Could it be that any criticisms of global feedback comes back to the error circuit; the extraction of the error is fraught with non-linearities, and so the feedback cannot be exploited correctly........
The transfer function of a long tail pair is S shaped, like a BL magnetic curve.
Hugh
This is my very old observation too that SS power amps which have diff pair at input measure OK but usually sounds like crap , far far away from life like sound , maybe that diff.pair S shaped transfer function have some bad influence , or who knows what else is...? ,
SS Amps with Simple Singleton CFA input stage usually sounds much more accurate to me in comparison with live sound , they can measure very good too .
And yes maybe that you looking for is something in middle between SS amps and tube amps , some simple Hybrid amp with Sigleton tube input (not tube diff.pair)stage driving SS OPS , altogether with manually adjustable both GNFB+GPFB level for best closed loop performance and best final sound too .
Best Regards !
The two best articles I have seen on Feedback are Nelson's (already linked) and this one:
Feedback and fidelity part 1
IMO feedback gets a bad rap and deservedly so not because it is inherently bad but because quite often the feedback loop is improperly designed.
I think also Norman Crowhurst should not be ignored yet I've not seen any of his comments linked or quoted thus far. He readily pointed out a variety of problems with the concept (for example, upon its use, the noise floor of the amplifier becomes that of a set of harmonic and inharmonic low level distortions) yet at the same time seemed quite willing to use it.
I personally don't like feedback as I feel it 'violates' a fundamental rule of human hearing; that being that the ear/brain system uses higher ordered harmonics to determine sound pressure. So if these harmonics are altered by the amplification (or made unmaskable by the amplification) the result will be that the music has a loudness to it that is not part of the original signal and a harshness or hardness will also be imposed. I don't think this is so much a fault of the feedback as it is a failure to do the feedback correctly. Since so many can't do it right, the result is there is a lot of brightness and harshness in the world.
My concern about the use of feedback has a lot to do with whether the amp thus behaves as a voltage source or not. This is an important aspect in amplifier design as the vast majority of speakers since the early 1960s are intended to be 'voltage driven' which is to say the amplifier can put out the same voltage regardless of load. Especially in a tube amplifier you need a bit of feedback to do that.
Turns out in high end audio not all speakers are designed to be voltage driven; some are intended to be 'power driven' by an amplifier that acts as a power source (and to do that has to have no feedback). Many of the horn systems of yesteryear are based on this idea as well. These days the reason for using the 'Power Paradigm' as I have come to call it is to take advantage of the fact that the ear/brain system has tipping points- one of them being that it translates distortion into tonality, so much so that it might actually favor a tonal coloration due to distortion over an actual frequency response error.
Since brightness is such an endemic coloration, if you use power paradigm concepts you can actually build a system that seems less colored by getting away from GNFB; relaxed neutral and free of brightness and harshness.
I am not saying that you can't do this with GNFB; I am saying its easier to do without. But now that we all have nice computers to work with I am expecting to see amps that employ GNFB that will sound perfectly fine.
Feedback and fidelity part 1
IMO feedback gets a bad rap and deservedly so not because it is inherently bad but because quite often the feedback loop is improperly designed.
I think also Norman Crowhurst should not be ignored yet I've not seen any of his comments linked or quoted thus far. He readily pointed out a variety of problems with the concept (for example, upon its use, the noise floor of the amplifier becomes that of a set of harmonic and inharmonic low level distortions) yet at the same time seemed quite willing to use it.
I personally don't like feedback as I feel it 'violates' a fundamental rule of human hearing; that being that the ear/brain system uses higher ordered harmonics to determine sound pressure. So if these harmonics are altered by the amplification (or made unmaskable by the amplification) the result will be that the music has a loudness to it that is not part of the original signal and a harshness or hardness will also be imposed. I don't think this is so much a fault of the feedback as it is a failure to do the feedback correctly. Since so many can't do it right, the result is there is a lot of brightness and harshness in the world.
My concern about the use of feedback has a lot to do with whether the amp thus behaves as a voltage source or not. This is an important aspect in amplifier design as the vast majority of speakers since the early 1960s are intended to be 'voltage driven' which is to say the amplifier can put out the same voltage regardless of load. Especially in a tube amplifier you need a bit of feedback to do that.
Turns out in high end audio not all speakers are designed to be voltage driven; some are intended to be 'power driven' by an amplifier that acts as a power source (and to do that has to have no feedback). Many of the horn systems of yesteryear are based on this idea as well. These days the reason for using the 'Power Paradigm' as I have come to call it is to take advantage of the fact that the ear/brain system has tipping points- one of them being that it translates distortion into tonality, so much so that it might actually favor a tonal coloration due to distortion over an actual frequency response error.
