Stereophile | January 2017 Issue

[KeithR]

What MikeL, Fremer, Harley etc. have chosen is of no consequence to me. If you read the Fremer review of the Lamm ML3 you can see the comments that indicate MF is aware of the tonal nature of the Lamm vs. the more skeletal "speed" approach of the darTZeel. MF is clearly distressed by amplifiers that measure "poorly" (his KR review many years ago showed this...he thought it sounded amazing but felt disgraced by the measurements) and so most likely has an ego issue regarding amp choice. He is just not thinking psychoacoustically though. No idea about MikeL or Harley.

Your point about transformers doesn't make sense. You are making an assumption that a large transformer is inherently slower than a small one? This has no meaning. Perhaps the transformers in your Tri SET amp were simply not very well designed. If the laminations are too thick or the winding not done well enough to prevent HF rolloff etc. But it is clear that to eliminate issues with saturation you need big iron...or go OTL and deal with other issues.

And I am coming to the conclusion SET amps have high, audible 2nd order harmonic distortion that is distracting in my listening and I haven't heard at live performance or with other topologies. I am not the only one who thinks this. So I guess we can agree to disagree.

On tformers, I'm saying weight isn't the only property to optimize sound - you seem enamored with it.

 

[morricab]

Suppose we accept the monotonic distortion decrease theory. Why does this pattern need to be natural (intrinsic) of the output stage? Why can we add it at the SE output and not at the input - an input stage with this pattern of distortion and other stages, including power with minimal distortion? How does the ear/brain know where the distortion is produced?

Do you consider feedback a nuisance because it changes this pattern of distortion or by another specific technical reason?

BTW, Mike Lavigne should be smiling - the DartZeel 458 is one of the few solid state amplifiers that approaches the monotonic distortion pattern!

Try explaining again your point as it doesn't really make sense as you have explained. Every stage will add some of its signature to the sound but the output stage, if push pull, will cancel even harmonics to a more or less degree, depending on how good things are matched. As it was shown also by Boyk and Sussmann, going away from Class A increases distortion significantly, especially at low signal levels.

Feedback was also demonstrated by Crowhurst to create a signal modulated "noise" floor...something Boyk and Sussmann briefly touch upon as well. This will affect dynamics and low level resolution.

As you can see from Boyk and Sussman, in non-simple circuits adding feedback does non-simple things and some high order harmonics actually increase. complementary output stages, differential input stages all have significant impact. As you have been fond of saying...its complicated.

No, it does not. If you look at Figure 9 in the Stereophile review, there is a picket fence of peaks going from 2nd harmonic right out to 20th harmonic. While they do decrease slightly, the drop should be precipitous and by 4th or 5th harmonic you should be just above the noise floor. There are high order harmonics poking up above -90db, which is quite possibly audible and definitely does not fit the aural harmonic pattern that Cheever derived from studies done on ear generated harmonics, which are rarely above 7th harmonic except at very high SPL.

Even worse possibly is the IMD spectrum (Figure 10). Two signals are creating a myriad of peaks as well as noise modulated peaks further down in the spectrum (probaly inaudible but there nonetheless). JA notes that it is consistent with frequency, which is a plus if it is inocuous in the first place

Now, these were done at high power but if you look at the THD + noise plots you will see that the dart has the lowest distortion just before clippping...so low power will not look substantially better from a relative perspective.

 

[microstrip]

Try explaining again your point as it doesn't really make sense as you have explained. Every stage will add some of its signature to the sound but the output stage, if push pull, will cancel even harmonics to a more or less degree, depending on how good things are matched. As it was shown also by Boyk and Sussmann, going away from Class A increases distortion significantly, especially at low signal levels.

Signatures are added, but can be masked by stronger ones. If some harmonics have much higher value in a stage the result will be dominated by this stage. Take a Devialet for example. If you add a tube input stage accomplishing your preferred model you get the model spectra you want.

Feedback was also demonstrated by Crowhurst to create a signal modulated "noise" floor...something Boyk and Sussmann briefly touch upon as well. This will affect dynamics and low level resolution.

