I came across this paper today, as it was referenced in another post discussing "Op-Amp myths".
The F-word - or, why there's no such thing as too much feedback
Traditionally, I'm a member of the "NFB is bad, mmkay?" crowd, but I'm always willing to listen and learn. So I grabbed a copy and read it over. Admittedly, much of the technical content is above my present level (no background in control theory), and while I do have enough grounding to take away a few relevant points, a deeper understanding will present a learning curve. So what else is new, right?
But from a purely logical standpoint, I take issue with one of his core conclusions, concerning TIM / SID and their relationships to NFB. Here is the relevant statement:
"SID can be eliminated without changing loop gain. Therefore, SID is not caused by negative feedback." (p.17)
To me, this simply does not logically follow. Is it not the equivalent of stating "Resistor overheating can be eliminated without changing power dissipation. Therefore, overheating is not caused by excessive current flow"? (Other measures, such as heat sinking or fan cooling will ameliorate this).
That aside, what you you all think of Putzeys' paper? As I said, I'm not in a position to make any genuinely insightful analysis, at least not right now.
The F-word - or, why there's no such thing as too much feedback
Traditionally, I'm a member of the "NFB is bad, mmkay?" crowd, but I'm always willing to listen and learn. So I grabbed a copy and read it over. Admittedly, much of the technical content is above my present level (no background in control theory), and while I do have enough grounding to take away a few relevant points, a deeper understanding will present a learning curve. So what else is new, right?
But from a purely logical standpoint, I take issue with one of his core conclusions, concerning TIM / SID and their relationships to NFB. Here is the relevant statement:
"SID can be eliminated without changing loop gain. Therefore, SID is not caused by negative feedback." (p.17)
To me, this simply does not logically follow. Is it not the equivalent of stating "Resistor overheating can be eliminated without changing power dissipation. Therefore, overheating is not caused by excessive current flow"? (Other measures, such as heat sinking or fan cooling will ameliorate this).
That aside, what you you all think of Putzeys' paper? As I said, I'm not in a position to make any genuinely insightful analysis, at least not right now.
Last edited:
I met him on BAF, and even have his lecture recorded. He is smart guy and a talented engineer. I did not read his paper, but I believe that it has nothing unknown to the science and engineering.
There are 2 anti-feedback points,
1. ) No know-how. It is hard to do anything when you don't understand what you are doing.
2.) Amps with feedback and without feedback misbehave differently. Tube amps with no feedback clip less nasty than amps with feedback, if the amp is not designed to clip gracefully from the beginning.
Amps add phase shift that is frequency dependent. If the gain over the loop on the frequency where the shift is 180 degrees is greater than 1 such a feedback obviously becomes positive and the amp oscillates. The measures have to be taken to prevent the gain > 1 on the frequency on which phase shift becomes 180 degrees.
The same about low frequencies where coupling RC networks play their roles.
It is easier to make the first stage faster than the last one, due to presence of an output transformer.
The result is, when the slower output stage gets overloaded the previous one may clip nasty limiting slew rate. But even without feedback it would be already limited...
I.e. feedback is not an evil. It is just a science+art to make high-end amps. You can achieve very clean reproduction of soft music, but have to design properly with overload in mind because the real music has some crest-factor, and an excuse "I always listen on 1W average power" does not work. If 1W means 75 dB in the listening position, what power would you need for forte-fortissimo? Triode no feedback amp may steal it's peak loudness clipping softly, but when compared to some other amp capable to reproduce it it is a completely different story!
There are 2 anti-feedback points,
1. ) No know-how. It is hard to do anything when you don't understand what you are doing.
2.) Amps with feedback and without feedback misbehave differently. Tube amps with no feedback clip less nasty than amps with feedback, if the amp is not designed to clip gracefully from the beginning.
Amps add phase shift that is frequency dependent. If the gain over the loop on the frequency where the shift is 180 degrees is greater than 1 such a feedback obviously becomes positive and the amp oscillates. The measures have to be taken to prevent the gain > 1 on the frequency on which phase shift becomes 180 degrees.
The same about low frequencies where coupling RC networks play their roles.
It is easier to make the first stage faster than the last one, due to presence of an output transformer.
