AFOM: An attempt at an objective assessment of overall amplifier quality

This is simply not practically possible - forget it. I mean in context of what is basically discussed here.

//

"What is being discussed here" has little to do with quality. It's just a selection of measurements selected out of the blue. Nothing to do with "quality"... unless you come up with a very specific, restricted definition of quality.

I'm stating a means of measuring QUALITY in the sense of discovering a technique by which we can ultimately measure the "quality" of any audio component,,, including amplifiers.... actually, circuits.

So, there you have it, Until you have a set of measurements that truly reflect what makes a good sounding circuit, you're just peeing into the wind.

Honestly, I think the old FCC standard of power/impedance/frequency, with preconditioning, extended to a waterfall over frequency with additional IM measurements, some impulse analysis ( no preringing! ) and perhaps some additional stress tests are the best we can do with our current understanding of the psycho acoustic science. Let the listeners figure out whether they like it or not.

A de-emphasis on the subjective aspect of listening is a huge step backwards.
 
I’ve added PMA’s load to the doc as the standard test load. The values are all in there. I will have a bash at making one in a few weeks with BOM. For a committed developers/reviewers, it’s not too difficult a task.

I will try posting the updated doc tomorrow. There is a surprising amount of work and cross-checking to do.
 
Last edited:
IMHO good amp characteristic very much depends from asociated PSU quality , where PSU internal low Z output is one very important factor , especially when PSU have to supply (a)B class power amp .
I think you'll find that Class A amps are even more fussy as to PSU quality. Try some EVIL SMPS with your favourite Class A amp. They will make an audibly significant and preferable difference 🙂
 
IMHO good amp characteristic very much depends from asociated PSU quality , where PSU internal low Z output is one very important factor , especially when PSU have to supply (a)B class power amp .
Cool. But it would be good that everyone that feels the need to talk implementation also at least come up with a test, in amp external interface, that can quantify the discussed issue.

//
 
John Atkinson tests the PSU by driving the amp at high powers at a low f offset by a few Hz from the mains fundamental - 50 Hz IIRC and then he looks at the mains harmonic content in the output. With low PSRR and/or high ripple PSU’s the noise is easily discernible.
 
I offer the below document as a/one view and a way to do some definitions. It is also a sort of comment on the xls doc presented. I don't claim expertness and lack in many areas. But I felt I needed to get this out of my system. See it as a 80/20 effort (love them - inject 20% and get 80, do 80 more for the remaining 20... )

//
 

Attachments

I offer the below document as a/one view and a way to do some definitions. It is also a sort of comment on the xls doc presented. I don't claim expertness and lack in many areas. But I felt I needed to get this out of my system. See it as a 80/20 effort (love them - inject 20% and get 80, do 80 more for the remaining 20... )

//

You list gain deviation as a parameter to be measured, but that s useless since any gain deviation will forcibly manifest by a given amount of distorsion, because that s exactly that, distorsion appear as a consequence of non linearity of the gain.
 
Do you mean linear distortion? If so yes - its an other way of describing linear distortion.

Just using the word "distortion" I think many people think 2nd, 3rd.... But just to make sure... : Say that you have basically an amp that has a BW of 100k and 0,001% THD for a 1k sine. Now you introduce a 1st order LP at 100Hz - do you now not measure 0,001 at 1k any longer? Even if now the gain at 1k i lower due to the filter...

I described what we usually call "FR was down 1dB at 20Hz" as a "gain error of -1 dB at 20Hz!... that was my intention.

//
 
In other areas of electronics, a lack of gain 'flatness' as a function of frequency would not generally be called distortion.
Distortion is when you have a nonlinear transfer function. So Vout is a.Vin +b.Vin^2 + c.Vin^3 ..... and b,c...are not all zero.
So a sine wave is distorted into a wave with harmonics
Two sine waves generate 2f1 -f2 components due to expanding the equation and the 3rd order component.
A sine wave becomes a 'funny' shape. It is distorted.

