To pick up where I left off with the comparison of the FiiO KA17 and the Hiby FC3 (or the investigation of the KA17's failings that make it sound veiled/warmer), I will note that I heard the difference between them a few times again while playing through headphones (HE-400i) but haven't gone through the pains of setting up a proper ABX test at a quiet time of day and in good biopsychological condition.
Meanwhile I thought to do more measurements-based comparisons to try to get more data on what tends to be different between these dongles' output and maybe get a better sense of how I should choose the right material to ABX.
Since I didn't have much success measuring device FRs with the Multitone tool, I thought to switch things up and put white noise through each device and loop it back in repeatedly, to show any defects better amplified and easier to spot. Easier to do with just Audacity and a music player. So for the Hiby FC3 what I got was what I already knew:
* it has no warm tilt, average magnitude response is flat all across the audio band
* in-band ripples are present but they're unlikely to be audible based on known research: 0.05 dB peak-to-peak on the FR curve (about 0.2 dB after 4 loopback passes as shown in the image) - though I wouldn't mind seeing this tested in listening comparisons by "golden ears" vs. some other DACs with even flatter filter response
View attachment 441496
OK, KA17, it's your time to shine:
View attachment 441497
* visible veiled tilt, mostly beyond 7.5k, overall about 0.2 dB from 0 Hz to 18 kHz in 4 passes, so about 0.05 dB when listening to music
Could this by itself be all the difference that I heard? It's true it only came out in some parts of some songs but still...
But wait, what's this? Why does the oscillogram look like that? That's not what I expect when looking at a white noise track, and it's not what the FC3 recordings looked like even after 4 passes:
View attachment 441498
Did it take that shape over time? What does the 1st pass look like?
View attachment 441499

6-second period infrasonics? WTF?
What if I put just a 1 kHz tone through and look at the FR?
View attachment 441500
Subharmonic distortion lobes of some kind, going up beyond -60 dB.

Unfortunately Audacity doesn't allow frequency zoom - that I know of - so I can't tell if the 0.167 Hz infrasonic component was produced directly or is some beat frequency between those other lobes.
Now bass distortion is the least audible of all, and this could just be the underlying mechanism that leads to that overall very mild warm tilt, but still... no well-behaved DAC should do this. What if it causes weird interactions with other frequencies depending on the song (and song section!) being played, resulting in audible artifacts beyond just a vague tilt? Could we see something if we put a song through and DeltaWave it vs. the original file?
View attachment 441501
OK, same vaguely warm tilt going into the treble, before the noise let through by the filter starts to dominate... but also massive sub-bass spikes.

Like, even if these aren't audible because my headphones can't produce them at the right volume, I'd bet they can affect how the rest of the frequencies are presented, by periodically pushing the membrane too close to xmax.
Also, the delta of spectra in spectrogram form gives another hint: the KA17's problems seem to come out more in the simpler sections of the song with the more sparse spectra (wherever you're seeing the thicker red lines in the sub-bass and extra energy shades going up across the whole band is where most instruments and voices went quiet).
View attachment 441506
Was this a one-off? What if I re-record the same song today?
View attachment 441507
Nope, pretty much the same. Weird sub-bass spikes galore. (Notice the fit quality doesn't make a difference, since we're not analyzing the time-domain delta, only the spectral delta.)
So going forward what this leaves me with is I should try more songs, and especially focus on more minimalistic ones - less symphonic, less white-noise-like - and check their spectra in DeltaWave, to find the best section of the best song to put in my ABX test later.
But regardless of what result I will get there, I think these spectrograms already show the hypothesis that "all you need is a frequency sweep or white noise test" is far from correct: even at today's level of technology DAC/amps can respond quite differently depending on the material they're being fed, showing
changes in tonality even within the same song.