Error signal on the majority of feedback amplifiers is created with a differential pair.
The transfer function of the differential pair (diff voltage in, diff current out) is S shaped, quite non-linear. This reflects the see-saw current sharing of the two transistors and skewed variation of Vbe across the two devices.
Thus, the error signal will not be too linear, particularly at high output levels.
Regardless of what we think of feedback, it cannot operate with good linearity if creation of the error signal is itself in error.
Perhaps if this flaw were eradicated feedback amps would get the rap they truly deserve......
Hugh
The transfer function of the differential pair (diff voltage in, diff current out) is S shaped, quite non-linear. This reflects the see-saw current sharing of the two transistors and skewed variation of Vbe across the two devices.
Thus, the error signal will not be too linear, particularly at high output levels.
Regardless of what we think of feedback, it cannot operate with good linearity if creation of the error signal is itself in error.
Perhaps if this flaw were eradicated feedback amps would get the rap they truly deserve......
Hugh
Is inverting topology (+input grounded) better than non inverting topology? The comparison is only made between 2 resistors (Rin vs Rfb)
AKSA said:
Hugh,
I beg to differ. Your observation about the non-linearity of the diff pair is correct, but not the conclusion that feebdack cannot operate with good linearity because of it.
I have to invoke the Black Box paradigm (David will love this - not!). You can mentally think of a Black Box with a non-linear transfer characteristic and then nfb around it to linearize it. It then doesn't matter whether the input stage of the Black Box is a diff amp or inverting input or a little guy with a fast potmeter. It is ONLY the excess gain (difference between OL gain and the feedback factor) that determines how much the original non-linearity is corrected.
The background behind this is that the non-linear diff pair is part of the non-linear gain block to be corrected. If we measure the OL non-linearity, that includes already the diff pair.
Now, if instead of a diff pair you come up with something that is more linear, that will be translated into a more linear / less non-linear OL gain. Assuming the same OL gain, the fb now corrects to the same order something that was less non-linear to begin with, so the result is better. But you could also have kept the non-linear diff amp, and improve the output stage. As long as the result (either better input stage or better output stage) is the same in the Black Box, the end-result is also the same. There is no special role for the diff input stage for error correction just because it sits in front.
Jan Didden
Hi Jan,
Thank you for your scholarly response. I can certainly follow your logic!
I guess we must agree to disagree?
Merry Christmas to you and your family,
Cheers,
Hugh
Thank you for your scholarly response. I can certainly follow your logic!
I guess we must agree to disagree?
Merry Christmas to you and your family,
Cheers,
Hugh
AKSA said:
Ahh, you follow my logic! Great 😉 . But do you then come to my conclusion as well? Seems logical...😉
Merry Christmas and also a good start for 2008 as well to you and your loved ones - stay healthy, stay sane!
Jan Didden
Jan,
I follow your logic, but I humbly suggest the next step......
The non-linearities in the diff pair are lumped in black box analogy with all the other non-linearities in the system, and are merely a scalar additon thereof. From a math standpoint, you correctly point out that only the scalar sum of these non-linearities is significant; their origin is not relevant.
But how about this one? Make the amp as linear as possible, then apply feedback to reduce the (moderate) distortion. I recall this sage advice has been around since Adam was a little boy.
A great deal of discussion here centers around reducing the Vbe distortion principally in the output stage. Soup du jour would appear to be error correction, in all its endlessly fascinating variants. But this distortion is trivial compared to distortions introduced by a common-emitter VAS which must swing rail to rail against a highly variable target impedance of an output stage. Can you not indulge me a little and concede that any attempt to remove non-linearities, wherever they appear, is laudable and should be encouraged?
I am trying to pursue a moderate course between high and low feedback designs. The golden mean, one might suggest......
May we all live longer to continue debating these weighty matters......
Cheers,
Hugh
I follow your logic, but I humbly suggest the next step......
The non-linearities in the diff pair are lumped in black box analogy with all the other non-linearities in the system, and are merely a scalar additon thereof. From a math standpoint, you correctly point out that only the scalar sum of these non-linearities is significant; their origin is not relevant.
But how about this one? Make the amp as linear as possible, then apply feedback to reduce the (moderate) distortion. I recall this sage advice has been around since Adam was a little boy.