A perfect/theoretical lowpass filter for example does not distort, but it has varying attenuation as a function of frequency
Two sine waves in the pass band will not (theoretically) generate 2f1-f2 components
A sine wave won't distort, but e.g. a square wave may change shape, but people like me don't call that 'distortion'.
That is linear. Only first ordr terms in the transfer function.

If gain varies with amplitude, that is not 'linear' so distortion is happening.
 
Thanks!

So you first and last statement is a bit contradicting as it says: in audio we called it distortion but not in other areas... but I get the gist.

My problem is I think that I could never make up my mind if the naming was due to the cause or effect. A non-linear function creates "new tones" distortion. 🙂

I find that I do not need to change anything in my doc wrt to this discussion. It ended up talking about errors rather than distortion which made me circumvent a potential "lingo" confusion ;-)


//
 
'Error' is another term we need to be careful with.
Multiple meanings perhaps?
An 'error signal' in a feedback loop can be a totally linear system
A phase error might be a linear filtering effect
Quantisation error might be a repeatable nonlinearity or pretty much random noise.
 
Agree - I made sure it was properly (I think) defined in the document. No term that isn't more or less problematic these days ... 😉

Also, as the doc only deals with a black box approach, most of what you mention is out of scope..

//
 
  • Like
Reactions: dkfan9
I offer the below document as a/one view and a way to do some definitions. It is also a sort of comment on the xls doc presented. I don't claim expertness and lack in many areas. But I felt I needed to get this out of my system. See it as a 80/20 effort (love them - inject 20% and get 80, do 80 more for the remaining 20... )

//
I don't want this to seem like a nitpick, but...

A fundamentally passive reference for the fundamentally active devices looks rather dubious.
This is roughly like using a reference that does not have any possible form of movement as a time reference.

There is a concept of "linear system" in systems theory, you even don't need additional "ideal" term for it because it's fundamentally ideal.
"A linear system is a mathematical model of a system based on the use of a linear operator".
The properties of a linear operator (or linear map) contain everything that is sufficient for an ideal or reference.
For example, additivity corresponds to an ideal (zero) IMD coefficient because additivity means f( x+y )=f( x )+f( y ) [weird emojis...], and if f() is a kind of ideal gain (which is a primary function of amplifier), f(x)=G*x, you have the precise description of reference.
If you pay attention to what x is (a time-dependent variable), then in terms of a linear system this variable can take on any values, the behavior of the linear system does not change. Here is a ready-made description of an ideal amplifier without clipping and without masking weak signals with its own noise.
And so on, all this has long been thought out in the theory of linear systems, including the fact that if a system is independent of time, it should be free from feedback (if you think about it, this is exactly the case).

The degree of deviation of a real system from an ideal linear one can be called an error, but this is precisely the degree of non-ideality. And there can be a lot of estimates of the degree of non-ideality, in addition, because they are precisely estimates of non-ideality, they depend on the main parameter - time (or frequency, frequency is also about time). Therefore, it is very difficult (if not impossible) to reduce each non-ideality estimate to a single value that allows unambiguous comparison.
For example:

this graph is THD(Frequency) of the Class D "single-chip" amplifier TPA3255:

Screenshot 2024-01-14 143741.png


and this graph is THD(Frequency) of the (venerable famous and bright) Quad 405:

QUAD405-thd.jpg


Both amplifiers are non-ideal (sorry for this trivia) but which is better in the sense of lower degree of non-ideality?
IDK.
And nobody knows.

And so on (I suspect it will be ad infinitum).

As for reactive load equivalents for testing amplifiers, this is just not interesting because there are even formal methods for generating models for different dynamic heads in different acoustic designs, there is no point in citing hundreds of links, but the most well-known are:
https://sound-au.com/project82.htm
https://sound-au.com/project216.htm
 
  • Like
Reactions: TNT