SY said:
Also, earlier in one of these threads I asked John how much feedback he used at 20 kHz. His original answer was 40 dB. I was quite surprised, as this implies an aggressive high-feedback design. It turns out there was a misunderstanding and the number he quoted was, I think, for low frequencies.
So if that number is correct for low frequencies, and using the 4 kHz number for OL BW quoted in the link above, this would indicate a unity loop gain frequency of around 400 kHz. That would in turn give a feedback factor at 20 kHz of 26 dB.
Johan Potgieter said:
Ultra low distortion in solid-state audio power amplifiers simply isn’t possible without ensuring that the amplifier is highly linear before applying global negative feedback. This has been well known for 30+ years.
Stability issues and laws of nature dictate the extent to which global negative feedback can be applied to reduce THD throughout the audio band. A bad designer simply cannot just endlessly ramp up the NFB factor to cure his designs linearity ills.
This is something, which gets implied around here endlessly, and it is complete and utter BS.
A high global NFB amplifier with <0.001% THD-20 would likely have significantly better open loop linearity than a zero global NFB amplifier with 0.1% THD-20.
In “Hi-End” HiFi, there is no shortage of high NFB amplifiers that attract critical acclaim from the subjectivists. Same, to a degree, with low NFB designs, or designs with MOSFET output stages or BJT output stages.
If a designer wishes to use only a little NFB, or do without it all together, then good luck to them. But to dogmatically declare “NFB=bad” is just plain daft – especially so, to the extent that it implies that high NFB designs must also be bad. To that extent, it just plain stinks.
Cheers,
Glen
I don't see why people feel that there's a conflict in John's personal philosophy and preferences versus some of the work he's done for others (e.g. Parasound). They say: We want thus and such. He provides a functioning design. They pay him. He pays his electricity bill and buys food.
This is hard to understand?
Grey
This is hard to understand?
Grey
G.Kleinschmidt said:
Okay, I just looked at the Stereophile distortion plots of the JC-1. I would expect that if the open-loop distortion were approximately independent of frequency, the closed-loop distortion at 20 kHz would be about 5 times worse than the midband distortion - due to the 4 kHz open-loop BW. But the data do not show this - the distortion at 20 kHz is only about 2x the distortion at 1 kHz. So either the open-loop distortion is getting less as frequency increases, or something is going on with Stereophile's measurements of high-frequency distortion. Maybe they are cutting off the high-order harmonics in the 20 kHz THD test.
One possible explanation for decreasing open-loop distortion with increasing frequency might be that he uses so many output devices, that the VAS distortion, which decreases with increasing frequency (due to Miller comp) is actually dominating over output stage distortion over some of the frequency range.
But yes, it does indicate very good open-loop linearity and a great job by John all around.
I too get fed up with the often repeated false dichotomy implying that high-feedback designs must have high open-loop distortion. It's much easier and less time-consuming to mindlessly repeat something than to properly explain it. That's why people who carefully explain things end up getting the short end of the stick in the propaganda wars around here.
Jahan, you have the right to live in ignorance of our experience, if you wish.
However, if I wanted to learn something, I would listen to someone with experience.
For example, a few blocks away is a restaurant that I seldom eat at, because it is so difficult to get a reservation and it is very expensive. It is run by a world famous cook named Alice Waters. She has several books out as well.
When I DO eat at her restaurant, just for lunch, I always find it to be an incredible experience. IF I wanted to emulate HER cooking, I would buy one of her books and take her advice, even if it did not initially make sense to me.
It is the same with audio design. I can't PROVE to you that negative feedback is often problematic. I just know this from experience, especially when in competiton with designs that use less or no global feedback than I do. You see, my associates like Charles and Nelson compete with me to make the best audio equipment possible. Sometimes, I win, most of the time, they win. Since they tend to use less feedback than I normally do, I think that this is an advantage for them. After all, we all know how to make high feedback amplifiers. I once made a 2000W current output power amp in 1969, with 4 separate feedback loops, 2 positive and 2 negative. Try that sometime! I even added a common mode servo with an IC in order that it did not 'latch up'. Of course, that was before I read Matti Otala's paper on amp design.