A great deal of discussion here centers around reducing the Vbe distortion principally in the output stage. Soup du jour would appear to be error correction, in all its endlessly fascinating variants. But this distortion is trivial compared to distortions introduced by a common-emitter VAS which must swing rail to rail against a highly variable target impedance of an output stage. Can you not indulge me a little and concede that any attempt to remove non-linearities, wherever they appear, is laudable and should be encouraged?
I am trying to pursue a moderate course between high and low feedback designs. The golden mean, one might suggest......
May we all live longer to continue debating these weighty matters......
Cheers,
Hugh
AKSA said:
Hi Hugh,
Yes I agree to this; any step forward is worthwhile. I also agree that often that Vas stage is the main culprit. Maybe we don't really disagree that much; it's just that I don't agree that the non-linearity of a diff input pair is to be avoided because it has a special role for feedback (because it is differential in / se out). It's just one of the links in the chain; and the best point of attack is usual the weakest link. Maybe that's the diff input pair, maybe the Vas, maybe the output stage.
Jan Didden
AKSA said:
Hi Hugh,
This is a very good point, and I agree with you. It exposes the need for amplifier designs with good input stage dynamic range and linearity. It certainly underlines the need for emitter degeneration in conventional BJT input differential pairs.
There is a big BUT here, however. It seems that most of those who argue against negative feedback do so with the implicit point of view that conventional distortion measurements do not capture or expose the presumed "badness" of NFB. The effect you describe is eminently measurable by THD-20, 19+20 kHz CCIF, etc. So we are still left with the question of why NFB amplifiers are bad (according to those who wrongly make this generalization) even though they measure better with known distortion measuring techniques?
Cheers,
Bob
I think this discussion can be separated into two (figurative) threads - one is the technical discussion about use/non use of feedback, whilst the other is a about basic premises. If the basic premise is that feedback is bad for amplifer sound and the argumentation is subjective, then it is unlikely that some middle ground will be reached by envoking a technical discussion.
It seems that the argument being put forward now from the zero feedback proponents is that one need not heed the opinion of a community sophisticated listeners who are independent of this discussion - it is purely up to ones subjective assessment of sound. And the conclusions of that individual, subjective assessment are that feedback is bad, despite a myriad of other technical factors known the affect the perception of sound quality.
However, for one last time, let me repeat the question, and I'll leave it to be proved either way:-
Can anyone explain why consistently in the subjective assessment of amplifiers by sophiticated listeners, it is both zero feedback amplifiers and feedback amplifiers that get excellent ratings - i.e. the hypothesis that feedback affects the sound of an amplifier negatively does not seem to hold.
Conversely, If it can be demonstrated that most amplifiers acheiving good or excellent ratings are zero feedback amps, the ZFC have won the day, and henceforth, all audio amps should be designed without feedback for best sound quality.
Note, that in order to answer this question and arrive at a conclusion, we do not need to envoke a technical discussion.
I rest my case.
It seems that the argument being put forward now from the zero feedback proponents is that one need not heed the opinion of a community sophisticated listeners who are independent of this discussion - it is purely up to ones subjective assessment of sound. And the conclusions of that individual, subjective assessment are that feedback is bad, despite a myriad of other technical factors known the affect the perception of sound quality.
However, for one last time, let me repeat the question, and I'll leave it to be proved either way:-
Can anyone explain why consistently in the subjective assessment of amplifiers by sophiticated listeners, it is both zero feedback amplifiers and feedback amplifiers that get excellent ratings - i.e. the hypothesis that feedback affects the sound of an amplifier negatively does not seem to hold.
Conversely, If it can be demonstrated that most amplifiers acheiving good or excellent ratings are zero feedback amps, the ZFC have won the day, and henceforth, all audio amps should be designed without feedback for best sound quality.
Note, that in order to answer this question and arrive at a conclusion, we do not need to envoke a technical discussion.
I rest my case.
Hi
If one paid enough money to the right people, they would no doubt have created the best "sounding" amp ever made and would be top seller regardless of actual measureable quality.🙄 However, with no real experimenting and no clear technical data to compare, the FB vs non FB argument goes in pointless circles. Bonsai has a good point about the nature of the debate hear. Perhaps a "simple" experiment could be designed by the knowledgeable folks here that could be performed by multiple people around the globe to attain actual data, then maybe there will be a conclusion to this tail-chasing argument that seems to pop up around here every now and then.