Since brightness is such an endemic coloration, if you use power paradigm concepts you can actually build a system that seems less colored by getting away from GNFB; relaxed neutral and free of brightness and harshness.
I am not saying that you can't do this with GNFB; I am saying its easier to do without. But now that we all have nice computers to work with I am expecting to see amps that employ GNFB that will sound perfectly fine.
Thinking about the above post, I realize that this may be a good reason to keep the output DC offset very low.
Comments?
Jan
Yes, any DC offset will cause a nonlinearity in the LTP.
Thus an equal DC offset generated by Any means can cancel the offset.
This will lower distortion where as a coupling capacitor will not.
If the ear uses a rising of the distortion-product-floor to identify loudness, could it be that NFB amplifier overload, or more likely, speaker overdrive, doesn't raise that floor in the same way (order of products and/or product rise level vs signal rise level) as the ear does when things get naturally loud? And perhaps amplifiers of some kinds of open loop nonlinearity, but without GNFB, DO more closely resemble the way distortion is created in the ear? Maybe the Zero GNFB lets things sound somewhat naturally loud at a lower actual level before speakers/amp clipping get a chance to sound ugly? Is there any way to prove or disprove such a conjecture?
I think it is true that we usually listen at home at less-than-live actual levels.
I think it is true that we usually listen at home at less-than-live actual levels.
Last edited:
If the distortion products are low, which they are in any good modern amp up to the clipping level, what does any of that matter?
We keep ignoring the one thing that human ears ARE sensitive to- frequency response. Feedback modifies that when real speaker loads are involved.
We keep ignoring the one thing that human ears ARE sensitive to- frequency response. Feedback modifies that when real speaker loads are involved.
I gather then we won't be seeing any PL or FW designs using 60 db!My view: The more linear the amplifier, the better feedback works, but also the less need for it.
BTW, my '81 Stasis 3 soldiers on trouble-free even after decades of driving reactive electrostats.
If the distortion products are low, which they are in any good modern amp up to the clipping level, what does any of that matter?
. Maybe the Zero GNFB lets things sound somewhat naturally loud at a lower actual level before speakers/amp clipping get a chance to sound ugly?
clipping behavior can be complicated in high global feedback amplifiers
and it can be controlled with intentional design for clipping recovery
or you could take the input limiting approach - soft or hard, supply tracking or not which get mention by Cordell, Self, and Lee
the negative feedback harmonic growth math is much subdued when you have more linear forward path before feedback
and high gain over audio frequencies is practical with higher order compensation and or nested feedback, "error correction" around output stage
but don't go overboard on the "more linear 1st" rubic without thinking it through, putting real options on the table
there are few option for "more linear" devices - exponential and power law are the options we have - low mu triodes are linearized by their internal plate V feedback
diff pair and CFA diamond input both use distortion cancelation through device matching giving even order cancellation
local feedback, degeneration in fact also causes harmonic growth
a textbook result is that global feedback gives the most desensitivitv, greatest distortion reduction for a fixed multistage active gain path compared to applying local feedback at intermediate stages
and it can be controlled with intentional design for clipping recovery
or you could take the input limiting approach - soft or hard, supply tracking or not which get mention by Cordell, Self, and Lee
the negative feedback harmonic growth math is much subdued when you have more linear forward path before feedback
and high gain over audio frequencies is practical with higher order compensation and or nested feedback, "error correction" around output stage
but don't go overboard on the "more linear 1st" rubic without thinking it through, putting real options on the table
To add to the conversation you really have to spell out how this “linearize 1st” can be done – otherwise the phrase is just empty noise
Cherry summed up the options to linearize a circuit as
increasing bias vs operating range (using less of the nonlinear transfer curve)
distortion cancellation as in diff pair or Class A complementary outputs
or negative feedback (in all of its forms)
yes you can choose devices - from a very restricted set of families of characteristics
tune bias conditions subject to power budget, implementation limits
use complements/diff pairs/symmetric circuits for even order cancellation
but these options are actually of limited use, are quickly exhausted - seldom give better than 20-30 dB "linearizion"
but feedback works for many orders of magnitude linearizion
so the real question in amplifier design is where to apply feedback, as local, nested, multiloop, global... for what result and at what cost in attaining circuit performance goals
there are few option for "more linear" devices - exponential and power law are the options we have - low mu triodes are linearized by their internal plate V feedback
diff pair and CFA diamond input both use distortion cancelation through device matching giving even order cancellation
local feedback, degeneration in fact also causes harmonic growth
a textbook result is that global feedback gives the most desensitivitv, greatest distortion reduction for a fixed multistage active gain path compared to applying local feedback at intermediate stages
but am I wrong when I say:
"Observe: one guy swears by a coloring SET amp, another mortgages his
house for a vintage super duper very high NFB Krell...
