Bob Cordell Interview: Error Correction

Re: Re: Re: Re: Re: EC Amp Topology

analog_guy said:
Edmond,

That sounds about right. I initially tried implementing a capacitive rolloff scheme.

Did you use an ideal opamp model or not? I found when I substituted an HSPICE macromodel for the EL2045 (available on the Intersil website) that I began to see stability problems at high amplitudes and edge rates. When simulating with a square wave take a close look at the edges to make sure there is no oscillation. It was the onset of stability (presumably caused by the current drive limitations of the EL2045) that prompted me to implement HF rolloff with the inductor.

Jeff

Hi Jeff,

I've spiced your circuit again, this time with the macro model for the EL2045. However, I didn't observe any stability problems when using the capacitor compensation (instead of inductor-wise). Probably due to the fact that I'm using a different output stage (2SA1407/2SC3601 and 2SJ201/2SK1530).

But I did observe quite a large difference in the required output current of the op-amp: at around 10MHz twice as much. So, using a coil is definitely a better solution to prevent the op-amp from running out of steam.

Cheers, Edmond.
 
darkfenriz said:
Hi Edmond

Well, maybe EC is mathematically equivalent, but there is an interesting feature of error correction and not for feedback loop:

Hi Adam,

If both techniques are mathematically equivalent, how on earth can one of them have an 'interesting feature'?


Error correction works as required around fairly linear and well defined stages, actually correction-efficiency is pretty proportional to initial linearity,

Does this mean that you adhere to the third view on HEC, that is, the magic of an adaptive feedback factor?
See Bob's summary of the three views:
http://www.diyaudio.com/forums/showthread.php?postid=1309536#post1309536


feedback in cotrary can linearize circuits that are terribly non-linear and even time variant.

Consequently, EC becomes a correction beast as the circuit is more linear initially (read: has more local feedback). That's why I think EC works best as a turbo-boost to feedback.

I'm afraid that I didn't get your point. If NFB can linearize a 'terribly' non-linear thing why shouldn't HEC be able do the same job?

Also, why can't NFB, just like HEC locally applied to the output stage, work as a 'turbo-boost' to (global) feedback?


All in all, any equivalence between EC and feedback is not very appealing to me, understanding EC as a different, alternative technique is damn appealing. If dr. Hawksford had given his article a name: "M.O.J. Hawksford: Well, just a bit different form of feedback, probably equivalent to tradinional one" would anyone ever get intersting in it ? 😉 😉

Best regards
Adam

Probably not. But I think he has cheated us a little bit by the title of his article: 'Distortion correction in Audio Power Amplifiers', (italics by me) while in fact his technique involves just error feedback and feedback is and stays feedback, no matter how it is implemented.

The only (minor) difference between HEC and NFB is that HEC is able to cancel the distortion completely, that is, at only one particular frequency (normally 0Hz), while NFB can only reduce distortion. In practice however, this difference vanishes almost completely due to other restrictions like limited bandwidth etc.

Also see:
J. Vanderkooy and S.P. Liphsitz, "Feedforward Error Correction in Power Amplifiers", JAES, Vol. 28, pp. 2-16, Jan./Feb. 1980.

D. Danyuk and G. Pilko, "Error Correction in Audio Amplifiers", JAES, vol. 44, pp. 721-728, Sept. 1996.

Cheers, Edmond.
 
Edmond Stuart said:


Hi Adam,

If both techniques are mathematically equivalent, how on earth can one of them have an 'interesting feature'?




Does this mean that you adhere to the third view on HEC, that is, the magic of an adaptive feedback factor?
See Bob's summary of the three views:
http://www.diyaudio.com/forums/showthread.php?postid=1309536#post1309536




I'm afraid that I didn't get your point. If NFB can linearize a 'terribly' non-linear thing why shouldn't HEC be able do the same job?

Also, why can't NFB, just like HEC locally applied to the output stage, work as a 'turbo-boost' to (global) feedback?




Probably not. But I think he has cheated us a little bit by the title of his article: 'Distortion correction in Audio Power Amplifiers', (italics by me) while in fact his technique involves just error feedback and feedback is and stays feedback, no matter how it is implemented.

