opamps THD+N/IMD/SNR measurements don't mean jack IRL, so let it go humm'kay?

[Head Injury]

Quote:

My current thinking is that if there is any relevant point to the discussion of measurements versus perception, it is exemplified by the point above, that (ignoring biases), finding measurements that explain why we perceive what we do is difficult and regular RMAA tests, and even scope measurements, don't tell the full story. The graph on that page, however, suggests to me that the 5532 under load has higher distortion in the treble, which might explain why some people perceive it, in some devices, to produce an unpleasant sound.

It's still under 0.003% at 20 kHz with a 500 ohm load. If you're suggesting that's audible, I'd need proof.

 

[BlackbeardBen]

Quote:

It's still under 0.003% at 20 kHz with a 500 ohm load. If you're suggesting that's audible, I'd need proof.

But what about intermodulation distortion?  That's just as, if not more, important than harmonic distortion thanks to the non-harmonically related distortion products.

 

[Head Injury]

Quote:

But what about intermodulation distortion?  That's just as, if not more, important than harmonic distortion thanks to the non-harmonically related distortion products.

Don't know what kind of distortion it is, the link only says "distortion". Presumably it's THD because they talk about THD on the previous page. The point is that Currawong pointed to that graph and said it might cause unpleasant sound. If it's THD, I highly doubt it. If it's IMD, I don't know. If IMD is high, let's see a graph of that instead.

 

[Willakan]

  
 
Quote:

 
My current thinking is that if there is any relevant point to the discussion of measurements versus perception, it is exemplified by the point above, that (ignoring biases), finding measurements that explain why we perceive what we do is difficult and regular RMAA tests, and even scope measurements, don't tell the full story. The graph on that page, however, suggests to me that the 5532 under load has higher distortion in the treble, which might explain why some people perceive it, in some devices, to produce an unpleasant sound.
 
Sound cards are a good example too. I'm sure that the manufacturer's measurements are under the most ideal of conditions. Even tests I've seen on computer review sites usually involve a dedicated computer with top-of-the-line components. How they'll perform in different computers with different power supplies isn't so clear. Considering the great variety of circuits an OPAMP might be used in, I don't see why the degree of variability is any less.
 
Just my $0.02

The great thing about opamps is that their performance is easily reproducible in different circuits, with everything they need to perform as advertised detailed on the datasheet. The mere fact that they tend to survive people shoving them into random places in their favourite pieces of hi-fi equipment is testament to this. Inside a computer is a little different, but most hi-fi equipment gets much cleaner power simply from some cheap generic voltage regulators. Even in a PC, the performance differences are unlikely to be very large, considering opamps generally have superb PSRR.
 
As for the 5532 graph, it is incredibly dangerous to just look at a graph and associate it with what people heard in sighted listening. The amount of distortion in the treble appears to be very small and unlikely to be audible. Additionally, the load the opamp is tested with is a very tough one (the tester notes it is the lowest impedance the opamp can cope with): few opamps are designed to directly drive even moderately low impedances, so distortion in many applications is likely to be lower. Furthermore, the distortion reaches this level at frequencies that are more difficult to hear.
 
This would all be different if anyone had ever heard this unpleasant sound under controlled conditions: as it stands, it appears to either be a function of bias or a preference for some degree of distortion/colouration, which is and can be measured. Then we could start rewriting the book on how humans perceive sound. As it stands, what is perceived can be explained perfectly by bias mechanisms or relatively easily measurable criteria and no evidence has been presented under even the ghost of a controlled condition to suggest otherwise.

 

[Chris J]

Quote:

So you can currently while parts are still available do much better designs
all discrete.
 

 

Implied in this statement is that the discrete audio circuit designer must know what they are doing.
I suspect some designs are designed to be discrete as a Marketing decision, i.e. it looks good in an ad:  
"All Discrete Design!"
"No IC Op Amps!"
This is NOT a dig at Kevin Gilmore
 
Quote:

From that article:
 
Sound cards are a good example too. I'm sure that the manufacturer's measurements are under the most ideal of conditions. Even tests I've seen on computer review sites usually involve a dedicated computer with top-of-the-line components. How they'll perform in different computers with different power supplies isn't so clear. Considering the great variety of circuits an OPAMP might be used in, I don't see why the degree of variability is any less.
 

