Comments on Audio Power Amplifier Design Handbook by Douglas Self

[Samuel Groner]

Hi

I'm sure that many members of this forum will be familiar with D. Selfs writing, in particular the audio power amplifier design handbook. I finally managed to finish a paper with a couple of complementing and contrasting thoughts on the 5th edition of this classic. I hope you enjoy reading it and appreciate further conversation on the topics covered!

audio_power_amp_design_comments.pdf

Samuel

[Telstar]

Wow, 50 pages!

Edit: it seems a very good read, will post some comments tomorrow.

[Bob Cordell]

Originally Posted by Samuel Groner

Hi

I'm sure that many members of this forum will be familiar with D. Selfs writing, in particular the audio power amplifier design handbook. I finally managed to finish a paper with a couple of complementing and contrasting thoughts on the 5th edition of this classic. I hope you enjoy reading it and appreciate further conversation on the topics covered!

audio_power_amp_design_comments.pdf

Samuel

Hi Samuel,

Thanks for doing this. I'll look forward to reading your comments.

Cheers,
Bob

[SSassen]

Thanks Samuel, that's an heroic effort!

Here's my 2 cents about audio amplifier design books.

What I find lacking in many books about power amplifiers is novel ways of reducing distortion, many just focus on the same mechanisms that have been attributed to being at the root cause of amplifier distortion for the past three decades. The same I see happening here on DIYaudio, the same designs get rehashed over and over again.

In the meantime the folks that design opamps haven't been twiddling their thumbs. Many modern opamps use novel topologies that would be applicable to power amplifiers too, so why not borrow some good ideas from them?

You'll find that among the solutions employed there you'll find key ingredients which can significantly improve the performance of classic topologies. Ingredients which I can't seem to find in many of the books about power amplifier design (this includes Douglas's and Bob's).

Just for the fun of it have a look at the Miller topology. In its basic form it has a few issues, but these can be overcome. Basically there's four major obstacles we need to tackle:

1) PSRR: with Ahuja compensation the PSRR can be raised to >110dB with ease.
2) Slew rate: use degeneration to up the slewrate and prevent it being directly coupled to the GBW.
3) Non linear capacitance parallel to Miller C: simplest fix is to cascode the current junction.
4) Non linear impedance at the input: use a bootstrapped cascode or a folded cascode.

These are all pretty common ingredients in opamp design, but especially bootstrapping and cascoding seem to be heralded as detrimental to an amplifier's sonic performance. As such this speaks volumes about the folks that seem to qualify these as such.

Regardless, it would be nice to see some of these approaches applied to power amplifiers as well, or at least discussed in amplifier design literature.

/steps off soapbox

Cheers,

Sander.

[jcx]

it is hard to be "systematic" when exploring even simple circuit topology alternatives

"design space" is multidimensional and not "simply connected" nor is there any unique global optimum - "stage by stage" optimization may ignore some interactions, "theoretically better" changes in one stage may be totally masked by other stage's limitations leading you to discard an improvement that would be valuable in a more globally oriented optimiztion - or it may remain totally moot with other stage's unavoidable limitations

as an example of rapidly increasing "dimensionallity"
the "enhanced Beta" VAS - should it viewed as a composite transistor? and then do you need to compare all other known 2 transistor composites that can substitute for a single Q: CPF, Cascode, super pair
then you can go to 3 transistor composites - cascode the "Darlington" beta enhanced circuit... and you can see some of the topological complexity you can get into in Hawksford's Slope Distortion paper - what about CFP for any or all of the Q in each version?

how about VAS Q degeneration, with the input buffer - do the combined linearizing, gain increases change the VAS Q type analysis?
(Cherry shows VAS Q emitter resistor as valuable for stability wihtout VAS linearity cost - I really don't understand Self's insistance the the R is "not local negative feedback, only for current limiting")


I'm a little frustrated by the near complete lack of circuit theory based analysis - I like what Dr Cherry has done over his many articles in JAES but it is hard to penetrate - there should be a possible intermediate level of analysis

Cherry's ""Estimates of Nonlinear Distortion in Feedback Amplifiers" paper seems to be overlooked

maybe Middlebrook's "low entropy" equations?

the VAS stage analysis may benefit from even a simple circuit theory model that lets you more readily weigh topology change impacts

[Bob Cordell]

Originally Posted by SSassen

Thanks Samuel, that's an heroic effort!

