Feedback, all else being equal, has been a compromise, from my experience. However, it can be practical, and I still use it with the majority of my designs. Where I REALLY hold the line, is with open loop bandwidth. I try to make the open loop bandwidth as high as possible, preferably above the nominal audio bandwidth of 20KHz. This is not easy, and it demands a very high open loop bandwidth amplifier and limits the amount of feedback possible to use.
This is where I differ from typical op amp designers, who will be on me soon for this opinion.
Still, open loop operation without global feedback works best for me.
When it comes to 'group delay', as presented here, it is pretty much a manifestation of high bandwidth and intrinsic stability of the amplifier stage.
This is where I differ from typical op amp designers, who will be on me soon for this opinion.
Still, open loop operation without global feedback works best for me.
When it comes to 'group delay', as presented here, it is pretty much a manifestation of high bandwidth and intrinsic stability of the amplifier stage.
Being inquisitive I would ask why feedback has compromised the ultimate audio performance of your projects? I know this is a difficult question and one that should not be tossed with the simple platitudes we see here so often. If we accept that there is an issue sonically the next step is to tease out why. It will show a lot about the hearing mechanism and what we could do to make everything better.
Demian, I don't know if your question was addressed my John, or myself. And I have to say I have no scientific evidence as to why zero NFB sounds better to me, just a few non substantiated personal theories. But it's not just me that hold this opinion, the majority of hi-end DIY amps being built today by real enthusiasts are zero NFB. The last time I used NFB in a commercial project was a x1 bufferamp/filter as an output stage for an Upgraded SACD. A DIY friend heard it, and his comment was: "uses a lot of NFB, yes?"
And he was right, it was a two transistor complimentary pair with total NFB to reduce the gain to x1. When this circuit was changed to a two transistor darlington, the sound gained clarity and life, and lost the characteristics that identified the NFB to him.
Regards, Allen
Allen:
I think the issue is identifying what is different with negative feedback and why that is audible (or conversely, what is different without NFB and why that is audible).
It can be argued that a follower has a lot of feedback, however its a different case. And building an analog amplifying stage that is useful without any overt or inadvertent negative feedback is quite difficult. Its done in RF but even there a successful implementation is difficult.
Arguing the pros and con's of NFB is quasi-religious here and I don't find that interesting or useful. Perhaps John's criteria (low levels of higher harmonics, particularly 7th) may be more useful as a guidepost. Your comment about unity gain amp with NFB vs. follower may hold more info in understanding the differences. And euphonic components are a bit of a zinger here.
Several questions begging for answer are first, audible thresholds for distortion components: harmonic, intermodulation or otherwise (for example this Experimental Study : Distortion - Axiom Audio ) and then an understanding of how distortions effect an audio program signal as Neson Pass touches on here: http://www.passlabs.com/pdfs/articles/distortion_and_feedback.pdf and then we touch on the delicate issue of "better" sound vs. "more accurate" sound. The last can be philosophically the most difficult to untangle since reproduction is just that, its not the original in any sense and our hearing never exists in isolation from our other senses and experiences.
A magnificent meal with interesting company followed by a concert of acoustic instruments in a real space cannot ever be meaningfully compared to a dinner of leftovers followed by the best HiFi in the universe. Context has a profound impact on experience. Some recorded music never existed except in that synthetic recorded space making any judgement fundamentally subjective. It doesn't exist until its played back.
On the one hand I'm interested in learning more since it may lead to a better "sounding" product and I don't by the simple NFB is bad line, I think the issue is far more complex.
I think the issue is identifying what is different with negative feedback and why that is audible (or conversely, what is different without NFB and why that is audible).
It can be argued that a follower has a lot of feedback, however its a different case. And building an analog amplifying stage that is useful without any overt or inadvertent negative feedback is quite difficult. Its done in RF but even there a successful implementation is difficult.
Arguing the pros and con's of NFB is quasi-religious here and I don't find that interesting or useful. Perhaps John's criteria (low levels of higher harmonics, particularly 7th) may be more useful as a guidepost. Your comment about unity gain amp with NFB vs. follower may hold more info in understanding the differences. And euphonic components are a bit of a zinger here.