Douglas Self alludes to this in his book "Audio Power Amplifier Design Handbook":
Chpater 8, Power Supplies and power supply rejection.
Chapter 13:   PCB design, grounding and mechanical design.
Poor PCB layout and poor component selection can ruin a good Op Amp based design.
Analog Devices and National Semiconductor have published many articles about this.
An audio amp design using a good Op Amp can also be ruined by adding a poorly designed discrete output stage.


Quote:

But what about intermodulation distortion?  That's just as, if not more, important than harmonic distortion thanks to the non-harmonically related distortion products.

Nelson Pass has an interesting article in Intermodulation Distortion.
Interesting, how often do you see IMD listed in Op Amp spec sheets?
http://www.passlabs.com/pdfs/articles/distortion_and_feedback.pdf
 

 

[Adda]

Quote:

Nelson Pass has an interesting article in Intermodulation Distortion.
Interesting, how often do you see IMD listed in Op Amp spec sheets?
http://www.passlabs.com/pdfs/articles/distortion_and_feedback.pdf
 

Very interesting article, it proves that my preferences in audio gear has merit, apart from what my ears already told me.
 
Feedback systems trade THD for IMD, opamp systems have feedback.
 
No feedback systems may have higher THD, but low IMD, most good discrete systems are no feedback designs.
 
High circuit complexity adds distortion, to counter this you can use feedback, but this adds IMD.
Opamp systems are complex circuits by their very nature, so they need feedback to counter this, but that adds IMD.
Discrete circuits tend to be very simple, so they are not dependent on feedback to attain low distortion levels.

 

[Willakan]

It is really nothing like that simple. Lots of Feedback = Audible IMD is simply not true.

 

[BlackbeardBen]

 
 


Quote:

Originally Posted by Adda /img/forum/go_quote.gif
 
 
Quote:

 
Nelson Pass has an interesting article in Intermodulation Distortion.
Interesting, how often do you see IMD listed in Op Amp spec sheets?
http://www.passlabs.com/pdfs/articles/distortion_and_feedback.pdf
 

 
Very interesting article, it proves that my preferences in audio gear has merit, apart from what my ears already told me.
 
Feedback systems trade THD for IMD, opamp systems have feedback.
 
No feedback systems may have higher THD, but low IMD, most good discrete systems are no feedback designs.
 
High circuit complexity adds distortion, to counter this you can use feedback, but this adds IMD.
Opamp systems are complex circuits by their very nature, so they need feedback to counter this, but that adds IMD.
Discrete circuits tend to be very simple, so they are not dependent on feedback to attain low distortion levels.

Actually, I posted a link to that article earlier in this very thread...
 
And I have absolutely no idea how you came to the conclusion that that white paper states that, "Feedback systems trade THD for IMD, opamp systems have feedback."  That not at all in any way what Nelson Pass was talking about.  In fact, Mr. Pass stated that feedback reduces all forms of linear and non-linear distortion (although of course not necessarily shaping it in a pleasing manner, at least not if you don't reduce it low enough).  Rather, the increase in IMD he shows is the result of multiple gain stages with the same harmonic distortion.  He doesn't compare negative feedback versus no feedback as far as IMD is concerned - and he definitely does not state that IMD is increased with feedback.  Like I said already, he says that all forms of distortion (linear and non-linear) are decreased with negative feedback, although distortion complexity is increased of course.  But you can see that with sufficient negative feedback, you eventually may reduce even the peaks of harmonic distortion below the noise floor of the no-feedback design.

 

[Adda]

 
Quote:

Actually, I posted a link to that article earlier in this very thread...
 
