Bob Cordell Interview: Negative Feedback

[lumanauw]

Mr. Cordell, is there any advantage or disadvantage to have the same amount of feedback for low frequencies (like50hz) and the high frequencies (like 20khz)?
Usually this means to make the OL curve as flat as possible from low to high frequencies. A trick to do this is to put heavy local feedback (emitor resistor) in transistors from differential pair upwards.

 

[jcx]

covered somewhat earlier in this thread

try ~ p35 +/-
http://www.diyaudio.com/forums/showthread.php?postid=1052464#post1052464

although touched on before and after for a few pages at a time

this thread is long enough, do we really need to go in circles?

 

[Nixie]

Originally posted by jcx
do we really need to go in circles?

Welcome to diyaudio!

Actually, this forum would benefit greatly from the ability to search within specific threads.

 

[Bob Cordell]

Mr. Cordell, is there any advantage or disadvantage to have the same amount of feedback for low frequencies (like50hz) and the high frequencies (like 20khz)?
Usually this means to make the OL curve as flat as possible from low to high frequencies. A trick to do this is to put heavy local feedback (emitor resistor) in transistors from differential pair upwards.

There are some who advocate wide open-loop bandwidth, and there are also those who advocate low or no negative feedback. The amount of feedback that you can have at 20 kHz is essentially limited by stability considerations. With a typical 6 dB/octave open-loop gain roll off, an amplifier with 20 dB of NFB at 20 kHz will have to be stable out to a gain crossover frequency of 200 kHz. An amplifier with 40 dB of NFB at 20 kHz will have to be stable out to a gain crossover frequency of 2 MHz, which is more difficult to achieve. Most non-low-feedback solid state amplifier designs have between 20 dB and 40 dB of NFB at 20 kHz.

Once you establish how much NFB you have at 20 kHz, then whether you have more feedback or about the same amount of feedback at lower frequencies is a matter largely of philosophy: i.e., whether or not you want to deprive the lower frequencies of the greater amount of NFB that is naturally available in order to get a high open-loop bandwidth or not.

Much confusion originated here with Otala's mistaken assertion that one needed high open loop bandwidth to achieve low TIM (e.g., high slew rate). All else remaining equal, high open loop bandwidth has nothing to do with achievable slew rate or lower HF distortion.

So it all ends up being the philosophical choice of the designer, just like the choice of a designer as to whether he wants a no-feedback design, a design with modest feedback, or a design with high feedback. Since if there is feedback, it is often a function of frequency, one must be clear about what they mean when they say high or low feedback. Are they talking about NFB at 20 kHz, or are they talking about NFB at 1 kHz.

Bob

 

[ingrast]

Bob:

I cannot completely agree with some points in your previous post.

While I recognize some designers are willing to choose how much feedback to apply, this does not really make sense "per se" unless qualified by some other type of consideration.

What we are seeking to improve is overall performance, and this is the bottom line, no matter the means employed.

If a certain design strives for the largest possible gain so as to apply the maximum amount of correction, but in doing so introduces system nonlinearities that eat up most of that resource, it will certainly perform worse than an alternative design where native linearity is carefully adressed, something that may imply a tradeoff in gain and thus in available negative feedback.

My point is the amount of negative feedback in isolation is not an issue in itself, but should be the result of an optimum balance between native nonlinearity - unavoidable and frequently worse with higher open loop gain - and correction.

A further consideration is that large amounts of negative feedback applied to a questionable basic amplifier, unavoidably leads to a compound forest of multiple low level IM products, the type of syndrome frequently attributed to amplifiers that measure very well in single tone tests yet perform very bad at auditioning as compared with other designs of inferior measured performance. This is particularly true with the first solid state designs of 20-30 years back.

Rodolfo

 

[Nixie]

What we are seeking to improve is overall performance

But that has not even been properly defined here. I remind you there is yet to be formulated a metric that correlates perfectly with blind testing results of perception of distortion. Thus, every time you make a comparison, such as when you use the word "worse" later in your post, you have some benchmark in mind which, by not being made explicit, is not questioned as to its validity.

 

[Bob Cordell]

Bob:

I cannot completely agree with some points in your previous post.

While I recognize some designers are willing to choose how much feedback to apply, this does not really make sense "per se" unless qualified by some other type of consideration.

What we are seeking to improve is overall performance, and this is the bottom line, no matter the means employed.

If a certain design strives for the largest possible gain so as to apply the maximum amount of correction, but in doing so introduces system nonlinearities that eat up most of that resource, it will certainly perform worse than an alternative design where native linearity is carefully adressed, something that may imply a tradeoff in gain and thus in available negative feedback.

My point is the amount of negative feedback in isolation is not an issue in itself, but should be the result of an optimum balance between native nonlinearity - unavoidable and frequently worse with higher open loop gain - and correction.