Since you ask, I am pretty certain we would categorize Krells as being in
the moderate feedback camp.
...the negative feedback harmonic growth math is
much subdued when you have more linear forward path before feedback
and high gain over audio frequencies is practical with higher order
compensation and or nested feedback, "error correction" around output
stage...
there are few option for "more linear" devices - exponential and power law
are the options we have - low mu triodes are linearized by their internal
plate V feedback
diff pair and CFA diamond input both use distortion cancelation through
device matching giving even order cancellation
local feedback, degeneration in fact also causes harmonic growth
a textbook result is that global feedback gives the most desensitivitv,
greatest distortion reduction for a fixed multistage active gain path
compared to applying local feedback at intermediate stages
All correct points. (except for the spelling of desensitivitv )
some examples of these 'power paradigm' speakers?
Most ESLs (Quad ESL57 and 63, Audiostatic, Sound Lab, Acoustat), most horns (especially old school) but many newer horns too- anything that works with an SETs (as SETs are power paradigm technology as well), a variety of box speakers (the Acoustic Research AR-1 being an example) including certain bass reflex designs. All single-driver full range designs like PHY or Lowther. Despite usually being used with solid state amps, planar magnetics are power paradigm technology too, owing to the fact that the impedance curve is not based on a driver in a box.
The crossover rules for a power paradigm speaker are a little different, as a low output impedance and decreasing power with increasing impedance is not a guarantee. Older speakers with midrange and tweeter controls had them not to adjust the speaker to the room but to adjust it to the voltage response of the amplifier. Obviously you don't need them if the speaker is expecting the amp to be a voltage source.
If you mix power paradigm technology with voltage paradigm technology the usual result is a tonal aberration due to frequency response error.
Owdeo, keep on using your personal opinions. Your comment of trading loop feedback for local input stage feedback makes complete sense.
First, you linearize the input stage, source of almost all TIM and PIM. There is a 'price' to pay in that the Xover distortion of the output stage 'could' become more troublesome and audible. However, with good output biasing, a linear input stage is the best thing that you can have. That is why many use complementary differential, either fet or bipolar. It is even more linear than a single diff pair, and has double the forward gain, too, so you can degenerate the input stage even more and still have enough feedback to look good on the spec sheet.
First, you linearize the input stage, source of almost all TIM and PIM. There is a 'price' to pay in that the Xover distortion of the output stage 'could' become more troublesome and audible. However, with good output biasing, a linear input stage is the best thing that you can have. That is why many use complementary differential, either fet or bipolar. It is even more linear than a single diff pair, and has double the forward gain, too, so you can degenerate the input stage even more and still have enough feedback to look good on the spec sheet.
wading in here with trepidation
I'm reiterating stuff mentioned before and said elsewhere, but in an attempt to summarize:
What is the easiest thing to hear? I said elsewhere recently that it was frequency response, but should have preceded that with differences in loudness. On comparisons you must level-match, but it is often not done in demos.
Then there is frequency response, or so-called linear distortion. The mention of the effect of loudspeaker frequency-dependent loading on amplifier outputs is crucial. And of course this complicates the level matching.
Then there are the nonlinear distortions. These generate energy at frequencies other than the input ones, whether it be harmonic distortion, IM distortion, and species of the latter ones that include crossover distortion and clipping distortion. And the clipping distortion effects need to include the short-term history of the events and the degree to which recovery from such overloads is swift and trouble-free. With sufficiently nasty behavior you can even get period doubling energy. Loudspeakers can do this on their own too.
Within the mechanisms for the generation of nonlinear distortions, in addition to the basic device transfer functions, we may need to consider thermal distortions arising from device signal-induced self-heating, which have a very messy dependence on time, and distortions arising from voltage-dependent capacitances in active devices when voltage swings are substantial. Both of these can be almost arbitrarily reduced by design. Note that both of them are small effects with vacuum tubes, unless the latter are driven so hard as to momentarily strip away the cathode space charge layer, or because the thermal time constants are so large, and the capacitance changes so small.
Behind all of these considerations are expectation biases, and the general psychoacoustic effect of hearing successive differences as improvements---unless a later presentation is just terrible. B usually sounds better than A.
Then there is the theory of the pleasures of cognitive efficacy, which has yet to be developed, but falls into the category of being pleased to hear identifiable differences that are then confirmed. Very close to the effects box problem, or even fortuitous cancellation of one property by another.
