Which is the threshold for audible THD in % for untrained audiophiles ?

[CG]

THD is a red herring.

 

Measuring THD is a good practice to ensure that each unit coming off the production line is pretty much equivalent to the others and to the original design intent.  That's it.  This has been known for more than half a century.

 

I'd suggest reading this, primarily the first page and the relevant references shown in the footnotes:

 

http://www.hopkins-research.com/information/electroacoustics/A new distortion measurement.pdf

 

You are certainly welcome to disagree with this, and I won't argue with you.  Personally, I'll stick with Shorter, Brockbank, and Wass.  I've found the latter two to be quite accurate with regard to what I do in my day job, which is not the audio biz.

 

[CG]

This is from Belcher's paper.  

 

"One interesting result given by Brockbank and Wass is that if the programme signal is assumed to be represented by n tones, each of equal power, and if n is greater than 30, then the contribution to the total distortion power due to harmonic products is at least two orders of magnitude less than that due to intermodulation products. For complex signals such as those produced by speech and music, n is generally large enough for the distortion power level contributed by harmonic products to be negligible. "

 

(From http://www.hopkins-research.com/information/electroacoustics/A new distortion measurement.pdf)

 

So, harmonic distortion is largely irrelevant in and of itself.  A proper measurement of harmonic distortion over various levels and frequencies *might* provide some insight about the IMD properties, but I'm not so sure about that.

 

BTW:  Brockbank and Wass's paper is available for purchase from the IEEE.  No, it's not free.  Yes, the math still holds.

[CG]

Yes, it is caused by the non-linearity of the transfer function.

 

But, how it manifests is not exactly obvious from a harmonic measurement test or simulation.  A few hours in front of a spectrum analyzer measuring a multi-tone communications system will prove that to anybody who actually is interested.  No math analysis required.

 

My point is that while a simulation of pure harmonic distortion is interesting, it doesn't really replicate what happens due to that non-linearity in an actual system.  (System meaning a a bunch of components put together to produce a result - for audio, it's to eventually move some air based to stimulate a listener's aural system.)

[CG]

6 minutes ago, pkane2001 said:

My simulation, for example, creates a variable, user-controlled non-linearity and applies it to any test signal, including the best hi-res audiophile recording or a 1000 multi-tone test signal. There's no frequency-domain manipulation involved, other than to display the resulting spectrum.

 

Is the distortion frequency dependent in any way?

 

Slightly OT, but I'm curious...  Is there any kind of "memory" dependency available for the distortion profile?  Example: Thermal effects.

[CG]

38 minutes ago, pkane2001 said:

 

The simulation has a 'feedback' simulation that simulates the effect of a frequency-dependent negative of positive feedback. This is not meant to perfectly reproduce any specific device, but rather to let me play with various types and amounts of distortions and test myself if I can hear them. Jitter is another distortion that can be simulated and mixed with the effects of harmonic distortion, as an example.

 

OK - good to know.  

 

I'm sure you know that basic device linearity is not only a function of operating conditions, intrinsic device design, but also of frequency.  That certainly would affect the results of your simulation.  (That wasn't specifically directed at you, but at people reading...)

 

Is the jitter actual phase/frequency modulation of the entire spectrum or is it simulated by adding the resulting noise sidebands to each tone?

[CG]

1 minute ago, pkane2001 said:

Jitter in DISTORT is the actual timing error added to the desired test signal. The timing of each sample is adjusted in accordance with selected modulation noise/signal (from random noise to 1/f noise to configurable sine-wave frequencies, as well as correlated signal). The sidebands (and other distortions) produced are the effect of modifying the timing of samples, these are not artificially inserted

 

Very good!

[CG]

1 hour ago, pkane2001 said:

There was a reference posted to an old paper proposing a multi-tone test for harmonic distortion. Here's a monster test signal that I'm currently using  Not something that was easy to generate in 1978, when the paper was written! This is showing non-linearity, so plenty of harmonic and inter-modulation distortion, but all below -100dB.

 

This test looks very promising.  Audio Precision offers something similar, for a price.

 

In your example, some of the distortion tones are about 65 dB below the level of the "desired" tones.  If you add the power of all these up, it really is imposing, since there's so many distortion products.  Just as the paper spelled out.

 

I'm not the guy doing the work, so this is only me wondering aloud...

 

What would be really cool is the ability to shape the amplitude profile of the tones.  My reasoning is that music and most other sound has a definite spectral profile.  You just don't find as much 19 KHz energy in your average concert or park setting compared to 2 KHz.  So, although having the test tones be equal in level certainly stresses the system, it may not be realistic enough to provide much insight to the system as a sound reproducer, as opposed to a test signal reproducer.  Probably, somewhere, somebody has compiled a representative profile of sound amplitudes under various conditions.  Maybe that curve could be a built-in option.  You could call it PK-weighting or something like that.  Be famous in your own time.

 

In communications systems, these kids of tests are often used for the reasons you suggest.  What is done is to turn off a couple of the tones and measure the level of IMD where that tone used to be.  Often, bandpass filters are used before the spectrum analyzer so that the other tones don't cause distortion within the spectrum analyzer itself.  (A real problem, btw...)