A further consideration is that large amounts of negative feedback applied to a questionable basic amplifier, unavoidably leads to a compound forest of multiple low level IM products, the type of syndrome frequently attributed to amplifiers that measure very well in single tone tests yet perform very bad at auditioning as compared with other designs of inferior measured performance. This is particularly true with the first solid state designs of 20-30 years back.

Rodolfo

I don't think you are disagreeing with any of my points, but rather elaborating on them or adding the usual necessary caveats. Of course we should not sacrifice open loop linearity just to increase NFB. Unfortunately, in some cases the opposite is true; in some misguided designs, people load the VAS in order to get wide open-loop bandwidth, and in so doing make the VAS work harder, actually resulting in more open loop nonlinearity. It is generally a misconception when people think that one somehow has to strain to get open loop gain; it is not hard to get with great linearity. Indeed, in my amplifiers, I start with an open loop amplifier that is more linear before I apply NFB than many no-feedback practitioners do. I think you are preaching to the choir.

Cheers,
Bob

 

[ingrast]

I don't think you are disagreeing with any of my points, ....

Apologies Bob, you are right and I may have been careless in my wording. What struck me was your:

.....
Once you establish how much NFB you have at 20 kHz, then whether you have more feedback or about the same amount of feedback at lower frequencies is a matter largely of philosophy....

Which I interpreted as if it were a relevant designer's choice by itself to select such or such amount of feedback without reference to other contextual variables.

On second reading it is apparent that was not what you meant.

Rodolfo

 

[forr]

Ingrast
---large amounts of negative feedback applied to a questionable basic amplifier, unavoidably leads to a compound forest of multiple low level IM products---

This seems to be in disagreement with the Baxandall papers on amplifier design.

 

[ingrast]

But that has not even been properly defined here. I remind you there is yet to be formulated a metric that correlates perfectly with blind testing results of perception of distortion. Thus, every time you make a comparison, such as when you use the word "worse" later in your post, you have some benchmark in mind which, by not being made explicit, is not questioned as to its validity.

While I agree there is not a widely **accepted** - and I want to remark the "accepted" part - objective metrics, it is also true only objective metrics can be used as unquestionable comparison references, in the strict sense that nobody will question that 2+2=4.

These objective metrics are what I implicitely refer to as performance results, or better - worse evaluation results.

If subjective evaluation is allowed to figure in, then there is no hope of reaching consensus, for it involves personal preferences, expectatives, experience, natural aptitudes and so forth.

Rodolfo

 

[ingrast]

Ingrast
---large amounts of negative feedback applied to a questionable basic amplifier, unavoidably leads to a compound forest of multiple low level IM products---

This seems to be in disagreement with the Baxandall papers on amplifier design.

I am not aware of the papers you mention, but it has been widely shown and can be easily simulated, that significant nonlinearities in the forward path, give rise iteratively under close loop conditions to large amounts of low level spurious components. Some people use to refer to a "veil" probably in connect¡on with this.

This nonwithstanding, it is obvious that if spurious products are really low, i.e. below 100 dB or better dynamic range, chances are they will neither be of consequence.
If a sufficiently high open loop gain can be attained within practical stability constraints to achieve this level of performance, then the native nonlinearities can be considered to be adequately tamed, and may be this is what your mentioned papers made reference to.

Rodolfo

 

[lumanauw]

A further consideration is that large amounts of negative feedback applied to a questionable basic amplifier, unavoidably leads to a compound forest of multiple low level IM products, the type of syndrome frequently attributed to amplifiers that measure very well in single tone tests yet perform very bad at auditioning as compared with other designs of inferior measured performance. This is particularly true with the first solid state designs of 20-30 years back.

This is very likely if this high global feedback poweramp is made with bad output stage. The problematic section is output stage, but the troubleshooting is made in differential pair+VAS, building huge OL gain in these front stage. It's like a car having problem in the tyres, but you keep fixing the steering wheel. Not shooting the real problem.
Better to have good output stage first (like using classA or using EC) and then using suitable gain front stage than building bad output stage with enormous front end gain.

 

[darkfenriz]

... in some misguided designs, people load the VAS in order to get wide open-loop bandwidth, and in so doing make the VAS work harder, actually resulting in more open loop nonlinearity. It is generally a misconception when people think that one somehow has to strain to get open loop gain; it is not hard to get with great linearity.

High open loop bandwidth often means also well-defined open loop bandwidth and well-defined pole frequencies, which is helpful in reducing dynamic distortion mechanisms, PIM included.

About VAS loading:
D. Self uses darlington output stage, so output stage input impedance is probably around 50k-100kOhm with 8ohm load and good Hfe trasistors. This impedance is as we all know terribly non-linear with reflecting crossover mechanisms, capacitive and modulated by speaker impedance, right?
Now I use tripple darlington output stage based on high Hfe transistors (2n5401/2n5551, 2sa1837/2sc4793, 2sa1943/2sc5200), so input impedance of output stage is around 10Mohm probably with all those nonlinear mechanisms, but is paralleled with 67kohm most linear electrical device around.
Which one would you prefer, input impedance being of comparable value?

regards
Adam

 

[Bob Cordell]

High open loop bandwidth often means also well-defined open loop bandwidth and well-defined pole frequencies, which is helpful in reducing dynamic distortion mechanisms, PIM included.