And finally (for now), there is a tendency to believe that the source material is flawless, and of course it is not. I quote a friend who responded to a post I made about a Steely Dan album on Facebook:
"One of the miserable outcomes of the much vaunted DIY movement of the last 10 years is the proliferation of recordings made by morons who figure what's the big whoop, I can do it myself. There is no greater contrast to all the execrably executed recording jobs we hear on the radio daily than the magnificent tracks recorded and mixed by Roger Nichols for all those Steely Dan records. He used very little EQ in the mix, his attitude was let's capture each instrument well in the first place, so he did it all with microphone choice and placement. And what mixes!"
I'm reiterating stuff mentioned before and said elsewhere, but in an attempt to summarize:
What is the easiest thing to hear? I said elsewhere recently that it was frequency response, but should have preceded that with differences in loudness. On comparisons you must level-match, but it is often not done in demos.
Then there is frequency response, or so-called linear distortion. The mention of the effect of loudspeaker frequency-dependent loading on amplifier outputs is crucial. And of course this complicates the level matching.
Then there are the nonlinear distortions. These generate energy at frequencies other than the input ones, whether it be harmonic distortion, IM distortion, and species of the latter ones that include crossover distortion and clipping distortion. And the clipping distortion effects need to include the short-term history of the events and the degree to which recovery from such overloads is swift and trouble-free. With sufficiently nasty behavior you can even get period doubling energy. Loudspeakers can do this on their own too.
Within the mechanisms for the generation of nonlinear distortions, in addition to the basic device transfer functions, we may need to consider thermal distortions arising from device signal-induced self-heating, which have a very messy dependence on time, and distortions arising from voltage-dependent capacitances in active devices when voltage swings are substantial. Both of these can be almost arbitrarily reduced by design. Note that both of them are small effects with vacuum tubes, unless the latter are driven so hard as to momentarily strip away the cathode space charge layer, or because the thermal time constants are so large, and the capacitance changes so small.
Behind all of these considerations are expectation biases, and the general psychoacoustic effect of hearing successive differences as improvements---unless a later presentation is just terrible. B usually sounds better than A.
Then there is the theory of the pleasures of cognitive efficacy, which has yet to be developed, but falls into the category of being pleased to hear identifiable differences that are then confirmed. Very close to the effects box problem, or even fortuitous cancellation of one property by another.
And finally (for now), there is a tendency to believe that the source material is flawless, and of course it is not. I quote a friend who responded to a post I made about a Steely Dan album on Facebook:
"One of the miserable outcomes of the much vaunted DIY movement of the last 10 years is the proliferation of recordings made by morons who figure what's the big whoop, I can do it myself. There is no greater contrast to all the execrably executed recording jobs we hear on the radio daily than the magnificent tracks recorded and mixed by Roger Nichols for all those Steely Dan records. He used very little EQ in the mix, his attitude was let's capture each instrument well in the first place, so he did it all with microphone choice and placement. And what mixes!"
I personally don't like feedback as I feel it 'violates' a fundamental rule of human hearing; that being that the ear/brain system uses higher ordered harmonics to determine sound pressure. So if these harmonics are altered by the amplification (or made unmaskable by the amplification) the result will be that the music has a loudness to it that is not part of the original signal and a harshness or hardness will also be imposed.
But isn't it the other way around? If an amplifier distorts, it upsets the harmonic structure of the original making it sound different than it should, if your reasoning is correct.
If feedback prevents this, and helps to preserve the original harmonic structure through the equipment, isn't that what you want?
Jan
Since you ask, I am pretty certain we would categorize Krells as being in
the moderate feedback camp.
OK, point taken, I will next time insert a REAL high feedback example amp
Jan
ESLs are a leading rather than lagging load. But I can't see anything else that makes sense there.
People often think I am making this up. Google 'Fisher 55-A' and you will see an amplifier with a variable feedback control known as a 'damping control'. It has voltage feedback at one extreme and current feedback at the other, both canceling in the middle. If you can get a good look at it you will see that those settings are marked 'constant voltage', 'constant current' and 'constant power' respectively. This amplifier was built when the industry was in transition from the power paradigm to the voltage paradigm.
ESLs have an impedance curve based on a capacitor rather than a driver in a box. As a result the impedance curve is not also an efficiency curve (which is the foundation of the voltage drive approach). So amps that can make constant voltage on an ESL tend to sound too bright and thin in the bass.
If you look into history you will see that EV and Mac led the movement to go from power drive to voltage drive, in the late 1950s and into the 1960s (it is they who established the voltage paradigm rules). Essentially though if you look at amplifiers with a fairly high output impedance such as an SET, which also does not employ negative feedback, you have to ask yourself how such an amplifier can be expected to get flat frequency response; the answer is that speakers were and are made to do just that. And those speakers often will not have flat frequency response if used with an amplifier that has a low output impedance and is capable of acting as a voltage source.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
- Home
- Amplifiers
- Solid State
- Global Feedback - A huge benefit for audio