10 tones played together, each with 4V peak or 1W @ 8 ohm, could sum to 40V peak and reach the clipping limit of a 100W amplifier.This is probably why multi-tone tests are now becoming more popular. You have a lot of tones mixing to create IM products. My only complaint about some of these tests is they are done at very low power levels where the total tone power is low ie the amplifier is not operating in its large signal regime where large signal non-linearity may be contributing significant distortion.
This is interesting because the amplifier does actually traverse a wide range of voltages, but the relative time spent near those peaks would be a lot lower. Therefore, such a test could be tuned to have a statistical distribution of output voltages that more closely resembles music or whatever "program material" the amplifier is designed for.
I’ve posted the first cut proposal up in the first post. Why don’t we look at that and have a discussion about the scoring on the first two amplifier types?
There currently is no consensus on what the ideal distortion test would look like. The current proposals in the document are THD+N at 1dBV out into 8 ohms, and then the same measurement at rated power into 8 ohms. The second proposed test in the document is 19+20 kHz IMD at 80% rated power.
There currently is no consensus on what the ideal distortion test would look like. The current proposals in the document are THD+N at 1dBV out into 8 ohms, and then the same measurement at rated power into 8 ohms. The second proposed test in the document is 19+20 kHz IMD at 80% rated power.
My first comment - why separate on amplifier type. I thought we seemed to agree on that the test should be implementation independent?
//
//
We said earlier on in the thread that if we had a single assessment standard, the result would be only one type of amplifier. I don’t think zero global feedback amplifiers should be penalised. Ditto vacuum tube amps. A well engineered 0.5% tube amplifier is still a valid amplifier and should be assessed on its merits within what is possible with the chosen technology.
Bart Kosko said in the article I linked:...(central limit theorem)...
One response to this call for tail-thickness sensitivity is that logic alone can decide the matter because of the so-called central limit theorem of classical probability theory. This important "central" result states that some suitably normalized sums of random terms will converge to a standard normal random variable and thus have a normal bell curve in the limit. So Gauss and a lot of other long-dead mathematicians got it right after all and thus we can continue to assume normal bell curves with impunity.
That argument fails in general for two reasons.
The first reason it fails is that the classical central limit theorem result rests on a critical assumption that need not hold and that often does not hold in practice. The theorem assumes that the random dispersion about the mean is so comparatively slight that a particular measure of this dispersion — the variance or the standard deviation — is finite or does not blow up to infinity in a mathematical sense. Most bell curves have infinite or undefined variance even though they have a finite dispersion about their center point. The error is not in the bell curves but in the two-hundred-year-old assumption that variance equals dispersion. It does not in general. Variance is a convenient but artificial and non-robust measure of dispersion. It tends to overweight "outliers" in the tail regions because the variance squares the underlying errors between the values and the mean. Such squared errors simplify the math but produce the infinite effects. These effects do not appear in the classical central limit theorem because the theorem assumes them away.
The second reason the argument fails is that the central limit theorem itself is just a special case of a more general result called thegeneralized central limit theorem. The generalized central limit theorem yields convergence to thick-tailed bell curves in the general case. Indeed it yields convergence to the thin-tailed normal bell curve only in the special case of finite variances. These general cases define the infinite set of the so-called stable probability distributions and their symmetric versions are bell curves. There are still other types of thick-tailed bell curves (such as the Laplace bell curves used in image processing and elsewhere) but the stable bell curves are the best known and have several nice mathematical properties. The figure below shows the normal or Gaussian bell curve superimposed over three thicker-tailed stable bell curves. The catch in working with stable bell curves is that their mathematics can be nearly intractable. So far we have closed-form solutions for only two stable bell curves (the normal or Gaussian and the very-thick-tailed Cauchy curve) and so we have to use transform and computer techniques to generate the rest. Still the exponential growth in computing power has long since made stable or thick-tailed analysis practical for many problems of science and engineering.
https://www.edge.org/response-detail/11715
https://arxiv.org/abs/1908.03580#:~:text=The Generalized Central Limit Theorem is a remarkable generalization of,to a special family known
So now we are at the defining moment...