Maybe we just need more permanent threads on NF.
🙂
If one paid enough money to the right people, they would no doubt have created the best "sounding" amp ever made and would be top seller regardless of actual measureable quality.🙄 However, with no real experimenting and no clear technical data to compare, the FB vs non FB argument goes in pointless circles. Bonsai has a good point about the nature of the debate hear. Perhaps a "simple" experiment could be designed by the knowledgeable folks here that could be performed by multiple people around the globe to attain actual data, then maybe there will be a conclusion to this tail-chasing argument that seems to pop up around here every now and then.
Maybe we just need more permanent threads on NF.
🙂
My understanding is that very high feedback has two effects; it reduces existing open loop distortion by roughly the feedback factor, but it also produces a spray of intermodulation artefacts which can reach out to high orders, albeit at very low levels.
Thus the nature of the OL distortion is important in terms of magnitude and order, at least as important as the feedback factor.
Is there any evidence that an amp with OL distortion of 10% and 60dB of feedback sounds subjectively superior to one with 1% OL distortion and 40dB of feedback?
Of course, the nature of the distortion to begin with would have to be identical; that is, the transfer function of both amps would need to differ only in terms of OL distortion magnitude.
The likelihood of performing an experiment like this is rather small; it's a tall order.
Hugh
Thus the nature of the OL distortion is important in terms of magnitude and order, at least as important as the feedback factor.
Is there any evidence that an amp with OL distortion of 10% and 60dB of feedback sounds subjectively superior to one with 1% OL distortion and 40dB of feedback?
Of course, the nature of the distortion to begin with would have to be identical; that is, the transfer function of both amps would need to differ only in terms of OL distortion magnitude.
The likelihood of performing an experiment like this is rather small; it's a tall order.
Hugh
Bonsai,
The question you pose looks reasonable on the face of it, but ignores reality. There are very, very few solid state zero feedback amps available in the marketplace. The overwhelming majority of designers do not know how to build circuits that test decently without using feedback, are so indoctrinated with the idea that NFB is an "of course" sort of thing, or bow to market pressures such as THX approval.
If a reviewer has only feedback amplifiers to listen to, then s/he will write reviews exclusively about feedback amplifiers. If, say, 1% of the amps available on the market are zero feedback, with another two or three percent being at 20dB or less (just making up numbers that I think most people would agree are "low" feedback), then you might expect roughly that same percentage of amps to achieve good reviews if they sound about as good as feedback amps.
But for 1% of the available amps to achieve 10 or 20% percent of the favorable reviews, including (apparently) Stereophile's amp of the year or whatever it was--this indicates better than average sound. There are statistical methods for analyzing this sort of thing, but I doubt you're interested in approaching it from that angle. It'd be a lot of work and you wouldn't accept the results as scientifically valid, no matter how the tallies were done.
Tube amps might or might not count, depending on your point of view. Audio Research used about 12dB NFB and CJ about 10dB NFB the last time I surveyed such things. Clearly pretty low rates of feedback in both cases. However, the mere mention of tubes disqualifies circuits in some folks' eyes.
Incidentally, an earlier post claiming that tube circuits couldn't muster enough gain for fairly high rates of feedback only served to show the poster's ignorance of tubes. Even setting aside hybrid circuits in which a P device is used to actively load the plate of a tube, if you use as many gain stages as, say, Mark Levinson does, and used 12AX7s, you could easily get 50-80dB feedback. If you start throwing in other tricks, such as positive feedback in any of its various guises (e.g. bootstrapping), then you could probably get triple digit feedback without breaking a sweat. The reason tube designers rarely resort to such tricks is because they don't need to. They get fair specs and excellent sound without having to resort to huge amounts of feedback.
If you were to include tube circuits as representatives of low feedback, then it's a landslide. Low-to-no feedback wins the "most favorable reviews" sweepstakes by a country mile.
I prefer to just listen and see what I think rather than sit around counting reviews or spend all my time reading distortion off my HP.