The only (minor) difference between HEC and NFB is that HEC is able to cancel the distortion completely, that is, at only one particular frequency (normally 0Hz), while NFB can only reduce distortion. In practice however, this difference vanishes almost completely due to other restrictions like limited bandwidth etc.

Also see:
J. Vanderkooy and S.P. Liphsitz, "Feedforward Error Correction in Power Amplifiers", JAES, Vol. 28, pp. 2-16, Jan./Feb. 1980.

D. Danyuk and G. Pilko, "Error Correction in Audio Amplifiers", JAES, vol. 44, pp. 721-728, Sept. 1996.

Cheers, Edmond.


Hi Edmond,

Boy, you must really hate Bob Carver, who is famous for coming up with neat-sounding, imagination-stimulating names for good, but straightforward techniques. I remember his "auto-correlator" noise reduction system, for example. His "magnetic field amplifier" was another good one.

I still say that Hawksford's use of the term correction was a good thing, in that it stimulated our thinking and imagination to go beyond the plain old NFB thinking inside the box.

I think his work was also around the time of the buzz about the Quad amplifier with feedforward error correction. Although you might argue that that was a "legitimate" use of the term correction, in practice its performance was actually quite inferior to what Hawksford's arrangement yielded.

Indeed, for practical reasons, virtually every form of feedforward error correction is only able to "cancel the distortion, that is, at only one particular frequency".

Feedback may always be feedback, but some approaches to feedback have advantages in the details of how they are implemented. Hawksford's error correction is one such example. Your more conventional local feedback approach in the output stage also has some similar advantages that lie in the details (e.g., tight loop, stable to quite wide bandwidths, etc.).

Cheers,
Bob
 
Edmond Stuart said:
If both techniques are mathematically equivalent, how on earth can one of them have an 'interesting feature'?

Simply, electronics is not mathematics.😉
If both techniques are mathematically equivalent, how on earth can one of them have an 'interesting feature'?
Not, realy, but this view is intuitional, that's nearly as important as being valid.
I'm afraid that I didn't get your point. If NFB can linearize a 'terribly' non-linear thing why shouldn't HEC be able do the same job?
EC around a buffer assumes it is a buffer i.e. gain is close to 1 at every point. NFB can make a buffer from a circuit of any high enough gain, linear or not.

Also, why can't NFB, just like HEC locally applied to the output stage, work as a 'turbo-boost' to (global) feedback?
That would be nesting, 'turbo-boosting' is what I call the fact, that feedback already extracts error signal and you can simply take and subtract it in earlier stage. (do you ramember my topology?)
Also see:
J. Vanderkooy and S.P. Liphsitz, "Feedforward Error Correction in Power Amplifiers", JAES, Vol. 28, pp. 2-16, Jan./Feb. 1980.

D. Danyuk and G. Pilko, "Error Correction in Audio Amplifiers", JAES, vol. 44, pp. 721-728, Sept. 1996.
Have read the first, do you have an electronic copy of the second?

Best regards,
Adam
 
fairy tales

Bob Cordell said:
Hi Edmond,

Boy, you must really hate Bob Carver, who is famous for coming up with neat-sounding, imagination-stimulating names for good, but straightforward techniques. I remember his "auto-correlator" noise reduction system, for example. His "magnetic field amplifier" was another good one.

Hi Bob,

Quite the opposite. I like that guy just as I like Hawksford. He (the latter) is just doing what he is paid for: writing articles that draw as much attention as possible, which, of course, also applies to the title itself.

As Adam pointed out, his article wouldn't attract that much attention if he had chosen a more applicable but less striking title.

Anyhow, in the summary he called it (happily) a "A modified feedback technique", which is more to the point.


I still say that Hawksford's use of the term correction was a good thing, in that it stimulated our thinking and imagination to go beyond the plain old NFB thinking inside the box.

I disagree, see below.


I think his work was also around the time of the buzz about the Quad amplifier with feedforward error correction. Although you might argue that that was a "legitimate" use of the term correction, in practice its performance was actually quite inferior to what Hawksford's arrangement yielded.