Several questions begging for answer are first, audible thresholds for distortion components: harmonic, intermodulation or otherwise (for example this Experimental Study : Distortion - Axiom Audio ) and then an understanding of how distortions effect an audio program signal as Neson Pass touches on here: http://www.passlabs.com/pdfs/articles/distortion_and_feedback.pdf and then we touch on the delicate issue of "better" sound vs. "more accurate" sound. The last can be philosophically the most difficult to untangle since reproduction is just that, its not the original in any sense and our hearing never exists in isolation from our other senses and experiences.
A magnificent meal with interesting company followed by a concert of acoustic instruments in a real space cannot ever be meaningfully compared to a dinner of leftovers followed by the best HiFi in the universe. Context has a profound impact on experience. Some recorded music never existed except in that synthetic recorded space making any judgement fundamentally subjective. It doesn't exist until its played back.
On the one hand I'm interested in learning more since it may lead to a better "sounding" product and I don't by the simple NFB is bad line, I think the issue is far more complex.
We have discussed the use of global negative feedback in audio systems, long and hard, over the decades.
At first, global negative feedback (hereafter called 'feedback') seems to be the answer to almost every engineer's dream, lower distortion, more long term stability, extended frequency response, etc.
In fact, many engineers, even from the first years, perhaps 70 years ago, thought that negative feedback would allow for relatively easy class B operation, pentode tube operation, rather that triode tube operation, for more power, less tube wear, and lower heat and input power requirements. You can easily understand this as important in a submarine, for example.
Well, things evolved where a compromise vacuum tube output stage, ultra linear, became dominant, sort of a combination of pentode and triode. However, with tubes, negative feedback was limited by a virtual brick wall, that was low frequency 'motorboating'. This is a low frequency oscillation due to too much feedback and caused by too many caps and transformers in the feedback loop. Generally, more than 20 dB and low frequency instability was probable.
Now what did this limit us too? Well, 20 dB feedback meant 10 times lower distortion, 10 times extended frequency response, 10 times higher damping factor, etc. Not bad.
However, that also meant that 1% distortion ONLY dropped to .1%, and this was still considered marginal and 1% distortion was not that easy of achieve open loop with class B, etc.
So, most tube amps from Heathkit to Marantz had to TRY to minimize the distortion in their open loop amps, and they were all fairly successful. A secondary factor was that the OPEN LOOP BANDWIDTH had to remain fairly high. Typically, 10KHz, which was necessary to achieve a 100KHz closed loop bandwidth, nominally.
Now with transistors, almost everything changed, higher feedback was possible, BUT this also usually necessitated lower open loop bandwidth, and many other problems. More later.
At first, global negative feedback (hereafter called 'feedback') seems to be the answer to almost every engineer's dream, lower distortion, more long term stability, extended frequency response, etc.
In fact, many engineers, even from the first years, perhaps 70 years ago, thought that negative feedback would allow for relatively easy class B operation, pentode tube operation, rather that triode tube operation, for more power, less tube wear, and lower heat and input power requirements. You can easily understand this as important in a submarine, for example.
Well, things evolved where a compromise vacuum tube output stage, ultra linear, became dominant, sort of a combination of pentode and triode. However, with tubes, negative feedback was limited by a virtual brick wall, that was low frequency 'motorboating'. This is a low frequency oscillation due to too much feedback and caused by too many caps and transformers in the feedback loop. Generally, more than 20 dB and low frequency instability was probable.
Now what did this limit us too? Well, 20 dB feedback meant 10 times lower distortion, 10 times extended frequency response, 10 times higher damping factor, etc. Not bad.
However, that also meant that 1% distortion ONLY dropped to .1%, and this was still considered marginal and 1% distortion was not that easy of achieve open loop with class B, etc.
So, most tube amps from Heathkit to Marantz had to TRY to minimize the distortion in their open loop amps, and they were all fairly successful. A secondary factor was that the OPEN LOOP BANDWIDTH had to remain fairly high. Typically, 10KHz, which was necessary to achieve a 100KHz closed loop bandwidth, nominally.