And I have absolutely no idea how you came to the conclusion that that white paper states that, "Feedback systems trade THD for IMD, opamp systems have feedback."

I'll read it again...
 
Edit: well I just mixed up IMD and high order harmonics.

 

[SanjiWatsuki]

The basic conclusion is that massive amounts of negative feedback is the best and results in no higher level harmonics, feedback in moderation from 0dB to 70dB creates higher level harmonics which are reduced as feedback increases, and no feedback results in no higher level harmonics but more distortion at the more natural lower level harmonics (Fig 10). Furthermore, these higher level harmonics can combine when a complex combination of tones (the basic definition of music) is played to create a large amount of distortion. Feedback in general reduces the overall measured distortion. Very cool stuff.

 
In this way, we're both right. If negative feedback is engineered incorrectly and in the wrong amounts, it will add higher level harmonics which can combine to very high amounts of distortion and make it sound worse, despite having better measurements. With properly implemented negative feedback, this effect can be entire mitigated, it's just a matter of engineering enough negative feedback or a circuit which mitigates the potential for more higher level harmonics. Still, feedback is not evil. 
 
It is worth noting that in the complex tone graphs we are dealing with much higher amounts of power and an 8 ohm speaker load. Op amps struggle with this and I agree 100% that op amps should not be used in that context and that discrete amplifiers have the advantage. The raw numbers suggested by the article will likely not be close to the actual numbers experienced by an amp based design for headphones. 

 

[Willakan]

Luckily, unless I'm much mistaken, opamps use utterly monstrous amounts of negative feedback - I can rest in peace in the knowledge that, contrary to the opinion of leeperry, the NE5532s that fill my gear are not eating away at sound quality.

 

[leeperry]

 

If negative feedback is engineered incorrectly and in the wrong amounts, it will add higher level harmonics which can combine to very high amounts of distortion and make it sound worse, despite having better measurements.

Which nicely sums up my assertion that measurements don't tell the whole story, and that throwing low THD+N/IMD and high SNR figures at your customers doesn't prove a g*oddamn thing as far as subjective listening is concerned. The only way to know whether something sounds "good" is to listen to it with your own ears, how empirical I know. Thanks for the heads up, much appreciated

 

[SanjiWatsuki]

It's not a universal, but most measurements of the complete circuits of the amplifiers in the good low-cost op-amp based amplifiers show very little distortion at upper harmonics. For example, the Asus Essence STX, which is often quoted, measured with a 0.00002% or -133dBr distortion at the 4th harmonic. This is not a complete guarantee, though! There are plenty of poorly designed amplifiers which can show high amounts of harmonic distortion at the higher harmonics into low ohm loads, like headphones. Granted, the Essence STX also has to do with the electrical grounding the computer itself, but it shows a significant amount of negative feedback for the most part. Most of the issues with the upper level harmonics distorting the sound only would occur with a smaller amount of feedback.
Quote:

Luckily, unless I'm much mistaken, opamps use utterly monstrous amounts of negative feedback - I can rest in peace in the knowledge that, contrary to the opinion of leeperry, the NE5532s that fill my gear are not eating away at sound quality.

 

 

[SanjiWatsuki]

Sorry for butting heads with you so much before, I just didn't understand the points you were trying to make with the sources you quoted. I completely see your point about relying entirely on THD+N specs. I don't 100% agree that the only way is to listen with your ears first for these types of things, but I do now think it is impossible to tell the whole story with only THD+N, IMD, and SNR figures without having the entire graph in front of you.
 
Quote:

 

Which nicely sums up my assertion that measurements don't tell the whole story, and that throwing low THD+N/IMD and high SNR figures at your customers doesn't prove a g*oddamn thing as far as subjective listening is concerned. The only way to know whether something sounds "good" is to listen to it with your own ears, how empirical I know. Thanks for the heads up, much appreciated

 

 

[Adda]

Quote:

NE5532s that fill my gear are not eating away at sound quality.

How can be sure of that? what else have you compared your setup to? what have you done to train your hearing to pick out differences in sound?