As you can see from Boyk and Sussman, in non-simple circuits adding feedback does non-simple things and some high order harmonics actually increase. complementary output stages, differential input stages all have significant impact. As you have been fond of saying...its complicated.

Yes, it is complicated. My point here is that these conclusions have not been confirmed subjectively.

No, it does not. If you look at Figure 9 in the Stereophile review, there is a picket fence of peaks going from 2nd harmonic right out to 20th harmonic. While they do decrease slightly, the drop should be precipitous and by 4th or 5th harmonic you should be just above the noise floor. There are high order harmonics poking up above -90db, which is quite possibly audible and definitely does not fit the aural harmonic pattern that Cheever derived from studies done on ear generated harmonics, which are rarely above 7th harmonic except at very high SPL.

Even worse possibly is the IMD spectrum (Figure 10). Two signals are creating a myriad of peaks as well as noise modulated peaks further down in the spectrum (probaly inaudible but there nonetheless). JA notes that it is consistent with frequency, which is a plus if it is inocuous in the first place

Now, these were done at high power but if you look at the THD + noise plots you will see that the dart has the lowest distortion just before clippping...so low power will not look substantially better from a relative perspective.

Thanks for analyzing the Stereophile review. So we need a well defined attenuation of harmonics, not just being monotonic and a limit. What is the best slope according to your findings? Around 10 dB per harmonic? Or just scaled such that the 5th is around -90dB?

Can you point measurements of one amplifier you consider exemplary?

 

[morricab]

Haven't we already debunked Cheever? I mean come on Brad. It's a dude who wrote a thesis based on a 5 person test on 2 crappy amps 20 years ago. It was never reviewed or presented to AES or anything. The guy isn't even in the audio world today, but a consultant. Yet you ignore Nelson Pass who says a third of people prefer 3rd harmonic and has 100x over the experience of Cheever.

Ironically you haven't posted measurements of the Aries Cerat which you are so in love with now. I'm going to die laughing if it has a bit of NFB lol.

Come on Keith, I haven't seen anything conclusive debunking Cheever...maybe in your mind that is the case...whether his test results are valid or not doesn't change the basics of his theoretical approach at obtaining an metric that fits very logically and is more comprehensive than anything coming before. He isn't the first, Btw, to have similar thinking...look at D.E.L Shorter (BBC engineer) and Jean Hiraga (he talked about monotonicity being a key to good sound) among others. Cheever tied it together with aural harmonics in a quite elegant way as a justification for WHY this pattern should be the most natural because it is related to how human hearing actually works. BTW, hte data on the hearing was not generated by him but by Olson. On last point, what does it matter that Cheever is not in the audio world today? Does that make the paper less valuable? Since this was a thesis, it was likely approved by his department and filed in the library...just like my Ph.D thesis. They are not published to the world at large normally.

I do not ignore Pass but Pass has done no controlled listening tests confirming his assertion, which is strictly anecdotal...if has the data he isn't sharing. He does criticize feedback and multiple stages with feedback (same white paper you are getting his anecdote from.

"Anecdotally, it appears that preferences break out roughly into a third of customers liking 2nd harmonic types, a third liking 3rd harmonic, and the remainder liking neither or both. Customers have also been known to change their mind over a period of time." So, completely uncontrolled.

"Audiophiles have been accused of using 2nd or 3rd harmonic distortion as tone controls to deliberately alter the sound. I suppose that there are people who like it that way, but I don't think this is generally the case. For reasons which will become clearer when we talk about inter-modulation distortion, high levels of any harmonic become problematic with musical material having multiple instruments, and the argument that 2nd or 3rd adds “musicality” doesn't quite hold up."

This is why you have to look at the whole pattern. Your 2nd harmonic is hurting your listening...just doesn't add up as Pass says.

"The high harmonic content of Class B amplifiers brings us to the word monotonicity. Monotonicity describes the relationship between the distortion level and the output level. The smooth transfer curves of Class A amplifiers have a characteristic which is monotonic, that is to say the distortion goes down as the output declines. It implies low order harmonic characteristic, which we have previously agreed is sonically preferred."

