But Olive & Toole concluded in their own research that good measured performance, including low THD, is correlated positively with a good listening experience.
An amplifier that has low THD is closer to an ideal amp, i.e., an amp that provides Vout = A*Vin. This can be proven mathematically. That makes THD relevant in my view. If you would rather trust the marketing verbiage of the manufacturer that's your choice. You're free to make that choice; just as I am free to measure THD and make design decision based on those measurements.
Tom
Have you read Earl Geddes on this subject? If not, it matters how much is noise, and how much is signal. It also matters which harmonics are involved (and in some cases their relative phases, at least according to some authors). Those who have said a spectrum graph is okay are far, far closer to being on the right track. I would like to see a distortion residual too.
I report THD at 1 W and at full output power, THD+N vs output power, THD+N vs frequency, harmonic spectrum, four kinds of IMD, .....
I agree that THD in isolation, while relevant, does not tell the full story. As I've said ad nauseam, you have to look at the complete picture, i.e., many measurements.
Throwing out one measurement because it doesn't provide a complete picture is pretty silly.
Tom
Sounds like you have all the right answers. Would you care to share some references?
Tom
But noise is not part of THD. It is part of THD+N. Could it be that you're conflating the two terms?
What do you mean by that?
I don't see how, say, 0.1 % THD can be "misleading by an order of magnitude". It's not difficult to measure THD ... at least not until you push past -100 dBc or so.
Tom
Some lossy compressed audio can be 50% distorted and yet not be deemed objectionable. Even for uncompressed audio, it might take 10% or more 2nd harmonic for some people to notice. Yet, .01% 7th harmonic can be heard by maybe 20% of the population and possibly deemed objectionable. Earl Geddes points out these problems in his own words. Sean Olive was trying to make the same point. It matters a lot what the specific distortion is.
If and when we add noise in, it just gets worse.
If and when we add noise in, it just gets worse.
This is from The Perception of Distortion on Geddes' website: The Perception of Distortion
A correlation of -0.423 is a moderate-to-large correlation according to Cohen (1988). So I would argue that Geddes' own data say that THD is predictive of the subjective rating.
But of course Geddes would rather sell you on his Gm number. And it does indeed appear to show a much stronger correlation with the subjective rating in his experiment than THD or IMD, but that does not in and of itself make THD or IMD irrelevant.
A distortion pedal for a guitar distorts well over 100 % depending on the setting. Some of us - me included - like the sound of a distorted electric guitar. Does that mean that we should pass all music through a distortion pedal?
It seems like you're trying to create a dichotomy that does not need to exist. I think there's room for nuance here.
Tom
A correlation of -0.423 is a moderate-to-large correlation according to Cohen (1988). So I would argue that Geddes' own data say that THD is predictive of the subjective rating.
But of course Geddes would rather sell you on his Gm number. And it does indeed appear to show a much stronger correlation with the subjective rating in his experiment than THD or IMD, but that does not in and of itself make THD or IMD irrelevant.
That may be. But how does that apply to, say, amplifiers? Would you be happy with an amplifier that had 50 % harmonic distortion?
A distortion pedal for a guitar distorts well over 100 % depending on the setting. Some of us - me included - like the sound of a distorted electric guitar. Does that mean that we should pass all music through a distortion pedal?
It seems like you're trying to create a dichotomy that does not need to exist. I think there's room for nuance here.
Tom
Interesting that it is a negative correlation. IME sometimes people prefer a little HD to add false clarity if the system is in some way muddy sounding, which is something that can occur with some dacs without showing up as PSS distortion.
Anyway, the problem with THD is if you never heard an amplifier and you are told it has 1% THD, can you tell from that information whether most people would think it sounds good, or that most will think it sounds bad? If you can't say because there isn't enough information, then why use THD at all. The value of scientific model is its ability to predict the future.
As many here have explained to you THD is just part of a large set of measurements. E.g in AFOM by @Bonsai. Even in ASR THD is never given as single number as all devices are measured with a set of measurements. But even as a single number THD is objectively much better than nothing where "nothing" refers to subjective opinion based on sighted listening.