What I am trying to point out is there are some real, experienced engineers, who know just about anything you might know about feedback, but try to reduce or avoid it.
Do you think that we just do this to confuse the issue?
However, if I wanted to learn something, I would listen to someone with experience.
For example, a few blocks away is a restaurant that I seldom eat at, because it is so difficult to get a reservation and it is very expensive. It is run by a world famous cook named Alice Waters. She has several books out as well.
When I DO eat at her restaurant, just for lunch, I always find it to be an incredible experience. IF I wanted to emulate HER cooking, I would buy one of her books and take her advice, even if it did not initially make sense to me.
It is the same with audio design. I can't PROVE to you that negative feedback is often problematic. I just know this from experience, especially when in competiton with designs that use less or no global feedback than I do. You see, my associates like Charles and Nelson compete with me to make the best audio equipment possible. Sometimes, I win, most of the time, they win. Since they tend to use less feedback than I normally do, I think that this is an advantage for them. After all, we all know how to make high feedback amplifiers. I once made a 2000W current output power amp in 1969, with 4 separate feedback loops, 2 positive and 2 negative. Try that sometime! I even added a common mode servo with an IC in order that it did not 'latch up'. Of course, that was before I read Matti Otala's paper on amp design.
What I am trying to point out is there are some real, experienced engineers, who know just about anything you might know about feedback, but try to reduce or avoid it.
Do you think that we just do this to confuse the issue?
andy_c said:
AP has a 22kHz lpf that is often engaged for THD measurements. I don't know if that is the case here (I lost the link to the Stereophile test) but that could explain it. Anybody has that link somewhere so I could take a look? The Stereophile search function doesn't help me.
Jan Didden
john curl said:
John,
Nobody disputes your experience and your successful designs. But being what they are, people who build their own equipment always ask WHY things are the way they are. That's what is sometimes frustrating. Not only tell us: "this is how it is" but please tell us why. Or tell us that it is your experience and you are not sure of the technical reasons. That's OK too, and would stop our speculations about the reasons for silence.
Jan Didden
High NFB and Low NFB - what do we mean?
What do we mean by a high feedback versus a low feedback amplifier design?
Clearly, the amounts of feedback at low frequencies and at 20 kHz must play a role in this definition, as may the open-loop bandwidth.
Let’s first consider amplifiers with wide open-loop bandwidth exceeding 20 kHz. We can probably agree that such a design with 10-20 dB of NFB across the audio band is a low feedback design. I’m also inclined to believe that an amplifier with 30-40 dB of NFB across the band would be considered by most to be a high feedback design. Note that an amplifier with simple dominant pole compensation starting at 20 kHz will have a rather aggressive 2 MHz closed loop bandwidth.
But now let’s include the situation where open loop bandwidth is less than 20 kHz, sometimes by a lot. Consider an amplifier with 20 dB of feedback at 20 kHz and a 4 kHz open-loop bandwidth. It will have about 34 dB of NFB at frequencies below 4 kHz. Is this a low feedback design or a high feedback design? Do we assign the designation based on the high-frequency feedback value or the low-frequency value?
One reason this semantical conundrum may matter is that it may get to the root of the reasons why high or low feedback designs might sound better or worse. At least it raises some questions that may be relevant to a better understanding.
For example, suppose a design with 20 dB NFB at 20 kHz and 40 dB at LF with a 2 kHz open-loop bandwidth (OLBW) is designated by NFB-averse designers to be a high-feedback design that sounds bad because it is a “high feedback” design. Assume also that the designer has done a good job of linearizing the open loop (as most of us agree he should).
Does this amplifier sound bad because of the extra feedback at LF? Does it sound bad because of the limited 2 kHz open-loop bandwidth? If so, then how will the amplifier sound if we increase the feedback at 20 kHz to 40 dB? It will now clearly be a “high-feedback” design. Will it sound great because the OLBW has been increased? If it does, we now have an example that shows that the generalization of “high NFB is bad” is wrong. Perhaps, then, the high amount of negative feedback per se is not the problem.