SQ or tech?
It's your metric and you cry if you want to....
Me, I'll leave the party 😉 it's doomed... ;-D
//
SQ or tech?
It's your metric and you cry if you want to....
Me, I'll leave the party 😉 it's doomed... ;-D
//
About THD (so if you're tired of it you can ignore me now) I just went through the whole thread, THD is one of the things to be measured in Post #1 and was mentioned throughout, but it was only after 180-200 posts before someone dared to question its use. I see Markw4 brought up the usual criticisms and mentioned the GedLee metric.
But to the op, please do NOT use THD in an amplifier quality metric. Rather than just being negative, I'll offer a quick alternative.
Get the harmonic distortion spectrum from an FFT or filters or whatever. Weigh the amplitude of each harmonic by its harmonic number, so for example the 7th harmonic amplitude counts as 3 1/2 times the 2nd harmonic amplitude. You might even make the function so the higher harmonics are weighed even more than that. Then do the square root of sums of squares thing. It's bound to correlate better to amplifier sound than THD.
But to the op, please do NOT use THD in an amplifier quality metric. Rather than just being negative, I'll offer a quick alternative.
Get the harmonic distortion spectrum from an FFT or filters or whatever. Weigh the amplitude of each harmonic by its harmonic number, so for example the 7th harmonic amplitude counts as 3 1/2 times the 2nd harmonic amplitude. You might even make the function so the higher harmonics are weighed even more than that. Then do the square root of sums of squares thing. It's bound to correlate better to amplifier sound than THD.
True but that ignores IMD where 2nd order distortion is not so well hidden by auditory masking as in the single-tone case, and can dominate the IMD spuriae.
Rather disappointed so many people here regard DBLTs as 'anecdotal'. MY DBLTs, conducted over nearly 2 decades are certainly NOT anecdotal. A DBLT is a MEASUREMENT. Your instrument is your DBLT panel. It has an accuracy which needs to be checked regularly. I've mentioned certain aspects but properly conducted DBLTs are a HUGE topic and I've been doing them a LOT longer than the false Harman prophets Sean Olive etc.
And YES. Doing proper DBLTs is VERY expensive & time consuming which is why you record ALL details, including da statistical analysis .. and you still have to repeat them every now & then ... cos they usually come under 'sparse data' analysis.
But to come back to the original topic, it is obvious the faults picked up by DBLTs, ie the stuff people can hear, are usually quite crude IF you know how to look for them. I'd like to think any FOM will at least put some importance to these things.
Anyone have a link to GedLee's follow up article where he claims you can hear THD <0.01% ?
And YES. Doing proper DBLTs is VERY expensive & time consuming which is why you record ALL details, including da statistical analysis .. and you still have to repeat them every now & then ... cos they usually come under 'sparse data' analysis.
But to come back to the original topic, it is obvious the faults picked up by DBLTs, ie the stuff people can hear, are usually quite crude IF you know how to look for them. I'd like to think any FOM will at least put some importance to these things.
Anyone have a link to GedLee's follow up article where he claims you can hear THD <0.01% ?
Last edited:
Coming back to THD, in da 60s & 70s, THD <0.01% under ALL conditions was quite difficult. Modern devices sadly many of them becoming Unobtainium, make this straightforward.
I'll stick my neck out and claim <0.01% under ALL conditions is inaudible. Between 0.1% & 0.01%, there are amps which can be distinguished and preferred by da true golden pinnae. But also loadsa stuff and those amps with >0.1% THD, which da wannabe Golden Pinnae claim supa dupa but whose proponents can't pick out reliably in DBLTs ... and are disliked by da true golden pinnae.