Earl Geddes explained it this way: https://en.wikipedia.org/wiki/Streetlight_effect
That said, I don't object to spectra nor to time-domain distortion residuals. The problem is when THD or THD+N is used alone.
That said, I don't object to spectra nor to time-domain distortion residuals. The problem is when THD or THD+N is used alone.
Mark, I really don't understand what you're trying to accomplish here. You drop that Sean Olive quote in every thread that relates to harmonic distortion and you seem completely unwilling to listen to reason. What are you trying to gain with this?
And what are you trying to say with your repeated mention of the Streetlight Effect? It sounds like you're trying to say that those of us who measure THD (along with other things) are ignorant. Is that your intended message? I hope you can see how this doesn't exactly put you in the best light. No pun intended.
While we're in the psychological corner, you might want to look up the Better-than-Avererage-Effect.
Also, as many of us have tried to explain to you, an ideal amplifier, i.e., one that provides Vout = A*Vin has no harmonic distortion. I aim to make my amplifiers close to ideal as possible, so I aim for as little distortion as I can get. As I've explained ad nauseam, optimizing for low harmonic distortion brings with it many other advantages, including high PSRR, low IMD, and in my case also low noise and low output impedance (-> high damping factor), which I measure as well.
Another value of the scientific method is that it's self-correcting. If the data show a negative correlation between the subjective rating and the harmonic distortion, then that's what the data show ... even if it does not support your subjective experience.
Tom
And what are you trying to say with your repeated mention of the Streetlight Effect? It sounds like you're trying to say that those of us who measure THD (along with other things) are ignorant. Is that your intended message? I hope you can see how this doesn't exactly put you in the best light. No pun intended.
While we're in the psychological corner, you might want to look up the Better-than-Avererage-Effect.
Yes I can actually. As you can see in Geddes' plot, there is a moderate negative correlation between the subjective rating (i.e., how much someone likes it) and the measured harmonic distortion. Higher harmonic distortion -> lower subjective rating. So, yes. I can say that an amp with 1 % THD will be rated worse than one with 0.1 % THD by a large group of people. That's what his data show. This is one of the many reasons a measurement of THD is relevant.
Also, as many of us have tried to explain to you, an ideal amplifier, i.e., one that provides Vout = A*Vin has no harmonic distortion. I aim to make my amplifiers close to ideal as possible, so I aim for as little distortion as I can get. As I've explained ad nauseam, optimizing for low harmonic distortion brings with it many other advantages, including high PSRR, low IMD, and in my case also low noise and low output impedance (-> high damping factor), which I measure as well.
Because, as I and many others have said repeatedly, it is one of the measures that describes how close to ideal a particular piece of audio gear is. And by "ideal" I mean in the Vout = Vin or Vout = A*Vin sense.
Yes. And from Geddes data I predict that audio amplifiers with higher harmonic distortion will be rated as sounding worse than those with low harmonic distortion. It is relevant to me whether my products are perceived as sounding good and therefore a measurement of THD remains relevant to me.
Another value of the scientific method is that it's self-correcting. If the data show a negative correlation between the subjective rating and the harmonic distortion, then that's what the data show ... even if it does not support your subjective experience.
Tom
Even if the 1% is 2nd harmonic? How many people like the sound of NP amplifiers? Probably lot of people prefer that sound to the sound of a Topping class-D with low measured THD, which they did to game the rating system.
OK so now we're back to "I only believe in science when science supports my opinion". Hey... You were the one who suggested I go read Geddes on the topic. So I did. His data don't support your opinion. If you have issues with how he ran his experiment, I suggest you take it up with him directly.
My question still stands: What are you looking to gain from your repeated posting of that Sean Olive quote and your repeated references to the Streetlight Effect?
Can we get back to discussing distortion of capacitors now? Or must all such discussions be derailed because you don't believe distortion is a relevant measure?
Tom
My question still stands: What are you looking to gain from your repeated posting of that Sean Olive quote and your repeated references to the Streetlight Effect?