Then we see that the semantics are important, because the NFB-averse designers might then say that what they really mean in the high NFB designation was amount of NFB at low frequencies.
Or, will the feedback-averse designers say that the above amplifier with 40 dB NFB flat to 20 kHz will sound better if that number is reduced to 20 dB NFB across the band.
I don’t know. But I do think that these kinds of Devil’s Advocate questions can lead to some thought-provoking discourse that may lead to a better understanding of the merits and demerits of NFB, and a better definition of the technical and philosophical positions of the feedback-is-good and the feedback-is-bad advocates.
Cheers,
Bob
What do we mean by a high feedback versus a low feedback amplifier design?
Clearly, the amounts of feedback at low frequencies and at 20 kHz must play a role in this definition, as may the open-loop bandwidth.
Let’s first consider amplifiers with wide open-loop bandwidth exceeding 20 kHz. We can probably agree that such a design with 10-20 dB of NFB across the audio band is a low feedback design. I’m also inclined to believe that an amplifier with 30-40 dB of NFB across the band would be considered by most to be a high feedback design. Note that an amplifier with simple dominant pole compensation starting at 20 kHz will have a rather aggressive 2 MHz closed loop bandwidth.
But now let’s include the situation where open loop bandwidth is less than 20 kHz, sometimes by a lot. Consider an amplifier with 20 dB of feedback at 20 kHz and a 4 kHz open-loop bandwidth. It will have about 34 dB of NFB at frequencies below 4 kHz. Is this a low feedback design or a high feedback design? Do we assign the designation based on the high-frequency feedback value or the low-frequency value?
One reason this semantical conundrum may matter is that it may get to the root of the reasons why high or low feedback designs might sound better or worse. At least it raises some questions that may be relevant to a better understanding.
For example, suppose a design with 20 dB NFB at 20 kHz and 40 dB at LF with a 2 kHz open-loop bandwidth (OLBW) is designated by NFB-averse designers to be a high-feedback design that sounds bad because it is a “high feedback” design. Assume also that the designer has done a good job of linearizing the open loop (as most of us agree he should).
Does this amplifier sound bad because of the extra feedback at LF? Does it sound bad because of the limited 2 kHz open-loop bandwidth? If so, then how will the amplifier sound if we increase the feedback at 20 kHz to 40 dB? It will now clearly be a “high-feedback” design. Will it sound great because the OLBW has been increased? If it does, we now have an example that shows that the generalization of “high NFB is bad” is wrong. Perhaps, then, the high amount of negative feedback per se is not the problem.
Then we see that the semantics are important, because the NFB-averse designers might then say that what they really mean in the high NFB designation was amount of NFB at low frequencies.
Or, will the feedback-averse designers say that the above amplifier with 40 dB NFB flat to 20 kHz will sound better if that number is reduced to 20 dB NFB across the band.
I don’t know. But I do think that these kinds of Devil’s Advocate questions can lead to some thought-provoking discourse that may lead to a better understanding of the merits and demerits of NFB, and a better definition of the technical and philosophical positions of the feedback-is-good and the feedback-is-bad advocates.
Cheers,
Bob
Market what? PMA LOW FEEDBACK? Why would I do that when I use lots of it in my commercial designs that I actually get some money for? Sounds like anti-marketing to me.
I don't mind or care that details of how to add more or better negative feedback solutions are discussed here. This is a negative feedback thread, is it not?
However, it is my opinion that negative feedback in audio designs is problematic, along with several others contributing here, who also have been successful in the audio design business. That is all there is to it, and I will leave it at that.
I don't mind or care that details of how to add more or better negative feedback solutions are discussed here. This is a negative feedback thread, is it not?
However, it is my opinion that negative feedback in audio designs is problematic, along with several others contributing here, who also have been successful in the audio design business. That is all there is to it, and I will leave it at that.
john curl said:
Hi John,
This is not meant to be a negative feedback thread solely for the purpose of promoting negative feedback. It is definitely intended to discuss and understand the differences among no-NFB, low-NFB and high-NFB approaches.