Now with transistors, almost everything changed, higher feedback was possible, BUT this also usually necessitated lower open loop bandwidth, and many other problems. More later.
a common objection is nfb "increases higher order distortion" - which is most true at low levels of loop feedback - such as the 20 dB John just mentioned
feedback proponents, aware of the several fine analysis of the problem (Baxandall, Cherry in the audio specific world, the Cordell feedback thread here) suspect some simply haven't "gotten over the hump" and applied enough loop gain to be solidly in the region of unambiguous improvement
a little feedback is provably "worse" in multiplying the levels of higher order harmonics relative to the (now reduced) closer in low order distortion products
enough feedback can drive all audio frequency distortion products below the noise floor -but requires designing for high feedback from the beginning - claiming to add 10-20 dB to a low feedback design is setting up conditions for "failure"
feedback proponents, aware of the several fine analysis of the problem (Baxandall, Cherry in the audio specific world, the Cordell feedback thread here) suspect some simply haven't "gotten over the hump" and applied enough loop gain to be solidly in the region of unambiguous improvement
a little feedback is provably "worse" in multiplying the levels of higher order harmonics relative to the (now reduced) closer in low order distortion products
enough feedback can drive all audio frequency distortion products below the noise floor -but requires designing for high feedback from the beginning - claiming to add 10-20 dB to a low feedback design is setting up conditions for "failure"
To not get ahead of ourselves, when we seriously went into audio design with solid state in the mid 1960's, most designers were either former tube designers or video or instrumentation designers, and there was little experience yet in op amp design, and IC op amps were just being introduced to the military, and even then, we found them hopelessly plagued with slew rate limitations, compared to the discrete designs that we were making then. We did not consider op amps as useful then, for audio design, until the 1970's, and even then it was problematic. More later.
Last edited:
In the 1960's, most excellent audio design with bipolar transistors was done by professional audio manufacturers, such as Ampex, Neve, and Studer. These companies normally used a 'ring of 3' type audio circuit, composed of an input transistor, a complementary second stage transistor, and an output transistor used as an emitter follower. These circuits used a single power supply and only coupling caps on the input and output, outside the feedback loop. This allowed for more negative feedback than the typical vacuum tube circuits, BUT they generally did not slew rate limit. They usually just rise-time limited to a few microseconds rise-time. This was a blessing in disguise, and many of these more than 40 year old designs can sound pretty good, even today. OP amp design came later, and generally compromised audio designs, especially with the use of available IC's that were fairly slow and with non-class A output stages.
Around 1970, advanced IC's could be purchased at a premium, perhaps $5 each. These devices were not perfect, but significantly better than the 741 type devices made by most American manufacturers. The 741 type devices were used in a lot of audio equipment, unfortunately. Better designers stuck with discrete designs. Some of these designs used a lot of negative feedback and very slow output transistors. They usually did not sound very good. Other discrete designs were faster, with slew rates greater than 30V/us, and they became the reference designs of the 1970's. While these designs used feedback, they usually had a fairly high open loop bandwidth and a minimum amount of feedback, typically 40dB, but sometimes only 20dB. The so called Otala amplifier used only 20 dB, and it sounds very good, even today. Many other 'high feedback' amplifiers made at the same time period, are now forgotten like old automobiles of lesser manufacture.
Feedback, further analyzed, has some serious drawbacks, that must be taken into account. First, is TIM distortion, that is the IM or harmonic distortion created by the non-linearity of the extra effort the input stage (usually) has to expend in order to drive its own internal capacitance, according to the formula dV/dt (slew rate)= I/C. It is this simple formula that defines slew rate almost always. However, the SAME slew rate does NOT always predict the effect of distortion at levels below slew rate. It is quite possible that a class B type input stage might be used and distortion actually higher from dv/dt related signals at low levels, than even somewhat higher levels. That is why measurement can still be necessary, and not just a limiting number used, such has 10V/us.
It has been shown by several researchers that 5V/us for a preamp and 50V/us for a power amp is a reasonable minimum to strive for, with a class A input and second stage design. Seems easy enough doesn't it? Then why the differences in design quality?
It has been shown by several researchers that 5V/us for a preamp and 50V/us for a power amp is a reasonable minimum to strive for, with a class A input and second stage design. Seems easy enough doesn't it? Then why the differences in design quality?
Last edited:
PMA, discussion at a very high level, without the grounding that it takes to really understand where we have researched, and what we have learned, tends to lead to conclusions from different groups that is unproductive, as they are not easily resolvable.
For example, high vs low feedback, high vs low open loop bandwidth, higher order distortion generation, etc.
I prefer to start from first principles, learned by me, historically, as to what has worked and what has not, in audio design.
For example, high vs low feedback, high vs low open loop bandwidth, higher order distortion generation, etc.