This is totally consistent with Hiraga and with Cheever. Cheever mentions SPL because that affects how much power the amp has to produce, which affects the distortion...

"Negative feedback can reduce the total quantity of distortion, but it adds new components on its own, and tempts the designer to use more cascaded gain stages in search of better numbers, accompanied by greater feedback frequency stability issues.

The resulting complexity creates distortion which is unlike the simple harmonics associated with musical instruments, and we see that these complex waves can gather to create the occasional tsunami of distortion, peaking at values far above those imagined by the distortion specifications.
"

Consistent with the theory of aural harmonics don't you think?

 

[morricab]

Haven't we already debunked Cheever? I mean come on Brad. It's a dude who wrote a thesis based on a 5 person test on 2 crappy amps 20 years ago. It was never reviewed or presented to AES or anything. The guy isn't even in the audio world today, but a consultant. Yet you ignore Nelson Pass who says a third of people prefer 3rd harmonic and has 100x over the experience of Cheever.

Ironically you haven't posted measurements of the Aries Cerat which you are so in love with now. I'm going to die laughing if it has a bit of NFB lol.

Forgot the last bit. I haven't posted what I haven't got. I have not seen any Aries Cerat measurements. It is not the only amp I like but it is almost certainly without NFB.

 

[morricab]

Signatures are added, but can be masked by stronger ones. If some harmonics have much higher value in a stage the result will be dominated by this stage. Take a Devialet for example. If you add a tube input stage accomplishing your preferred model you get the model spectra you want.


Yes, it is complicated. My point here is that these conclusions have not been confirmed subjectively.



Thanks for analyzing the Stereophile review. So we need a well defined attenuation of harmonics, not just being monotonic and a limit. What is the best slope according to your findings? Around 10 dB per harmonic? Or just scaled such that the 5th is around -90dB?

Can you point measurements of one amplifier you consider exemplary?

According to Cheever's calculation if 2nd harmonic is 5% (-26 db)at 90db then the third should be at -60db, fourth at -90db or so and the 5th at about -110db, 6th at -125db and 7th around -140db, beyond that they just get lower and lower. essentially nothing much beyond 7th. dartzeel runs right up to 20th and the IMD is harmonics everywhere.

Cheever's slope is SPL defined and it changes somewhat at lower and higher volumes. How stressed your amp is at those SPLs will depend on the sensitivity and load of your speakers. So, to match this slope from Cheever's calculation your amp will need to make this pattern (or the pattern shifted lower in level...like 1% 2nd and subsequent drop in all other harmonics) at the given SPL to match the ear's self-generated pattern.

I will have to see if I can find a good example. I seem to remember the Lamm ML2 was pretty good in this regard though.

 

[amirm]

According to Cheever's calculation if 2nd harmonic is 5% (-26 db)at 90db then the third should be at -60db, fourth at -90db or so and the 5th at about -110db, 6th at -125db and 7th around -140db, beyond that they just get lower and lower. essentially nothing much beyond 7th. dartzeel runs right up to 20th and the IMD is harmonics everywhere.

Do you know where he got those numbers? I ask because the shape of auditory mask differs with both level and frequency. As such there cannot be such rules across the band.

Here is the graph for different levels at 1 Khz for example:

At 100 db SPL for example, masking continues to threshold of hearing at -65 db. Nothing below that would be audible in the presence of the 100 db 1 Khz tone.

In addition, the maximum dynamic range of the ear is about 116 db SPL. It is a windowed function so doesn't simultaneously hear all of that (due to electromechanical aspects of OHC). As such I am very surprised that someone would be talking about distortion products down to -140 db and such.

So love to see the research that says otherwise.

 

[microstrip]

According to Cheever's calculation if 2nd harmonic is 5% (-26 db)at 90db then the third should be at -60db, fourth at -90db or so and the 5th at about -110db, 6th at -125db and 7th around -140db, beyond that they just get lower and lower. essentially nothing much beyond 7th. dartzeel runs right up to 20th and the IMD is harmonics everywhere.