Can we get back to discussing distortion of capacitors now? Or must all such discussions be derailed because you don't believe distortion is a relevant measure?
Tom
I would like to see is move ahead and develop better methods and metrics that correlate with what we hear. The first problem is to get people to let go of past metrics that were developed at a time when THD+N could be measured with a voltmeter and a notch filter. Whatever voltage was left with the notch in place had to be THD and or N. It was very crude, but it was done because the light was better there. It was possible to do at low cost so everyone could measure it. However, it doesn't have all that much to do with how people actually hear. It doesn't tell you if you can predict that your amp won't sound bad because of various problems that are not PSS. Bob Cordell makes that plain in chapter 16 of the 2nd edition of his book. There are mechanisms that can affect amp sound that are not PSS. Also, the order of harmonics and their phases can matter. Phase is the difference between all the different types of white noise, and they don't all sound the same. So let's try to move beyond the mid-20th century here.
Also, I do believe distortion is a relevant measure, so is signal-correlated noise, so it the statistical type of and color of random noise, and so are time-domain distortion residuals. The only time I mention a problem with THD or THD+N is when someone mentions it as though its all that matters. Its not all that matters.
Last edited:
What do you base that on?
A just as valid argument could be that it was done because an amp with low distortion and noise is closer to an ideal amp than one with higher distortion and noise. They looked there because it was relevant to do so.
A similar valid argument, based on Geddes' data, is that harmonic distortion was measured because it correlated with the perceived experience, therefore, if one designed an amp with lower harmonic distortion it would be judged to provide better sound quality than one with higher harmonic distortion.
Belcher (1974) arrived at the same conclusion. THD correlates with the perceived experience; higher THD -> lower subjective ratings. He further argued that the result of a multi-tone test (in his case, three tones) correlated even more strongly with the perceived experience, which is one of the reasons I started including multi-tone tests in my product characterization and nudged Amir at ASR to do the same.
Geddes' data says otherwise. As do Olive & Toole's data.
Did you read that chapter? All the distortion mechanisms mentioned in Ch. 16 (2nd ed) of Cordell, "Designing Audio Power Amplifiers" result in harmonic distortion. All the graphs in that chapter show THD vs frequency. Also, all those distortion mechanisms are PSS (assuming you by PSS mean periodic steady-state). I don't see any mention of the perceived sound quality in that chapter, so, honestly, you seem to be contradicting yourself here.
There's a direct link between THD and IMD. The only other distortion mechanism mentioned by Cordell in that chapter is transient intermodulation distortion (TIM) also known as slewing-induced distortion (SID). You can go back to Jung, Stephens, & Todd (1979) for a good description of the underlying mechanisms of that. It's a solved problem, by the way, which is why TIM isn't measured much anymore.
And you believe the best way to do this is to derail all discussion about harmonic distortion with that Sean Olive quote?
Tom
This is pure gobbledygook.
Intuition and a solid technical explanation aren’t the same thing. Neither is claiming DA produces distortion that can’t be measured. I would have thought a multi-tone test would have exposed the claimed problem quite easily
Are you claiming that an amplifier that measures 1 ppm distortion that uses capacitors for signal coupling is seriously degraded due to DA but unfortunately it isn’t measurable?
It is interesting for me to follow this discussion, although I do not have the theoretical knowledge like you. But that's why I've been working and repairing amplifiers for the last 35 years of my life, and that's why I have one interesting thing.
About 25 years ago, me and a friend had a small hi-fi cd shop. At that time, I made Pass's ZEN from parts from the drawer on a universal PCB, nothing exotic, just to make it work.
I brought that amplifier to the store in the listening room and it was better than all the amplifiers we had on the shelf at the time. There was also a 5000euro Krell and a top model Onkyo Integra.
Pass's ZEN, which has a much higher THD than all those amplifiers, sounded much better, which is not only my opinion, but about 50 or so people who passed through the listening room at that time.
Now how to explain that the amplifier with higher THD was better in the listening test than those with low THD in all the people who listened to the tests?