Gotta remember that its quite difficult (??!) to do an amp with THD <0.01% under all conditions that has yucky xover, loadsa high order harmoncs bla bla 🙂
I'll stick my neck out and claim <0.01% under ALL conditions is inaudible. Between 0.1% & 0.01%, there are amps which can be distinguished and preferred by da true golden pinnae. But also loadsa stuff and those amps with >0.1% THD, which da wannabe Golden Pinnae claim supa dupa but whose proponents can't pick out reliably in DBLTs ... and are disliked by da true golden pinnae.
Gotta remember that its quite difficult (??!) to do an amp with THD <0.01% under all conditions that has yucky xover, loadsa high order harmoncs bla bla 🙂
How many test subjects, how were they selected, how was the test gear qualified, what was the distortion and noise of the system and recordings upstream of the amplifiers, what equipment was used for testing, any ground loops in the system, was the system tested for signal correlated noise, were the test subjects trained on detecting lower and lower levels of IMD with appropriate musical recording segments, what kind of transducers were used, did you assume gaussian distributions, discard any outliers, etc.?MY DBLTs, conducted over nearly 2 decades are certainly NOT anecdotal.
You understand, we were not there. We don't know what you did or didn't do. We don't know how to reproduce what you did in order to see if we can replicate your results. There are many potential concerns by modern standards. No offense intended, of course. Just interested in good science relative to what is known about human perceptual testing today.
Also, what about a quote from Earl Geddes:
In our next paper we will show that .01% THD of one type of nonlinear system can be perceived as unacceptable while 10% THD in another example is perceived as inaudible.
http://www.gedlee.com/Papers/Distortion_AES_I.pdf
Last edited:
I used to test anyone who visited the factory and/or our R&D labs. The amplifier tests involved a much smaller group of people cos only certain people can hear small differences. My panel has at times included Big Ears Banks, and Bat Ears Faulkner. I've also tested JA who is perhaps the ONLY reviewer I regard as better than da man (or more particularly woman) in da street. BTW, even da best ears are prejudiced so there are people, including these 3, whose opinions I trust blind but take what they say sighted with a big pinch of salt 😱
That's why I can claim da stuff in #136. This isn't anecdotal. If you place a BIG enough $$$ wager to make it worthwhile for me, I'll repeat the tests for you.
The selection is simply those who perform consistently on DBLTs. Unlike da false Harman prophets, I found no difference in preference between men/women in da street and 'trained' listeners. You just have to configure the test correctly. eg. Listeners choose THEIR music. Each person is tested individually. THEY have complete control over switching.
The amp tests were ABC (as all my tests) with 2 of the presentations the same. BUT THE LISTENER IS NEVER TOLD THIS. They may take up to 1/2 day to do ONE test .. only cos I can't afford to pay my staff for longer sessions. Some of the tests were repeated 3x. You discount ALL those who don't pick the odd man out on all three tests (though they don't know this).
You ONLY ask a preference on those who pass this. For the others, you just tell them loudly & clearly that your stuff is hand carved from Unobtainium and solid BS 🙂
I could go on but I think you see why proper DBLTs are much more $$$ than the latest AP.
I could go on ... To some the other points in your question, the listeners were only asked to rate the presentations out of 10 and give a preference (which can be different from their scores). They are not even told what they are listening to. We started off with very complex forms but quickly came down to this as most useful.
The tests check two things. Firstly, is there an audible difference? And this also depends on the listener. But if my A team don't hear differences, I'm willing to wager $$$ no one else can.
Secondly, IF and ONLY IF, the first answer is yes, do check the results to see if there is a consistent preference. One of the surprising results is that when there was a general audible difference, da hard rock fan, da classical musician, da speaker designer, da recording engineer (who might insist on playing his own recordings) they all preferred the same thing.
The tests spanned copy master tapes, vinyl, CDs, and some early digital recordings with a Sony PCM-F1 and the prototype Mk4 Soundfield. Bat Ears used some of his own recordings but wasn't involved in da amp DBLTs
That's why I can claim da stuff in #136. This isn't anecdotal. If you place a BIG enough $$$ wager to make it worthwhile for me, I'll repeat the tests for you.
The selection is simply those who perform consistently on DBLTs. Unlike da false Harman prophets, I found no difference in preference between men/women in da street and 'trained' listeners. You just have to configure the test correctly. eg. Listeners choose THEIR music. Each person is tested individually. THEY have complete control over switching.
The amp tests were ABC (as all my tests) with 2 of the presentations the same. BUT THE LISTENER IS NEVER TOLD THIS. They may take up to 1/2 day to do ONE test .. only cos I can't afford to pay my staff for longer sessions. Some of the tests were repeated 3x. You discount ALL those who don't pick the odd man out on all three tests (though they don't know this).
You ONLY ask a preference on those who pass this. For the others, you just tell them loudly & clearly that your stuff is hand carved from Unobtainium and solid BS 🙂
I could go on but I think you see why proper DBLTs are much more $$$ than the latest AP.
I could go on ... To some the other points in your question, the listeners were only asked to rate the presentations out of 10 and give a preference (which can be different from their scores). They are not even told what they are listening to. We started off with very complex forms but quickly came down to this as most useful.
The tests check two things. Firstly, is there an audible difference? And this also depends on the listener. But if my A team don't hear differences, I'm willing to wager $$$ no one else can.
Secondly, IF and ONLY IF, the first answer is yes, do check the results to see if there is a consistent preference. One of the surprising results is that when there was a general audible difference, da hard rock fan, da classical musician, da speaker designer, da recording engineer (who might insist on playing his own recordings) they all preferred the same thing.
The tests spanned copy master tapes, vinyl, CDs, and some early digital recordings with a Sony PCM-F1 and the prototype Mk4 Soundfield. Bat Ears used some of his own recordings but wasn't involved in da amp DBLTs
Wanna explain how these affect the conduct and/or results of DBLTs? Especially da 'modern standards'how was the test gear qualified, what was the distortion and noise of the system and recordings upstream of the amplifiers, what equipment was used for testing, any ground loops in the system, was the system tested for signal correlated noise, were the test subjects trained on detecting lower and lower levels of IMD with appropriate musical recording segments, what kind of transducers were used, ....
.... There are many potential concerns by modern standards.
No need. PMA already tested me with one of his DBT unity-gain noninverting audio opamp buffer listening tests. Proceeding from my recollection, he asked for people to show by ABX that they could tell any two apart. These were digital recordings he made of different buffers. I couldn't tell them apart by ABX, so I tried to find the two that were the most different. In that effort I imported all the files into Reaper, a DAW, and starting trying to sort them so I could find the two the most different. Only bubble sorting worked, and it was very difficult. I quit before sorting the last one, too. Said to heck with it. Somebody else can do it. But I ended up with a sort order of difference in order perceived brightness (where the differences were very slight). Then I though I would send PMA the sort list even if I couldn't do what he wanted the ABX way. But first I had an important decision to make. Were the brighter files more clear or more distorted? They were slightly different but what was the difference? I decided to go with brighter as more distorted. I sent the list to PMA by PM before he announced which file was which. Turned I chose correctly, and the sort order I head was in fact in increasing distortion order, each opamp slightly more distorted than the one before it in the list. For that little effort PMA gave me a brief type of honorable mention in one post. PMA measured all the buffers himself. He knew what the distortion level were. They were what you would expect from audio opamp buffers. Beyond the buffers themselves there was his A/D distortion and by playback system I used involved. Speakers were NS-10s. Amp was Bryston 4-B. DAC was Benchmark DAC-1 or DAC-3, don't recall anymore. According to PMAs measurements I sorted all the ones I bothered to sort (all but one) in order of distortion. It was easier than ABX for me.I'll repeat the tests for you.