Cheever's slope is SPL defined and it changes somewhat at lower and higher volumes. How stressed your amp is at those SPLs will depend on the sensitivity and load of your speakers. So, to match this slope from Cheever's calculation your amp will need to make this pattern (or the pattern shifted lower in level...like 1% 2nd and subsequent drop in all other harmonics) at the given SPL to match the ear's self-generated pattern.

I will have to see if I can find a good example. I seem to remember the Lamm ML2 was pretty good in this regard though.

Just checked - it seems we can not compare the available data at Stereophile or Soundstage - the distortion data on the Dartzeel you refer (fig.9) is at 150w/8ohm, 300W/4ohm and the data on the Lamm's is at 1W/8ohm. Apples versus oranges, as they say.
All the peaks of the Dartzeel distortion at 1W are bellow -100dB.

 

[microstrip]

(...) Since this was a thesis, it was likely approved by his department and filed in the library...just like my Ph.D thesis. (...)

It was just a Master Thesis not a PhD Thesis - Keith is right putting it in the proper place. I have gone through it before and unless you are prepared to support his suppositions, you can not use it to firmly support anything. It was supposed to be a start for an interesting project that never went on ...

The bibliography section is really worth reading for those interested in the subject.

 

[KeithR]

I do not ignore Pass but Pass has done no controlled listening tests confirming his assertion, which is strictly anecdotal...if has the data he isn't sharing. He does criticize feedback and multiple stages with feedback (same white paper you are getting his anecdote from.

A 5 person test based on two crappy amps (1 was diy as I recall) is more conclusive than 30 years of potentially uncontrolled tests by one of the most creative amp designers in audio history?

And yes, I tend to agree with you that NFB is to be avoided and that simpler amps typically sound better - but you have taken it MUCH further.

ps. Daniel Cheever's site is conveniently down now for the peanut gang to read

 

[PeterA]

Haven't we already debunked Cheever? I mean come on Brad. It's a dude who wrote a thesis based on a 5 person test on 2 crappy amps 20 years ago. It was never reviewed or presented to AES or anything. The guy isn't even in the audio world today, but a consultant. Yet you ignore Nelson Pass who says a third of people prefer 3rd harmonic and has 100x over the experience of Cheever.

Ironically you haven't posted measurements of the Aries Cerat which you are so in love with now. I'm going to die laughing if it has a bit of NFB lol.

I remember seeing this, but can't recall if NP said it about people in general, audiophiles, or specifically, his amplifier customers. I did read that Jon Valin prefers the Pass X series (X350.5, class AB) to the XA series (class A) for instance. I don't know which line is more popular, but they do have distinct sounds, and I think Pass amplifiers have different ratios of 2nd versus 3rd harmonics, depending on the model. I presume this is all about subjective preference rather than measurable faithfulness (accuracy) to the signal.

 

[KeithR]

Do you know where he got those numbers? I ask because the shape of auditory mask differs with both level and frequency. As such there cannot be such rules across the band.

Here is the graph for different levels at 1 Khz for example:

At 100 db SPL for example, masking continues to threshold of hearing at -65 db. Nothing below that would be audible in the presence of the 100 db 1 Khz tone.

In addition, the maximum dynamic range of the ear is about 116 db SPL. It is a windowed function so doesn't simultaneously hear all of that (due to electromechanical aspects of OHC). As such I am very surprised that someone would be talking about distortion products down to -140 db and such.

So love to see the research that says otherwise.

I'd give you the link to the Cheever article, but the site appears down - I would love your feedback on it.

 

[KeithR]

Actually, here is another link that works:

http://milbert.com/Files/articles/DanCheever.pdf

Here is Cheever's "Study"

"Although a strict scientific experiment was conducted comparing the amplifiers I do not present an
detailed analysis here. In summary, the type 45 amplifier was chosen as preferable 100% of the time by all
the different listeners (5) in a “single-blind environment”, meaning, the listener toggled a remote push
button and either the amplifiers were swapped or not. A numeric display was incremented at each selection
and the listener noted if the amplifier changed or not, and if the change was to the one preferred. I attended
all sessions and verified that levels where matched, that there was no clipping, and the program material
kept within the flat pass-band of the type 45 amplifier-speaker combination. The source was a live piano
microphone feed"

Let's just call a spade a spade and just say this is a theory proposed by an EE, but there is really no hard evidence presented and even Cheever in the end says the application is limited as his formulas need much more broad testing due to age, sex, etc.

 

[morricab]

Just checked - it seems we can not compare the available data at Stereophile or Soundstage - the distortion data on the Dartzeel you refer (fig.9) is at 150w/8ohm, 300W/4ohm and the data on the Lamm's is at 1W/8ohm. Apples versus oranges, as they say.
All the peaks of the Dartzeel distortion at 1W are bellow -100dB.

If you look at THD vs. power you see that the relative distortion for the dart gets worse as power goes down and the Lamm gets better as the power goes down. This means the Lamms perform realtively better as things get softer and the darts relatively better as it gets louder.

 

[amirm]

I'd give you the link to the Cheever article, but the site appears down - I would love your feedback on it.

Whoa. You didn't tell me it was 87 pages! I read or should I say suffered through it.

His work is quite extensive but alas, it misses core fundamentals for such work. If one wants to correlate subjective listening performance to objective data, both of them need to be on solid ground. The former unfortunately is not remotely so. He relies on a bunch of anecdotal evidence such as so and so reviewer saying an amp is good. And his own test which is but a footnote:

"15 Although a strict scientific experiment was conducted comparing the amplifiers I do not present an
detailed analysis here. In summary, the type 45 amplifier was chosen as preferable 100% of the time by all
the different listeners (5) in a “single-blind environment”, meaning, the listener toggled a remote push
button and either the amplifiers were swapped or not. A numeric display was incremented at each selection
and the listener noted if the amplifier changed or not, and if the change was to the one preferred. I attended
all sessions and verified that levels where matched, that there was no clipping, and the program material
kept within the flat pass-band of the type 45 amplifier-speaker combination. The source was a live piano
microphone feed."

Lots and lots of detail is missing here. For one thing, single blind tests are invalid unless they are shown to be practically equiv. to double blind. This would require for example for the test conductor to be out of view, no communication between him and testers, etc. None of this is stated so the door is wide open for invalid outcome (see https://en.wikipedia.org/wiki/Clever_Hans).

Second, there has to be statistical rigor in the listening tests. Enough trials need to occur to reduce probability of chance to .05. We don't know from his footnote if testers got the outcome right once, or a dozen times.

Then there is this detail which doesn't come out until the end of the paper in conclusions:

"For example, a set of 1960’s
Radio Shack 16 ohm PA monitors was necessarily chosen in the listening tests because it
had a very little impedance variation over the range of frequencies used in the listening
tests- a grand piano. As shown in Fig 3-5, this speaker was flat within +/- 0.25dB from
180Hz to 18kHz, meeting the ABX [40] specifications for frequency response matching.
Loudspeaker efficiency is also an issue with amplifiers with power outputs of less than
5W. "

He used a PA speaker? And one that matched the output impedance of the tube SET amp to produce a smooth response? What did it do for the solid state?

He also says a bunch of stuff that despite the length of the paper, are unsupported:

"1) The ears’ self generated harmonics mask external harmonic distortion that has the
same character. The ears’ harmonic distortion is fully studied and falls off at a rate of
approximately 10n
, where the power n designates the harmonic number. I propose
that external harmonics strictly adhering to this envelope are fully “undistorted” by
our ear-brain system and are thus indistinguishable from pure tones. An analytical
derivation of conformance to this aural harmonic envelope is developed."

Who says? Where are the careful listening tests that demonstrate this? No, you can't compare one amp to another to prove this. You have to create synthetic tests with such characteristics and demonstrate this.

I am also troubled by him saying the ear's distortion masks equipment distortion. Distortion is additive. It is not subtractive.

"To claim that these distortions are in-audible is fallacious, as they modulate at higher and
higher non-linear degrees with the instantaneous signal level."