About 25 years ago, me and a friend had a small hi-fi cd shop. At that time, I made Pass's ZEN from parts from the drawer on a universal PCB, nothing exotic, just to make it work.
I brought that amplifier to the store in the listening room and it was better than all the amplifiers we had on the shelf at the time. There was also a 5000euro Krell and a top model Onkyo Integra.
Pass's ZEN, which has a much higher THD than all those amplifiers, sounded much better, which is not only my opinion, but about 50 or so people who passed through the listening room at that time.
Now how to explain that the amplifier with higher THD was better in the listening test than those with low THD in all the people who listened to the tests?
Tom, I don't entirely agree with some of established psychoacoustics. I am more in agreement with W. Dixon Ward, who's writings can be posted as excerpts if anyone is interested.
Regarding what Cordell calls distortion, he includes things like EMI-induced distortion, Grounding Distortion, Inductor and Magnetic distortions, Fuse, Relay and Connector distortions, etc. If you have any experience with some of these sorts of things they aren't always necessarily PSS effects. They can be more in the realm of signal-correlated noise since there can be some randomness or else pseudo-randomness to their behavior. A frying noise when you play music is not necessarily PSS, although it may show some of that behavior too. When such effects are partially random in nature or else partially correlated with some signal other than your test signal then the spurs associated with them should be smaller according to the extent to which they are non-PSS. If the spurs are small enough they may be deemed inaudible whether they are audible to some people or not. And this is just for amplifiers, not for more complex devices.
And that's not even getting into areas such as stereo imaging and sound stage where problems may not simply be related to distortion and noise. Asymmetrical crosstalk and or certain types of correlated noise can be significant problems that don't either show up or don't show up very well in standard distortion and noise measurements.
Regarding DA, IME it doesn't produce HD, although it can affect sound. And it can take some power-on time to reach some sort of sonic equilibrium. These things can be measured but the numbers tend to look small. Small numbers may be scoffed at simply because they are small numbers, not necessarily because they are inaudible effects.
Last for now, the reason I question great THD numbers is because of their limited predictive value, particularly into a resistive load at 1kHz only. Thus when someone says something like, "all a good amp needs is low enough THD and low enough noise and it can only be sonically transparent," my BS alarm goes off. At that point I say something about THD. What happens then is people jump out to defend THD as valuable. Well, its not all that valuable in the context I was objecting to.
Regarding what Cordell calls distortion, he includes things like EMI-induced distortion, Grounding Distortion, Inductor and Magnetic distortions, Fuse, Relay and Connector distortions, etc. If you have any experience with some of these sorts of things they aren't always necessarily PSS effects. They can be more in the realm of signal-correlated noise since there can be some randomness or else pseudo-randomness to their behavior. A frying noise when you play music is not necessarily PSS, although it may show some of that behavior too. When such effects are partially random in nature or else partially correlated with some signal other than your test signal then the spurs associated with them should be smaller according to the extent to which they are non-PSS. If the spurs are small enough they may be deemed inaudible whether they are audible to some people or not. And this is just for amplifiers, not for more complex devices.
And that's not even getting into areas such as stereo imaging and sound stage where problems may not simply be related to distortion and noise. Asymmetrical crosstalk and or certain types of correlated noise can be significant problems that don't either show up or don't show up very well in standard distortion and noise measurements.
Regarding DA, IME it doesn't produce HD, although it can affect sound. And it can take some power-on time to reach some sort of sonic equilibrium. These things can be measured but the numbers tend to look small. Small numbers may be scoffed at simply because they are small numbers, not necessarily because they are inaudible effects.
Last for now, the reason I question great THD numbers is because of their limited predictive value, particularly into a resistive load at 1kHz only. Thus when someone says something like, "all a good amp needs is low enough THD and low enough noise and it can only be sonically transparent," my BS alarm goes off. At that point I say something about THD. What happens then is people jump out to defend THD as valuable. Well, its not all that valuable in the context I was objecting to.
Last edited: