Interesting. So you're saying that thd at 20khz correlates with other types of distortion, like TIM? There's probably an obvious answer to this, but why not directly measure the other types of distortion instead of guessing that they will be audible? Also does the relationship between supra sonic THD and "other types" of distortion hold true for all amps? Or is it just some amps with a specific design? What about class D? Thanks
20kHz harmonic distortion correlates with TIM, specifically, as TIM appears when the required slew rate is about 10% or so (as a rule of thumb) of the amps slew rate- with lots of caveats based on specifics of the amps architecture.
There are a couple of tests that seem to correlate reasonably well with human perception- one is the magnitude of the high order odd harmonics- for example ninth- they don't seem to be well masked in the auditory process and can be perceptible even at relatively low levels. These harmonics are often generated in the application of negative feedback and the sensitivity grows at a rate higher than the harmonic ratio- for example a 9th harmonic might need to be > 40dB lower than a just perceptible 2nd harmonic in order to be imperceptible.
The feedback ratio generally (but not always) falls with increasing frequency in the audio band due to finite GBW products and finite gain for the amps, and the relative increase in harmonic distortion at 20kHz can provide a measure of this- or you could just look at the data sheet ;-).
The best test for correlation with audio quality seems to be a non-evenly spaced many tone input signal with a 1/f power spectrum to emulate music.
This input signal should extend to say, 40 kHz, as out of audio band signals can IM back into the audio band.
This type of input can degrade the SINAD of the output signal. introducing a sea of spurii that causes the effective noise floor to increase.
In the case of real music these randomized spurii appear as audible "fuzz", degrading resolution and imaging and introducing dischordance into the music. Well designed amps, even those with global negative feedback- if the feedback factor is large enough- or those with zero negative feedback, will pass this test and will be subjectively transparent, in my experience. Generally bipolar input stages are worse than FET ones due to the larger number of meaningful distortion components that are generated by the exponential gain characteristic versus a near square law gain characteristic.
However, the higher the feedback factor the smaller the input signal to the distorting input stage is, which reduces the distortion generated and hence the distortion generated by the distortion due to the application of the feedback loop- which reduces the generated higher order harmonics etc. (3rd harmonic distortion in the output, for example, generates 9th harmonic also when the feedback loop is closed).