The Mola Mola Tambaqui DAC

[ray-dude]

There is a whole field of Information Theory devoted to this subject.  See for example:

 

https://en.wikipedia.org/wiki/Whittaker–Shannon_interpolation_formula

 

Directly germane to audio, with sinc-based filters used to interpolate values in upsampled music (for example, as used in Chord DACs).  The impact on time-domain accuracy/reproduction is profound.

 

It is not the upsampling that intrinsically improves SQ, but the algorithms and filters used in upsampling that improve SQ.  A high resolution file can sound worse than redbook, or it can sound like the original studio master before it got processed down to redbook.  With the later, things can be magical.

[ray-dude]

To followup, this is sometimes easier to (literally) see if you look at image processing.  There are different ways to upsample an image, all of which have pluses and minuses:

 

https://clouard.users.greyc.fr/Pantheon/experiments/rescaling/index-en.html

 

Same is true with audio.

 

The nasty wrinkle here is that these algorithms work in reverse when down sampling.  When that DXD master file (24 bit 352kHz) gets processed to generation the CD quality file you stream or buy, depending on how that is done, there would be irreversible damage done to the original digital information.  As much care needs to be taken when downsampling as upsampling.  Net net is that upsampling sometimes brings magic, sometimes is "eh", depending on what happened to the digital file WAY before it got to you.  Different filters end up working with different albums because of this.

 

With the above link, it some some image down sampling without various filters.  You can visually see what the filters are doing to the original.

 

Not a perfect analogy, but gives a sense about why all upsampling isn't the same (or all downsampling)