The result is, when the slower output stage gets overloaded the previous one may clip nasty limiting slew rate. But even without feedback it would be already limited...
I.e. feedback is not an evil. It is just a science+art to make high-end amps. You can achieve very clean reproduction of soft music, but have to design properly with overload in mind because the real music has some crest-factor, and an excuse "I always listen on 1W average power" does not work. If 1W means 75 dB in the listening position, what power would you need for forte-fortissimo? Triode no feedback amp may steal it's peak loudness clipping softly, but when compared to some other amp capable to reproduce it it is a completely different story!
Was he discussing the same points in the paper?
Is the recording something you can pass along?
I will remove it as soon as we start watching a movie. Our cable at home has limited bandwidth. 🙂
Thank you for the link, well written, with lot of humor! 🙂
I disagree, though, with this phrase:
How nested feedbacks are implemented impacts on how the amp will misbehave on overload, that is very significant for sound quality. Especially, which dynamic distortions will occur in the amp that contains capacitively coupled stages.
I disagree, though, with this phrase:
How nested feedbacks are implemented impacts on how the amp will misbehave on overload, that is very significant for sound quality. Especially, which dynamic distortions will occur in the amp that contains capacitively coupled stages.
Thanks for posting it, very much appreciated.
My first d/l attempt broke at 192/506MB for no reason I can see. I'm trying it one more time. If it's OK by you, I'll re-up it to my shell, and make it generally available to the membership for time being.
ETA: Broke again at ~128MB, won't bother trying again.. but I would like to hear it at some point. Maybe compress it to mp3 for a quicker d/l?
Last edited:
Sure, feel free!
And thatks again for the link! It is a great article, except he assumes unlimited dynamic range of an amp in question. ;-)
The talk though is about feedback in class D amplifiers.
There is certainly such a thing as poorly-implemented feedback, though.
I have done a lot of testing with plate-grid feedback on pentodes (just like is done in a classic inverting op-amp stage) and it makes a KT88 perform better than a 300B. Lower distortion and higher gain thanks to the KT88s higher transconductance.
I got to thinking more about it and applied the same principle to CCS-loaded pentodes (yes, you read that right), which vastly increases the feedback by creating a bunch more open-loop gain. My results were that I could make EL84s perform better than any triode that I had ever seen, and they would swing a lot more volts. The distortion that I did measure was low order (yes, and lower order than many venerated triodes) and second harmonic dominant. Tons of feedback, beautiful result. Bandwidth from DC to 300kHz+, which I can easily limit to whatever I want. I don't know what else I could ask for.
However, you can screw feedback up and get a terrible result. The hard thing is when you are including many stages, you just get too much phase shift and have to limit feedback at high frequencies and get high distortion or bad behavior up there. It is a juggling game. But there is no reason to be afraid of feedback.
I have done a lot of testing with plate-grid feedback on pentodes (just like is done in a classic inverting op-amp stage) and it makes a KT88 perform better than a 300B. Lower distortion and higher gain thanks to the KT88s higher transconductance.
I got to thinking more about it and applied the same principle to CCS-loaded pentodes (yes, you read that right), which vastly increases the feedback by creating a bunch more open-loop gain. My results were that I could make EL84s perform better than any triode that I had ever seen, and they would swing a lot more volts. The distortion that I did measure was low order (yes, and lower order than many venerated triodes) and second harmonic dominant. Tons of feedback, beautiful result. Bandwidth from DC to 300kHz+, which I can easily limit to whatever I want. I don't know what else I could ask for.
However, you can screw feedback up and get a terrible result. The hard thing is when you are including many stages, you just get too much phase shift and have to limit feedback at high frequencies and get high distortion or bad behavior up there. It is a juggling game. But there is no reason to be afraid of feedback.