Not just buz. Peter Walker (I met him once, nice guy) published his article in WW six years before!
And Hawksford referred to it.


Indeed, for practical reasons, virtually every form of feedforward error correction

feedforward? feedback I suppose.


is only able to "cancel the distortion, that is, at only one particular frequency".

Feedback may always be feedback, but some approaches to feedback have advantages in the details of how they are implemented. Hawksford's error correction is one such example. Your more conventional local feedback approach in the output stage also has some similar advantages that lie in the details (e.g., tight loop, stable to quite wide bandwidths, etc.).

Cheers,
Bob

The error correction view has pushed the thinking on HEC in the wrong direction, resulting in fairy tales like "adaptive feedback" or "feedback on demand". The major contributor to this kind of thinking is probably Hawksford himself:

"The technique should find favor among designers who adhere to the low-feedback school of designs, as corrective feedback is only applied when distortion in the output is generated."

How naive can one be to get misled by his carefully chosen words?
Does Hawksford mean that there is no feedback if no distortion is generated in the output stage? wrong!
He is only saying that is there no feedback signal:

"If, therefore, the output stage... is designed to be as linear as possible,... then only minimal error signals results"

However, as Brian has also explained, the feedback circuit is still present and alert. With or without distortion of the output stage, the NFB factor (or NFB loop gain if you like) is still the same, a lot of NFB, over 40dB in a well balanced circuit.

As we all know, feedback systems, thus HEC too, need some form of frequency compensation, but Hawksford never breathed a word of it. Why? Is he deliberately misleading us by concealing something?

As for Bod Carver, this is really an amusing story:
http://www.carveraudio.com/CarverChallenge.pdf

Cheers, Edmond.
 
darkfenriz said:
[snip]
That would be nesting, 'turbo-boosting' is what I call the fact, that feedback already extracts error signal and you can simply take and subtract it in earlier stage. (do you remember my topology?)
[snip]
Best regards,
Adam

Hi Adam,

Ah, now I get it! You are referring to your little gem (for example), the unbeatable 10 transistors amp with less than 10ppm distortion.
This wasn't obvious to me right from the beginning, as the previous discussion (with Bob et al) was narrowed to only output stages.

Cheers, Edmond.
 
Re: fairy tales

Edmond Stuart said:


Hi Bob,

Quite the opposite. I like that guy just as I like Hawksford. He (the latter) is just doing what he is paid for: writing articles that draw as much attention as possible, which, of course, also applies to the title itself.

As Adam pointed out, his article wouldn't attract that much attention if he had chosen a more applicable but less striking title.

Anyhow, in the summary he called it (happily) a "A modified feedback technique", which is more to the point.




I disagree, see below.




Not just buz. Peter Walker (I met him once, nice guy) published his article in WW six years before!
And Hawksford referred to it.




feedforward? feedback I suppose.




The error correction view has pushed the thinking on HEC in the wrong direction, resulting in fairy tales like "adaptive feedback" or "feedback on demand". The major contributor to this kind of thinking is probably Hawksford himself:

"The technique should find favor among designers who adhere to the low-feedback school of designs, as corrective feedback is only applied when distortion in the output is generated."

How naive can one be to get misled by his carefully chosen words?
Does Hawksford mean that there is no feedback if no distortion is generated in the output stage? wrong!
He is only saying that is there no feedback signal:

"If, therefore, the output stage... is designed to be as linear as possible,... then only minimal error signals results"

However, as Brian has also explained, the feedback circuit is still present and alert. With or without distortion of the output stage, the NFB factor (or NFB loop gain if you like) is still the same, a lot of NFB, over 40dB in a well balanced circuit.

As we all know, feedback systems, thus HEC too, need some form of frequency compensation, but Hawksford never breathed a word of it. Why? Is he deliberately misleading us by concealing something?

As for Bod Carver, this is really an amusing story:
http://www.carveraudio.com/CarverChallenge.pdf

Cheers, Edmond.


Hawksford has done a great deal of good for the audio community, and his error correction scheme was an outstanding contribution. Its a shame you can't bring yourself to cut him some slack.