Here's my 2 cents about audio amplifier design books.

What I find lacking in many books about power amplifiers is novel ways of reducing distortion, many just focus on the same mechanisms that have been attributed to being at the root cause of amplifier distortion for the past three decades. The same I see happening here on DIYaudio, the same designs get rehashed over and over again.

In the meantime the folks that design opamps haven't been twiddling their thumbs. Many modern opamps use novel topologies that would be applicable to power amplifiers too, so why not borrow some good ideas from them?

You'll find that among the solutions employed there you'll find key ingredients which can significantly improve the performance of classic topologies. Ingredients which I can't seem to find in many of the books about power amplifier design (this includes Douglas's and Bob's).

Just for the fun of it have a look at the Miller topology. In its basic form it has a few issues, but these can be overcome. Basically there's four major obstacles we need to tackle:

1) PSRR: with Ahuja compensation the PSRR can be raised to >110dB with ease.
2) Slew rate: use degeneration to up the slewrate and prevent it being directly coupled to the GBW.
3) Non linear capacitance parallel to Miller C: simplest fix is to cascode the current junction.
4) Non linear impedance at the input: use a bootstrapped cascode or a folded cascode.

These are all pretty common ingredients in opamp design, but especially bootstrapping and cascoding seem to be heralded as detrimental to an amplifier's sonic performance. As such this speaks volumes about the folks that seem to qualify these as such.

Regardless, it would be nice to see some of these approaches applied to power amplifiers as well, or at least discussed in amplifier design literature.

/steps off soapbox

Cheers,

Sander.

Hi Sander,

These are good points, but I believe you must have missed them somehow when you read my book.

Although I am not familiar with "Ahuja compensation", I do discuss PSRR issues with Miller compensation that results from one end of the capacitor being referenced to the rail while the other is referenced to signal ground, and mention compensation approaches that avoid this, such as what I call Miller Input Compensation. This is what I used in my MOSFET power amplifier that employed error correction published in 1984.

Of course, my book also prominently covers Hawksford Error Correction (HEC).

The second issue, that of using emitter degeneration to increase slew rate, is also covered extensively in several places in my book. I also discuss the fact that the use of Miller Input Compensation frees the amplifier to have very large slew rate virtually independent of compensation. When used in my MOSFET amplifier, that small 50W amplifier achieved a slew rate of 300 V/us.

The third issue, that of nonlinear collector-base capacitance in parallel with the Miller capacitor, is also covered in my book as a potential problem with simple Miller compensation. I show that the elimination of its effect is accomplished by placing an emitter follower in front of the VAS transistor (sometimes referred to as a Darlington VAS, but not in the Darlington sense where both collectors are tied together).

The fourth issue, using a bootstrapped (or "driven") cascode to reduce distortion at the input stage is also covered in my book in a couple of places.

In fairness to Doug, I think he has covered some of these as well.

I do not recall seeing where anyone asserted that cascoding was detrimental to the sound.

Cheers,
Bob

[SSassen]

Originally Posted by Bob Cordell

Hi Sander,

These are good points, but I believe you must have missed them somehow when you read my book.

Although I am not familiar with "Ahuja compensation", I do discuss PSRR issues with Miller compensation that results from one end of the capacitor being referenced to the rail while the other is referenced to signal ground, and mention compensation approaches that avoid this, such as what I call Miller Input Compensation. This is what I used in my MOSFET power amplifier that employed error correction published in 1984.

Of course, my book also prominently covers Hawksford Error Correction (HEC).

The second issue, that of using emitter degeneration to increase slew rate, is also covered extensively in several places in my book. I also discuss the fact that the use of Miller Input Compensation frees the amplifier to have very large slew rate virtually independent of compensation. When used in my MOSFET amplifier, that small 50W amplifier achieved a slew rate of 300 V/us.

The third issue, that of nonlinear collector-base capacitance in parallel with the Miller capacitor, is also covered in my book as a potential problem with simple Miller compensation. I show that the elimination of its effect is accomplished by placing an emitter follower in front of the VAS transistor (sometimes referred to as a Darlington VAS, but not in the Darlington sense where both collectors are tied together).