It is definitely supposed to be a forum where NFB-averse designers can explain the bad things about NFB, whether it be overload behavior, concerns about stability, TIM, open-loop bandwidth or whatever, whether we agree on the merits of these or not. This is how we all learn.
The earlier presentation of the Baxandall data showing that NFB can actually increase the presence of some harmonics and maybe even create new ones under certain conditions is a perfect example of what should be included in this thread. Both sides trying to better understand the other's position is key to this, again, even if we don't agree.
Expression of opinion one way or the other based on experience has its place here, but because of its anecdotal nature there is less that we can do with it in the discussions. On the other hand, when one says that feedback is bad because it does thus and such (for example creates more 7th-order distortion), that is something that we can better get our teeth into.
The question that I posed above was an attempt on my part to keep this very interesting discussion going in a way that we all might learn from.
Cheers,
Bob
May I chime in and repeat a question/proposal?
Another thing about feedback in power amps: When we have a control loop, does voltage control do better than current control, or is a combined approach ("power" control) the optimum? With current control I mean a voltage controlled current pump (like a Howland or some sort of the circuit type I mentioned above) in a voltage feedback loop, while a voltage control is a mere unity gain current buffer in the loop. For the former the control voltage (output of the error amplifier, thus error voltage itself also) is also greatly function of load current, for the latter load current does not play a role (simplifying, of course). And in a power feedback, both current and voltage are reflected in the control voltage, in some optimized weighting. An example would be a current buffer/pump with a defined and constant output resistance, inside of the error control loop. In any case of course it would be good when the choosen control variable is as linear as possible, regarding its transfer function.
The thinking behind this is that speakers (that is, drivers, actually), though normally specified with constant voltage, are still current devices (and at times are deliberately run with non-zero generator-Z, positve and negative, though this has nothing to do with the amp's control mechanism, I still assume the normal zero Z as the design goal). I feel that the current nature (including stored energy effects, back-EMF), when reflected in the primary error voltage might relax the sound (such is reported for some current-controlled amplifiers).
I hope I could express my thoughts (not a native speaker + difficult subject == possible misunderstanding).
Regards, Klaus
That could be seen as linearizing a device, getting a very linear compound device which wouldn't need much global FB to work.
Another thing about feedback in power amps: When we have a control loop, does voltage control do better than current control, or is a combined approach ("power" control) the optimum? With current control I mean a voltage controlled current pump (like a Howland or some sort of the circuit type I mentioned above) in a voltage feedback loop, while a voltage control is a mere unity gain current buffer in the loop. For the former the control voltage (output of the error amplifier, thus error voltage itself also) is also greatly function of load current, for the latter load current does not play a role (simplifying, of course). And in a power feedback, both current and voltage are reflected in the control voltage, in some optimized weighting. An example would be a current buffer/pump with a defined and constant output resistance, inside of the error control loop. In any case of course it would be good when the choosen control variable is as linear as possible, regarding its transfer function.
The thinking behind this is that speakers (that is, drivers, actually), though normally specified with constant voltage, are still current devices (and at times are deliberately run with non-zero generator-Z, positve and negative, though this has nothing to do with the amp's control mechanism, I still assume the normal zero Z as the design goal). I feel that the current nature (including stored energy effects, back-EMF), when reflected in the primary error voltage might relax the sound (such is reported for some current-controlled amplifiers).
I hope I could express my thoughts (not a native speaker + difficult subject == possible misunderstanding).
Regards, Klaus
john curl said:
Hi John,
Thanks for this information, but it wasn't exactly what I was looking for. I wasn't asking any "what if" questions about increasing the NFB in the JC-1, I was just interested in knowing what the amount of NFB was at low frequencies and at 20 kHz, and how much the open loop bandwidth is. There is a bit of speculation in some of the postes that followed this, but it would be nice to have the answer from you if you were willing to share it with us.
I understand and appreciate your concern about the 7th harmonic distortion.
Bob