I don't share your rather cynical view of Hawksford's work. We are going to have to just agree to disagree on this issue. I will therefore consider myself to be in very good company when you don't cut me any slack. None of us is perfect.

I believe that I was the first to state the need for compensation of HEC, and the first to implement it. Nevertheless, I never interpreted his failure to mention it as anything deliberate.

In my earlier comment that you referred to, I did definitely mean to say feedforward; there was no mistake.

Cheers,
Bob
 
Hawksford has done a great deal of good for the audio community, and his error correction scheme was an outstanding contribution. Its a shame you can't bring yourself to cut him some slack. I don't share your rather cynical view of Hawksford's work. We are going to have to just agree to disagree on this issue. I will therefore consider myself to be in very good company when you don't cut me any slack. None of us is perfect.

Bob, this is the reflection of your views of Otala's, Curl's and Jung's work. How do you feel yourself in their shoes?
 
dimitri said:


Bob, this is the reflection of your views of Otala's, Curl's and Jung's work. How do you feel yourself in their shoes?


Hi Dimitri,

I've mellowed 🙂.

Seriously, this is a fair question. My original paper on Otala, "Another View of TIM" pretty much characterizes the way that I wrote about Otala's work. If you haven't read it, it is on my web site at www.cordellaudio.com under publications. I think if you read it you will find it is a fair and technical analysis of the TIM work of Otala.

I did not call him or his work a fraud or a hoax, and I did not attribute dishonesty to his intentions. He was just wrong in the conclusions he drew. He was also extremely stubborn in sticking to his original misguided conclusions about negative feedback. I have seen no such stubbonness in Hawksford.

Otala's work did force many in the audio community to look much harder at the issues he exposed, even if those efforts were in straightening out those aspects of his work that were wrong. For that I give him some credit. He did, however, in my opinion, mislead an entire generation of audio designers into wrongly thinking that NFB was a bad thing even when well-executed.

Cheers,
Bob
 
Re: Re: fairy tales

Bob Cordell said:
Hawksford has done a great deal of good for the audio community, and his error correction scheme was an outstanding contribution. Its a shame you can't bring yourself to cut him some slack.

I don't share your rather cynical view of Hawksford's work. We are going to have to just agree to disagree on this issue. I will therefore consider myself to be in very good company when you don't cut me any slack. None of us is perfect.

I believe that I was the first to state the need for compensation of HEC, and the first to implement it. Nevertheless, I never interpreted his failure to mention it as anything deliberate.

In my earlier comment that you referred to, I did definitely mean to say feedforward; there was no mistake.

Cheers,
Bob

Reading carefully somebody's else comment is not your forte, is it?
I said I like Hawksford.

>Cynical?

Perhaps you don't know, but in academic circles, one is supposed to write papers, the more the better and the rank on the citation index does matter, the higher the better. I'm cynical about this common practice, not about Hawksford's work.

I'm only critical about the way he has presented his paper. Is that forbidden? And why I'm critical? Because some people adhere to the nonsensical "adaptive feedback" or "feedback on demand" view on HEC. So I asked myself in how much his paper has contributed to that view. Is that forbidden too?

>Compensation?

Isn't it quite common to pay attention to the Nyquist stability in all instances that involve NFB? Not so with Hawksford and I consider this a serious omission, as the naive reader of his paper might conclude that NFB is not involved and compensation is not required.

Happily he complemented his paper with a stability analysis in a later publication.

>feed forward, no mistake?

For two reasons wrong.
1. It is in principle possible to completely null all errors due to the main amplifier (see Vanderkooy et al on error feed forward), so not for only one frequency or DC, which applies to error feedback
2. You wrote: "Indeed,.....", meaning that you were referring to and agreeing with me on that particular point about HEC, which is error feedback.

Cheers, Edmond.
 
Re: Re: Re: fairy tales

Edmond Stuart said:


Reading carefully somebody's else comment is not your forte, is it?
I said I like Hawksford.

>Cynical?

Perhaps you don't know, but in academic circles, one is supposed to write papers, the more the better and the rank on the citation index does matter, the higher the better. I'm cynical about this common practice, not about Hawksford's work.