The fourth issue, using a bootstrapped (or "driven") cascode to reduce distortion at the input stage is also covered in my book in a couple of places.

In fairness to Doug, I think he has covered some of these as well.

I do not recall seeing where anyone asserted that cascoding was detrimental to the sound.

Cheers,
Bob

Thanks Bob,

Reading your reply I realize I should've ended mine mentioning that it would be nice to see a comprehensive writeup encompassing a lot of the novel approaches into design examples. It might even be worthwhile to set a few design requirements for an amplifier and work through a complete design from the ground up, describing the design choices, motivate them and finish off with a well rounded design that meets all the design requirements. This also gives folks that do not have design experience a guideline to work from.

I have read parts of your book and Douglas's latest, but although both look to be excellent writeups, most of the design examples or methods to reduce distortion (HEC for example) are decades old and you both seem to be using the same approach. Which is fine, ultimately it is the end result that counts, however methods such as putting an emitter follower in front of the Miller transistor obviously are getting a little long in the teeth.

That obviously doesn't render them ineffective, but modern opamps (for example the OPA1632 and the National LME49xxx series) employ novel methods that are both interesting and worthwhile to investigate. I'm convinced that creating interest and understanding for these novel approaches on amplifier design will yield new insights into distortion mechanisms and new ways to combat distortion and will motivate designers to try these in their own designs.

As for cascoding and bootstrapping, I've seen quite a few topics here and on other forums that refer to anything but the minimum required number of transistors to build an amplifier as detrimental to its sound quality. As a result folks often refer back to designs which are reminescent of '60s amplifiers as the pinnacle in amplifier design.

Either way, I appreciate your reply, I just wanted to give some feedback from the perspective of the reader. I however realize that what I find interesting and would like to read more about might not be the same for other people.

Cheers,

Sander.

[Telstar]

Originally Posted by SSassen

That obviously doesn't render them ineffective, but modern opamps (for example the OPA1632 and the National LME49xxx series) employ novel methods that are both interesting and worthwhile to investigate. I'm convinced that creating interest and understanding for these novel approaches on amplifier design will yield new insights into distortion mechanisms and new ways to combat distortion and will motivate designers to try these in their own designs.

I believe that looking only at harmonic distortion wont bring us anywhere further. PPM THD is just a style exercise that has been done in books and on forums including this one, but does not bring any audible improvement over let's say .001%, as long as in BOTH cases high order harmonics are under the noisefloor.
Quality over quantity any time.
That's why sometimes simple designs sound better than very complex ones.

What is needed it to shift the focus to the quality of the distortion and to other distortion mechanisms that are ignored more often than not.
Also by opamp designers. I agree though, that a lot is to learn studying modern opamp circuits.

[SSassen]

Originally Posted by Telstar

I believe that looking only at harmonic distortion wont bring us anywhere further. PPM THD is just a style exercise that has been done in books and on forums including this one, but does not bring any audible improvement over let's say .001%, as long as in BOTH cases high order harmonics are under the noisefloor.
Quality over quantity any time.
That's why sometimes simple designs sound better than very complex ones.

What is needed it to shift the focus to the quality of the distortion and to other distortion mechanisms that are ignored more often than not.
Also by opamp designers. I agree though, that a lot is to learn studying modern opamp circuits.

I use distortion for lack of a better word to describe all non-linear transfer functions in an amplifier's circuit that get added to the output signal and weren't in the original input circuit. I think that was clear from the exact piece of my reply you quoted?

Cheers,

Sander.

[Bob Cordell]

Originally Posted by SSassen

Thanks Bob,

That obviously doesn't render them ineffective, but modern opamps (for example the OPA1632 and the National LME49xxx series) employ novel methods that are both interesting and worthwhile to investigate. I'm convinced that creating interest and understanding for these novel approaches on amplifier design will yield new insights into distortion mechanisms and new ways to combat distortion and will motivate designers to try these in their own designs.


Cheers,

Sander.

Hi Sander,

Can you elaborate on some of the modern op amp techniques that you feel we have overlooked for use in audio power amplifiers? This might make for some good discussion.

Cheers,
Bob