I played around with the AP Flex software a bit. Amir, WolfX-700 and SIY might be interested to check the below using their APx555.
AP's method of measuring THD is - these days- quite limited. Instead of simply taking the harmonics and calculating the respective THD values from within a high resolution FFT, like nearly everyone else does, and thus can offer THD measurements down to -160 dB easily, their THD measurement is limited to a lower FFT size, so can't separate noise and harmonics at levels that - again these days - even 200$ DACs reach easily (as proven by this forum).
Looking at the so called Distortion Product Ratio measurement tells it all: a stable 1st harmonic, and if below -120 dB heavily fluctuating harmonics 2 to x. Noise dominates them. So while the APx is famous for its analog high precision generator with about -150 dB THD, the APx itself is not able to measure that.
Which brings me back to my above post - I wanted to see if it is possible to add some workarounds so one can better separate THD and THD+N sweeps, to get a better impression what part is noise and what part is THD. That is indeed easily possible by reducing the captured noise.
Step 1: Using a 1 kHz sine, limit the Analyzer Bandwidth using elliptic high and low pass from 950 Hz to 10 kHz. This includes 10 harmonics, but removes a lot of the noise energy that taints the THD value.
Step 2: Set the Analyzer Settings (the detector) to Average mode instead of Flat. Average reduces noise, but keeps all steady tones on the same level as Flat does.
Pic 1 shows that these changes have no influence on normal, steady state signals and their levels. Note that red (Det AVG) includes the BW Limit.
As with the following pics this is a simple -60 dBFS to 0 dBFS sweep in 31 linear steps. The next pic shows how the ratio curve of THD is lowered by both changes by more than 10 dB. At levels of -10 dBFS THD takes over and they start to merge.
Another way to look at it is THD Level instead of Ratio. This one (IMHO) better visualizes that noise is the dominant and constant factor for most of the graph. It still shows the improvement by bandwidth limitation plus detector averaging to include less noise.
I exported the red curve from the THD Ratio measurement and imported it into the standard THD+N measurement of this DUT (Device Under Test).
While this measurement would still have the OP ask why all these graphs show less distortion at higher levels, I wonder if a measurement with such settings isn't a useful addition to the existing graphs, showing the influence of the distortion more clearly. I also expect this curve to be more low/detailed when not using a simple DUT in loopback as done here but the APx as source and analyzer.
And if you ever asked what we do on Christmas...here is your answer...