I'm only critical about the way he has presented his paper. Is that forbidden? And why I'm critical? Because some people adhere to the nonsensical "adaptive feedback" or "feedback on demand" view on HEC. So I asked myself in how much his paper has contributed to that view. Is that forbidden too?

>Compensation?

Isn't it quite common to pay attention to the Nyquist stability in all instances that involve NFB? Not so with Hawksford and I consider this a serious omission, as the naive reader of his paper might conclude that NFB is not involved and compensation is not required.

Happily he complemented his paper with a stability analysis in a later publication.

>feed forward, no mistake?

For two reasons wrong.
1. It is in principle possible to completely null all errors due to the main amplifier (see Vanderkooy et al on error feed forward), so not for only one frequency or DC, which applies to error feedback
2. You wrote: "Indeed,.....", meaning that you were referring to and agreeing with me on that particular point about HEC, which is error feedback.

Cheers, Edmond.


We agree to disagree; no more silly semantical games; the technical value of this disagreement has gone to zero, if not beyond.

Cheers,
Bob
 
janneman said:
I wish to recognize Mr. Cordell's almost heroic efforts to keep this thread afloat. Alas, it wasn't to be.

Jan Didden


Thanks, guys, but I hope you did not interpret my earlier remarks to mean that discussion of error correction should be ended.

I only meant that that part of the discussion focusing on characterizing the nature of Hawksford's contributions and arguing about whether HEC should be viewed only as a form of negative feedback has reached the point where Edmond and I should just agree to disagree and move forward.

Edmond has made a real contribution to the vitality of this thread in presenting a more conventional local output stage NFB approach that should be considered as competitive with HEC.

Cheers,
Bob
 
Jan,

I second your comment; all praise to Bob, who has always remained fair, unbiased, and more than personable.

Bob,

Could I ask a simple question, from a guy who sells commercial amps of vastly simpler design than anything discussed here?

If we concentrate EC techniques on the output stage we are in fact preventing waveform compression on a unity gain stage. This is particularly significant on high level signals as they approach the rail, of course, and laudable.

However, if we assume that in a DEF output stage the Vbe increases to say 2V at max signal current and the drop across say a 0.22R emitter resistor at 6A is 1.32V, then the total compression is 3.32V maximum for an output of say 45V peak (assuming 50V rails). This implies an input signal off the VAS of around 48.32V, and thus gives an overall gain at peak current of 0.93. EC corrects this drop in gain to unity.

However, ordinarily, this slight loss of gain would be compensated by the global negative feedback loop, and in truth this compensation would be a walk in the park. Of far more significance would be variations in feedback factor caused by a high output impedance voltage amplifier as it struggles to negotiate the highly variable Zin of a DEF output stage. (I assume here a regular, conventional DEF, with which many may disagree, but for the purposes of the argument it's useful.)

Insofar as the EC output stage presents a far more uniform impedance to the VAS, I believe this is where you will obtain the majority of your benefits rather than the prevention of waveform compression at high signal levels.

Would you care to comment on this? Gospel, or garbage?

Cheers,

Hugh
 
AKSA said:
[snip]
However, ordinarily, this slight loss of gain would be compensated by the global negative feedback loop, and in truth this compensation would be a walk in the park. Of far more significance would be variations in feedback factor caused by a high output impedance voltage amplifier as it struggles to negotiate the highly variable Zin of a DEF output stage. (I assume here a regular, conventional DEF, with which many may disagree, but for the purposes of the argument it's useful.)

Insofar as the EC output stage presents a far more uniform impedance to the VAS, I believe this is where you will obtain the majority of your benefits rather than the prevention of waveform compression at high signal levels.

Would you care to comment on this? Gospel, or garbage?

Cheers,

Hugh

Hi Hugh,

I have spiced a typical amp (Self's blameless with a DEF OPS) with/without an ideal EC circuit and with/without an ideal buffer between the VAS and OPS. Here are the THD20 results:

original: 137ppm
+ buffer: 136ppm
+ EC: 5.1ppm
+ EC + buffer: 4.4ppm

I hope this will answer your question.

Cheers, Edmond.