I prefer to start from first principles, learned by me, historically, as to what has worked and what has not, in audio design.
John:
My core question turns on why they work. Here is a core philosophical question: if the problem is addition (new harmonics) why after passing through a chain of IC based electronics upstream (recording, mixing and mastering stuff plus effects) does the very small added distortion affect the sound so much? Or vice versa do the added distortions somehow make a chain sound better when they are the right distortions? And how can we know?
The history helps ground the discussion.
Fairchild has a long complex history of making all manner of stuff. Fairchild Semiconductor was a spin-off Fairchild Camera and Instrument - Wikipedia, the free encyclopedia At one point Keith Johnson desinged tape recording electronics for them.
My core question turns on why they work. Here is a core philosophical question: if the problem is addition (new harmonics) why after passing through a chain of IC based electronics upstream (recording, mixing and mastering stuff plus effects) does the very small added distortion affect the sound so much? Or vice versa do the added distortions somehow make a chain sound better when they are the right distortions? And how can we know?
The history helps ground the discussion.
Fairchild has a long complex history of making all manner of stuff. Fairchild Semiconductor was a spin-off Fairchild Camera and Instrument - Wikipedia, the free encyclopedia At one point Keith Johnson desinged tape recording electronics for them.
Even if the signal passed through many opamps during recording process, the sonic result would be affected by specific stages, like CDP output stage, preamplifier, and power amplifier input stage. This is very interesting, and it indicates that neither PIM, nor high order harmonics are the answer.
Again, this is the problem. We can conjecture as to the best approach, especially from our own experience and what we have studied.
In any case put forward, so far, the understanding is not as complete as possible.
For example, one might think that low feedback, below 20dB is a problem, another might find that no feedback sounds best, yet another finds that measuring the harmonic order gives a powerful clue, etc. I still have to go forward with what we DO know about aberrations in the musical signal when feedback is applied, and as well note what appears as overwhelming distortion generation by some products that can still sound very good, appropriately used with other compatible equipment.
In any case put forward, so far, the understanding is not as complete as possible.
For example, one might think that low feedback, below 20dB is a problem, another might find that no feedback sounds best, yet another finds that measuring the harmonic order gives a powerful clue, etc. I still have to go forward with what we DO know about aberrations in the musical signal when feedback is applied, and as well note what appears as overwhelming distortion generation by some products that can still sound very good, appropriately used with other compatible equipment.
One historical understanding, more than 70 years old, is the relative importance of the harmonics generated, even when amplifying a single test tone. It can be shown that the harmonics are related to the amount of deviation, and the rate of change of the deviation from perfectly linear. In a Class A system, it is fairly easy to predict the distortion levels, once ONE level and its harmonic series is measured. Therefore, you can go up or down in level with predictable calculations. This can give a level of confidence that if you measure the distortion at 1V, for example, that lower output will have a predictably lower amount of distortion with every harmonic. Each harmonic, however, has a different rate of reduction, and that could be important. For example, at .1V you might find more 2'nd than 3'rd, but at 1V, you might find more 3'rd than 2'd, because it rises faster than second with level.
To understand what distortion does to effect the amplified signal, has been shown that higher order harmonics are MUCH more annoying, than 2'nd or 3'rd harmonic. In fact, 7th harmonic is pretty much intolerable, and makes music sound, metallic. This might be OK with some musical instruments, but you don't want human voices to be changed in that way, normally. There have been several weighting curves put forth over the decades to more accurately predict distortion detection, but none, to my knowledge, is built into test equipment, where it might be useful.
To understand what distortion does to effect the amplified signal, has been shown that higher order harmonics are MUCH more annoying, than 2'nd or 3'rd harmonic. In fact, 7th harmonic is pretty much intolerable, and makes music sound, metallic. This might be OK with some musical instruments, but you don't want human voices to be changed in that way, normally. There have been several weighting curves put forth over the decades to more accurately predict distortion detection, but none, to my knowledge, is built into test equipment, where it might be useful.
Nice but not useful comment. Vinyl has many distortions. I don't think we can infer from this that adding those distortions is the way forward. I have access to direct to disk recordings, master tapes and high resolution digital. I can't say that any of them are a shining beacon illuminating the direction of the future. However I do seem to get the most satisfaction from the high resolution digital when well done. However this doesn't answer any of the questions I asked.
- Status
- Not open for further replies.