About VAS loading:
D. Self uses darlington output stage, so output stage input impedance is probably around 50k-100kOhm with 8ohm load and good Hfe trasistors. This impedance is as we all know terribly non-linear with reflecting crossover mechanisms, capacitive and modulated by speaker impedance, right?
Now I use tripple darlington output stage based on high Hfe transistors (2n5401/2n5551, 2sa1837/2sc4793, 2sa1943/2sc5200), so input impedance of output stage is around 10Mohm probably with all those nonlinear mechanisms, but is paralleled with 67kohm most linear electrical device around.
Which one would you prefer, input impedance being of comparable value?

regards
Adam

It has been shown that having wide open loop bandwidth does not help PIM and related nonlinearities. It all comes out in the wash. I know this may seem non-intuitive. The less-well-defined pole has movement that has less effect on the result. A pole at 100 Hz that moves around 10% will have about the same effect as a pole at 10 kHz that moves around 0.1%. Putting in a shunt resistance that moves the pole from 100 Hz to 10 kHz and at the same time reduces pole uncertainty from 10% to 0.1% has no net effect. The underlying effect causing pole shift, e.g., nonlinear capacitance, has not changed, and its effect on propogation delay through the circuit has not changed by adding shunt resistance. SPICE it, you'll see.

A simple Darlinton output stage is completely inadequate for reasons of the nonlinear resistance you cite, at the very least. With a triple, this issue pretty much goes away. I'd go with the triple, with a VAS that is not artifically loaded by a resistor. The addition of that resistor will never decrease distortion; it will only increase it.

Bob

 

[john curl]

Actually, I once found a measureable example of adding a bandwidth widening resistor and having LOWER overall distortion. I attibute it to lowering the drive inpedance to the transistor follower output stage.
For the record, Bob's opinions directly contradict the opinions of Matti Otala and Charles Hansen. I'll stick with Matti, as I have done in the past, on these issues.

 

[darkfenriz]

Same example has been shown by MikeB on this forum. I also SPICED the fact, that sometimes adding resistive loading to VAS decreases distortion by several percent or so, not to mention, that it sounds good to me.

 

[mikeks]

Actually, I once found a measureable example of adding a bandwidth widening resistor and having LOWER overall distortion. I attibute it to lowering the drive inpedance to the transistor follower output stage.
For the record, Bob's opinions directly contradict the opinions of Matti Otala and Charles Hansen. I'll stick with Matti, as I have done in the past, on these issues.

Clearly Matti Otala did audio a great disservice with his wilful obscurantism.

 

[Bob Cordell]

Actually, I once found a measureable example of adding a bandwidth widening resistor and having LOWER overall distortion. I attibute it to lowering the drive inpedance to the transistor follower output stage.
For the record, Bob's opinions directly contradict the opinions of Matti Otala and Charles Hansen. I'll stick with Matti, as I have done in the past, on these issues.

Yes, John, my opinions do directly contradict those of Matti Otala. He misled a thousand audio engineers. Whether it was willful on his part, or due to a lack of experimental and scientific discipline, or something else, we may never know.

However, that does not mean that his work was without value. He deserves credit for casting a spotlight on some things that really needed examination, even if his proposed solutions were completely misguided. The technical discussions and arguments that took place as a result of his work did have worthwhile outcomes. He also deserves credit for at least putting forth a scientific postulation for his theories, and for proposing a measurement technique for measuring the effects he claimed to be curing (e.g., TIM, DIM). However, his very own objective measurement techniques were what provided the basis for disproving his "cures" for the problem.

It is very easy, using Matti's very own DIM test, to show that, given the same amount of HF NFB above 20 kHz, a wide open loop bandwidth does not improve slew rate and does not reduce measured TIM.

What he did, which was his own un-doing, is much more than we get from some audio folks today who postulate a bunch of pseudo-scientific cr@p and don't back it up with any means of proving it one way or the other. Half the oil in our community is snake oil; the hard part is distinguishing which is which.

I know this area is one in which we will continue to respectfully agree to disagree.

Cheers,
Bob

 

[john curl]

Who cares about more slew rate, IF you have virtually no TIM, already? THEN, other factors 'might' become more important.
Bob, you are stuck on 'TIM' and ignoring 'DIM' which Matti had already coined as early as 1976, when I worked with him.
I, too, did NOT like his ignoring slew rate as a number, and his insistence (in the late '70's) that high open loop bandwidth appeared to be necessary, but Walt Jung's and my work on op amps, both as IC's and discrete, tend to go in this direction, when listening to the results.

 

[Nixie]

when listening to the results.

And I trust the results are valid since the comparisons were done with blind testing?