Grey
The question you pose looks reasonable on the face of it, but ignores reality. There are very, very few solid state zero feedback amps available in the marketplace. The overwhelming majority of designers do not know how to build circuits that test decently without using feedback, are so indoctrinated with the idea that NFB is an "of course" sort of thing, or bow to market pressures such as THX approval.
If a reviewer has only feedback amplifiers to listen to, then s/he will write reviews exclusively about feedback amplifiers. If, say, 1% of the amps available on the market are zero feedback, with another two or three percent being at 20dB or less (just making up numbers that I think most people would agree are "low" feedback), then you might expect roughly that same percentage of amps to achieve good reviews if they sound about as good as feedback amps.
But for 1% of the available amps to achieve 10 or 20% percent of the favorable reviews, including (apparently) Stereophile's amp of the year or whatever it was--this indicates better than average sound. There are statistical methods for analyzing this sort of thing, but I doubt you're interested in approaching it from that angle. It'd be a lot of work and you wouldn't accept the results as scientifically valid, no matter how the tallies were done.
Tube amps might or might not count, depending on your point of view. Audio Research used about 12dB NFB and CJ about 10dB NFB the last time I surveyed such things. Clearly pretty low rates of feedback in both cases. However, the mere mention of tubes disqualifies circuits in some folks' eyes.
Incidentally, an earlier post claiming that tube circuits couldn't muster enough gain for fairly high rates of feedback only served to show the poster's ignorance of tubes. Even setting aside hybrid circuits in which a P device is used to actively load the plate of a tube, if you use as many gain stages as, say, Mark Levinson does, and used 12AX7s, you could easily get 50-80dB feedback. If you start throwing in other tricks, such as positive feedback in any of its various guises (e.g. bootstrapping), then you could probably get triple digit feedback without breaking a sweat. The reason tube designers rarely resort to such tricks is because they don't need to. They get fair specs and excellent sound without having to resort to huge amounts of feedback.
If you were to include tube circuits as representatives of low feedback, then it's a landslide. Low-to-no feedback wins the "most favorable reviews" sweepstakes by a country mile.
I prefer to just listen and see what I think rather than sit around counting reviews or spend all my time reading distortion off my HP.
Grey
AKSA said:
Hi Hugh,
Yes, and no. There are some common mis-conceptions about this, and it is at best a half-truth. I'll be posting some simulations and measurements in the permanent feedback thread over the next couple days to illustrate this. Baxandall illustrated it, and there was nothing wrong with what he did, but how some people have interpreted and generalized those particular results has been problematic.
Nevertheless, you are correct in observing that the magnitude and nature of the open loop distortion is important. Designers who use feedback and ignore the open loop get what they deserve and embarrass the rest of us who wisely use feedback as a very effective tool.
Cheers,
Bob
Bob, in his recent talk at the European Triode Festival, Jan Didden showed the infamous Baxandall chart of square-law device distortion versus feedback. He also showed some experimental results from an amp with and without a small (12dB or so) amount of negative feedback. The experiment did not match the theoretical predictions- all higher order terms were diminished with feedback compared to open loop.
I suppose real amps aren't pure square law devices, and the amount of feedback needed to knock any higher order distortion down can be significantly smaller than Baxandall's predictions. So, yes, a few boulders of salt are in order.
I suppose real amps aren't pure square law devices, and the amount of feedback needed to knock any higher order distortion down can be significantly smaller than Baxandall's predictions. So, yes, a few boulders of salt are in order.
SY said:
What? Reality didn't match theoretical predictions?!
Gasp!
Somebody had better slap reality upside the head and tell it to get with the program. This kind of behavior is completely unacceptable.
Grey
The math works, but perhaps the experiment was unsuccessful.
For the record, it is almost impossible to model a fet with a simple square law. This has been known for decades. Also, the same jfet number, made by two manufacturers, can be significantly different in its distortion products. This is one major reason why I have not depended on modeling for my designs, I just build it.
For the record, it is almost impossible to model a fet with a simple square law. This has been known for decades. Also, the same jfet number, made by two manufacturers, can be significantly different in its distortion products. This is one major reason why I have not depended on modeling for my designs, I just build it.
And real amplifiers, as opposed to isolated ideal devices, aren't usually particularly square-law-ish.
- Status
- Not open for further replies.