task. Control systems analogy would apply to speaker membrane feedback control.
Re: PIM with and without negative feedback
But emitter degeneration IS feedback, too, and applying more of it
to the "non-fb version" only would be denying a good thing to a classical Cdom-compensated amplifier.
(together with more tail current). And this is exactly the one good thing that helps slew rate.
It would be somewhat like giving the fruits of NFB to the wrong party.
Perhaps one could soften the tail current source or the mirror, if any. It's hard for me to be so destructive.
At least John hasn't, as he wrote a few days ago in a competing forum whose name I don't dare to pronounce here aloud.
Gerhard
Bob Cordell said:
But emitter degeneration IS feedback, too, and applying more of it
to the "non-fb version" only would be denying a good thing to a classical Cdom-compensated amplifier.
(together with more tail current). And this is exactly the one good thing that helps slew rate.
It would be somewhat like giving the fruits of NFB to the wrong party.
Perhaps one could soften the tail current source or the mirror, if any. It's hard for me to be so destructive.
Bob Cordell said:
At least John hasn't, as he wrote a few days ago in a competing forum whose name I don't dare to pronounce here aloud.
Gerhard
There has been an everlasting semantic discussion on term "feedback". Finally we agreed we meant global voltage parallel negative feedback from output to input.
Re: PIM with and without negative feedback
Thank you for that seminal paper, which demonstrates that PIM is really vanishingly insignificant in virtually all feedback amplifiers of even moderate competence.
I couldn't help thinking, however, that a valid comparison between the open-loop and closed-loop condition necessarily requires that the amplifier's forward path remain unchanged in both cases.
In other words, all compensation elements used with the major loop closed should, ideally, be retained when the system is operated without the major loop.
It is the difficulty of driving the open-loop AUT without first severely degenerating gain local to each stage in the forward path which prompted this rumination.
More info.
Bob Cordell said:
Thank you for that seminal paper, which demonstrates that PIM is really vanishingly insignificant in virtually all feedback amplifiers of even moderate competence.
I couldn't help thinking, however, that a valid comparison between the open-loop and closed-loop condition necessarily requires that the amplifier's forward path remain unchanged in both cases.
In other words, all compensation elements used with the major loop closed should, ideally, be retained when the system is operated without the major loop.
It is the difficulty of driving the open-loop AUT without first severely degenerating gain local to each stage in the forward path which prompted this rumination.
More info.
Bob,
I have myself thought for a long time about HOW TO perform a comparison between
C(-losed)FB and NoFB but my problem has been, the way I think of it, is how should we make the comparison, how should the NoFB amplifier look like.
I mean the way you made yours is the one way as I see it, by using local degeneration you decrease the gain until you reach the same output signal gain as for the CFB version.
My problem is that I see another way as well as how to perform these kind of comparisons, as in my opinion the other way measuring on a NoFB amplifier could be by just decreasing the input signal until we reach the same output signal level as for the CFB version, thereby all else is equal in the amplifier.
This might be a tricky case if the OLG is not flat out to 20 kHz (if 20 kHz is the highest frequency we want to analyze!).
How do you see on it?
I would at least make two versions of the NoFB amplifier and verify the results as I think when using degeneration variants we have so many ways how to degenerate an amplifiers internal circuits. And I guess I will do it some day.. 🙂
Cheers Michael
I have myself thought for a long time about HOW TO perform a comparison between
C(-losed)FB and NoFB but my problem has been, the way I think of it, is how should we make the comparison, how should the NoFB amplifier look like.
I mean the way you made yours is the one way as I see it, by using local degeneration you decrease the gain until you reach the same output signal gain as for the CFB version.
My problem is that I see another way as well as how to perform these kind of comparisons, as in my opinion the other way measuring on a NoFB amplifier could be by just decreasing the input signal until we reach the same output signal level as for the CFB version, thereby all else is equal in the amplifier.
This might be a tricky case if the OLG is not flat out to 20 kHz (if 20 kHz is the highest frequency we want to analyze!).
How do you see on it?
I would at least make two versions of the NoFB amplifier and verify the results as I think when using degeneration variants we have so many ways how to degenerate an amplifiers internal circuits. And I guess I will do it some day.. 🙂
Cheers Michael
Yes, I am an electrical engineer. I've decided to use the term "control system" rather than "negative feedback" because I want to emphasize the importance of control and the value of system thinking. It is sometimes narrows ones insights to consider only "global negative feedback".
My definition of "control system" is not limited to slow actuators and motors. For me it includes all systems from automobile cruise controllers to phase locked loops. The design challenge is unusual in audio because of the dominant importance of distortion.
I agree with you and disagree with you. 🙂 Obviously, modern parts operate at very high frequencies. Audio signals are relatvely slow but when distortion is introduced the speed needed by the control system to adequately correct errors becomes a significant problem. Parasitic components can slow down the system too.
Question. If your OL amp has -40dB distortion and you want it to have +30dB gain at 50kHz and -90dB distortion in CL, what OL bandwidth is required to ensure stability?
My definition of "control system" is not limited to slow actuators and motors. For me it includes all systems from automobile cruise controllers to phase locked loops. The design challenge is unusual in audio because of the dominant importance of distortion.
I agree with you and disagree with you. 🙂 Obviously, modern parts operate at very high frequencies. Audio signals are relatvely slow but when distortion is introduced the speed needed by the control system to adequately correct errors becomes a significant problem. Parasitic components can slow down the system too.
Question. If your OL amp has -40dB distortion and you want it to have +30dB gain at 50kHz and -90dB distortion in CL, what OL bandwidth is required to ensure stability?
traderbam said:
It depends on how many stages it has, what kind of stages, their parameters, that may be reflected in overal parameters of the opamp that must be known to answer your question. Some of them may be still out of scope. Recently I tried to find OpAmps for my new mixing console, but discovered that they don't exist: manufacturers neglect what is significant for me, but fight hardly for parameters that are not so significant for me.
Of course there is no "right" or "wrong" answer to my question. But what bandwidth do you think would be required of a practical circuit?
traderbam said:
When it starts rolling off, how does the frequency dependence of the gain look? How does the phase dependency look? How fast is differential stage? What transient processes occure when we compensate distortions?
No straight answer on your question.
Re: Re: Re: Re: PIM
Mike,
What about taking an amp that has no global feedback, and still a gain of say 30dB. That would be the open loop condition. (Don't worry about that for the moment, it can be done. Trust me 😉 )
To compare it to a global feedback version we cannot just wrap feedback around it as there is no excess gain to sacrifice.
Can we not add that excess loop gain with the feedback? Like putting a gain of 40dB preamp in front, and then enclosing the whole shebang in a 40dB feedback loop? Would that not be a valid comparison for the main amp with / without feedback? The 40dB preamp being beyond reproach in those areas we are interested in?
Jan Didden
mikeks said:
Mike,
What about taking an amp that has no global feedback, and still a gain of say 30dB. That would be the open loop condition. (Don't worry about that for the moment, it can be done. Trust me 😉 )
To compare it to a global feedback version we cannot just wrap feedback around it as there is no excess gain to sacrifice.
Can we not add that excess loop gain with the feedback? Like putting a gain of 40dB preamp in front, and then enclosing the whole shebang in a 40dB feedback loop? Would that not be a valid comparison for the main amp with / without feedback? The 40dB preamp being beyond reproach in those areas we are interested in?
Jan Didden