The most interesting thing to me that happed after that with PMA is he briefly announced he learned how to pass ABX. He said is was difficult, took sustained concentration, etc. Right out of the ABX denial playbook, as some folks once liked to say. But here was PMA admitting its hard to learn how to do it. Bravo to PMA!!!
After that I made repeated recommendations for improvements to Foobar ABX plugin, and also made recommendations for trying other test protocols such as sort order tests. Nobody cared. Nothing ever came of any of it. We're still at the I can make it so you can't pass my amateur ABX game. In my book its BS.
Separately, but not without merit, you and Earl Geddes seem to disagree somewhat. Which one of you is mistaken?
I did have a brief online conversation with Earl in the forum one day. I asked him if we really know what 100% of the human population can or can't hear in terms of distortion. He said he thought we had a pretty good idea of what about 95% of the population can hear. He said that to study the other 5% or so would take the development of new tests. He said a preliminary study might cost in the range of several thousand to several tens of thousands of dollars. To me Earl sounds more like a professional scientist than you do.
Last edited:
Claiming what THD is detectable without knowing harmonic profile is meaningless. 0.03% of odd harmonics is likely detectable, but may be undetectable/benign if only second/even harmonics are present.
Tried it with test cd and headphones.
THD as one single number is useless parameter.
Tried it with test cd and headphones.
THD as one single number is useless parameter.
Last edited:
1000 watts clipping at home hifi system? Now i heard everything.
When i play test cd with -3dB signal (pink noise for instance from ultracurve to set eq), at normal listening level, and i measure signal on speakers with oscilloscope...and calculating ~8 ohm for speakers, its 200 milli watts. Yes, milli watts. If i crank it up for loud music it can reach watt or two.
Now, when it comes to garage system or for bbq party, that's a different story. Well, who cares about sound quality then. No scoring needed. The more beer, the better it sounds.
When i play test cd with -3dB signal (pink noise for instance from ultracurve to set eq), at normal listening level, and i measure signal on speakers with oscilloscope...and calculating ~8 ohm for speakers, its 200 milli watts. Yes, milli watts. If i crank it up for loud music it can reach watt or two.
Now, when it comes to garage system or for bbq party, that's a different story. Well, who cares about sound quality then. No scoring needed. The more beer, the better it sounds.
One serious document at: https://www.itu.int/dms_pubrec/itu-r/rec/bs/R-REC-BS.1116-3-201502-I!!PDF-E.pdf They do explain what you should do, but unfortunately they don't go into all the supporting literature.Wanna explain how these affect the conduct and/or results of DBLTs? Especially da 'modern standards'
Other than that, I could probably put some references in my dropbox if you would like. Most likely its more than I can attach here. One big issue has to do with the very important distinction between preference testing and discrimination testing. Two very different things. For example, if I can ask what animal you hear playing (and its chicken clucking or a dog barking), the answer isn't I prefer cats meowing. Its not about what you prefer or don't prefer at all. Also, you have to be trained to make sure you know what a chicken and what a dog sound like. Then maybe we start to reduce the volume to find out the lowest volume level at which you can correctly discriminate. Again, nothing to do with preference.
Anyway, when it comes to detecting distortion, it should not be about whether the listener prefers it or not. It should be that they are trained to know how to discriminate distortion at lower and lower levels, and they are then incrementally tested until they just can't do it at any lower level.
Last edited:
Recently spent lots of time swapping preamps in my system. All else the same, eq flat every time.
I found the best sounding preamps were the ones with higher thd.
I found the best sounding preamps were the ones with higher thd.
Suggests your system might be a bit muddy. Mundorf caps in there or something like that?I found the best sounding preamps were the ones with higher thd.
Or maybe you are like wavebourn described: https://www.diyaudio.com/community/threads/euphonic-mechanisms-in-amplifiers.397918/post-7336457
- Home
- Amplifiers
- Solid State
- AFOM: An attempt at an objective assessment of overall amplifier quality