its more correctly stated that in principle equivalent (or better) linear operation mode desensitivity, distortion reduction, frequency response, output Z vs frequency can be had with only global feedback vs tying up lots of the gain in local feedback loops with "the same" total gain stages
but the practical limits are as Wavebourn states - mainstream, historically proven designs that use local feedback/degeneration multiple places avoid a zoo of large signal nonlinear problems topologically rather than with elaborate clamping, recovery circuitry
"too much" forward gain controlled only by a (possibly very complicated) global feedback may have both turn-on issues and clipping problems
less an issue with tubz - few use enough of them to get enough forward gain to get in deep trouble
and not many "nfb" fanboys recognize their beloved triodes come with built-in local negative feedback
but with ss it becomes size, power and cost practical to add even tens of clamping and recovery aid parts to allow use of really high, mostly global, loop gain
Bruno's work with self oscillating Class D makes him the expert in the large signal nonlinear issues - hard to point at any other "name" designer with the experience, successful designs in that area
but the practical limits are as Wavebourn states - mainstream, historically proven designs that use local feedback/degeneration multiple places avoid a zoo of large signal nonlinear problems topologically rather than with elaborate clamping, recovery circuitry
"too much" forward gain controlled only by a (possibly very complicated) global feedback may have both turn-on issues and clipping problems
less an issue with tubz - few use enough of them to get enough forward gain to get in deep trouble
and not many "nfb" fanboys recognize their beloved triodes come with built-in local negative feedback
but with ss it becomes size, power and cost practical to add even tens of clamping and recovery aid parts to allow use of really high, mostly global, loop gain
Bruno's work with self oscillating Class D makes him the expert in the large signal nonlinear issues - hard to point at any other "name" designer with the experience, successful designs in that area
Last edited:
I will do it tomorrow...
[root@www music]# ls -al bruno.*
-rwxr-xr-x. 1 root root 439403547 Nov 6 2012 bruno.wav.gz
[root@www music]#
Not much. 🙂
In high quality audio amplifiers, NFB has only one useful purpose: damping speaker's main resonance. Low distortion can be, and should be achieved without NFB, if we are talking about high quality sound reproduction.
CentOS ? 🙂
The feedback cant be much because nothing happens when the input voltage equals the output one. There is no current flow when two voltages are the same.
Indeed. 🙂
What happens? Nirvana! Clap by one hand? No voltage! Just a music. Like alive. ;-)
Sure, if you agree to burn electricity instead of an excess gain. Free cheese is only in a mousetrap. 😛
...but... still, you are hiding your feedback inside of your triodes. 🙂
I agree on a contest; your pure triode amp without intentional feedbacks VS my pentode amp with nested feedbacks. Are you ready?
Last edited:
But this is a correct statement, is it not? Overheating is caused by the heat flow into the R being larger than the heat flow out of the R to the environment.
If you isolate the R enough (thermally) low current can cause overheating. With good ventilation and/or heatsinking, it will not overheat even with large current.
I agree that Bruno's statement cuts some corners, has to if you want to cram years of study into a few pages, and you can always argue one way or another. It is almost always possible to logically find flaw if you make the field wide enough, but that would not do justice to the basic insight it provides re: SID. And that is that the statement: 'there is NFB therefor there will be SID' is incorrect. And I believe that is what Bruno wanted to debunk.
Jan
It is quite easy to avoid that the amp ever clips.
Positive feedback can give a negative resistance output and provide even more damping than NFB.
I can agree to the 'can be' part, but why the 'should be'? Where is the necessity coming from?
Jan
Indeed. And I would argue, as Wavy did, that this situation is Nirwana: Vout = Vin! That's what we wanted all along, no?
But we know it never happens exactly that way. There is always a tiny difference between the signal input of the amp and the input where the feedback is returned to. How much? That depends on the open loop gain (which determines the available feedback). If the OL gain is very, very high, the difference between the two inputs is tiny, and the difference between Vi and Vout is also tiny: the amp is extremely transparant, almost distortionless.
For a rule of thumb we often assume that the two amp inputs are equal, but by definition they MUST be different. How much different? Enough to support Vout via the OL gain.
Jan
why do you belive in "can be" jan? - care to list available gain devices fundamental relations?
I think you will find gm for all of them is a power law - not inherently linear
and not reliably correctable by even order cancellation since even fet and tubz "square law " is only an approximation
I think you will find gm for all of them is a power law - not inherently linear
and not reliably correctable by even order cancellation since even fet and tubz "square law " is only an approximation
Nothing is ideal, everything distorts. That's why it is still a science + art to minimize first of all most audible errors, to fool imagination as if the sound is real.
- Status
- Not open for further replies.