This is an assertion, not a statement backed by research and fact. It is the type of thing one would read on forums, not in a scientific paper. Show the fallacious aspect, don't claim it.

"I emphasize that the subjective
capabilities of audio amplification are far more strongly aligned with open-loop linearity
than magnificent closed loop single-sine bench test results. "

Again, an opinion. His thesis around feedback is just plain wrong. Feedback reduces the levels of overall distortion to well below masking. Once there, it doesn't matter what pattern it has. The components are inaudible regardless of pattern. Or else masking would be invalid. Feedback's harm comes when you clip. Then it produces far more horrendous results than not having it. But below that threshold, it produces goodness that is easily verified with controlled listening tests and psychoacoustics.

Finally in the conclusion of his paper we got to the nut of it:

"The output impedance of these amplifiers is in the ohms range, hundreds of times worse
than solid-state push-pull amplifiers. This requires careful mating with loudspeakers that
do not have great impedance variation with frequency. "

This is why you hear different sound which you may prefer out of a tube amp. Its high output impedance has high interaction with the dynamic impedance of the speaker creating a unique (and colored) combination of amp+speaker. This behavior is highly audible and can be backed left and right. Changes as low as 0.5 db in low frequencies for example are audible. All the rest of his paper is, pardon my directness, nonsense compared to this fact.

 

[microstrip]

If you look at THD vs. power you see that the relative distortion for the dart gets worse as power goes down and the Lamm gets better as the power goes down. This means the Lamms perform realtively better as things get softer and the darts relatively better as it gets louder.

We were just being thought that the important aspect for sound quality is not the distortion figure, but the relative levels of the harmonics in the distortion spectra. Are we back in the old way?

 

[amirm]

And here is an example speaker impedance changes with frequency: Wilson Alexa:

You see all the ups and downs of the solid line? All of that interacts with the SET amp's high output impedance to change the frequency response of the overall system. It is an in-built equalizer that cannot be disabled.

 

[morricab]

It was just a Master Thesis not a PhD Thesis - Keith is right putting it in the proper place. I have gone through it before and unless you are prepared to support his suppositions, you can not use it to firmly support anything. It was supposed to be a start for an interesting project that never went on ...

The bibliography section is really worth reading for those interested in the subject.

Have you done either a master's thesis or a PhD? I have done both and I am here to tell you that methodologically there is no difference only the size of the question to be answered and many master'theses turn into a PhD thesis as you go further down the rabbit hole.

That it is a master's thesis doesn't diminish the contribution.

He did enough in a limited way to show basic proof of principle. He has based his work on solid work from the past and has demstrsted in a limited way that he is on the right track.

Geddes also demonstrated no correlation with THD from 0.02% to 20% vs. listening and depending on the nature of the distortion came up with the Gedlee metric. I would be surprised if it gave radically different results.

Shorter had something less sophisticated over 60 years ago.

 

[KeithR]

And here is an example speaker impedance changes with frequency: Wilson Alexa:

You see all the ups and downs of the solid line? All of that interacts with the SET amp's high output impedance to change the frequency response of the overall system. It is an in-built equalizer that cannot be disabled.

To be fair, nobody would use a SET with that speaker Amir nor do most think SETs are universal.

 

[microstrip]

Have you done either a master's thesis or a PhD? I have done both and I am here to tell you that methodologically there is no difference only the size of the question to be answered and many master'theses turn into a PhD thesis as you go further down the rabbit hole.

That it is a master's thesis doesn't diminish the contribution.

He did enough in a limited way to show basic proof of principle. He has based his work on solid work from the past and has demstrsted in a limited way that he is on the right track.

Geddes also demonstrated no correlation with THD from 0.02% to 20% vs. listening and depending on the nature of the distortion came up with the Gedlee metric. I would be surprised if it gave radically different results.

Shorter had something less sophisticated over 60 years ago.

So we find we disagree on what is a master thesis and a PhD thesis ... This forum is surely not interested in it, I will skip this part ...

IMHO we can get information from these works, but not use them as proof of our arguments. YMMV.