I believe the reasoning for almost all DAC manufacturer's to choose the stop band at 0.546 Fs is this (see the left hand graph and ignore the right):
...as well as only 85dB (on datasheet) stopband attenuation and rather strong passband ripple. The former cannot be fixed after recording, the latter is fixable like what RME did in the ADI-2 Pro series. I tried to make an anti-ripple filter for similar AKM ADCs as well so the correction can be applied to other interfaces.
I remember that feature. It was a pain in the butt trying to explain it to customers when the sound difference was nil.
The DAC chips have choices because there are applications where it matters, and it's cheaper to make a single chip with selectable filters than to manage inventory of multiple pre-programmed versions. Given this situation, offering the tweaker crowd something to play with takes very little effort. Once one manufacturer starts doing it, the rest pretty much have to follow in order to maintain "feature" parity.
His job is to make things that sell. If having more knobs sells, they'll add more knobs.
Like what I mentioned on the first few pages of this thread, measurements and various quotes from some authorities won't make sense without involving listening and psychoacoustics. There are indeed some audio formats with configurable filtering options and the differences are really audible by everyone, and these settings are applied to the format decoder, not on the DAC.
The graphics is indeed from a TI ADC presentation. The D/A process is similar, but with the (normally assumed) guarantee that the digital input is 'properly sampled' (i.e. meet the band-limiting requirement of the sampling theorem). Here is some similar DAC graphics from Maxim that also shows the reconstruction filter roll-off starts at Fs/2.
Got it. So then we are back to the filter being fine if the digitizing is limited to below ~0.46*fs (and probably still fine otherwise since low levels would be expected at these freqs anyway). But if there is content approaching fs/2 then having more attenuation at fs/2 makes sense, and so OP has a point that this would be a better tradeoff. Sorry, I know all this was discussed already but I was hoping that we weren't deciding the debate on this point was won by the chip makers.
The image at 25 kHz is inaudible. It can only become a problem if the input content just below Nyquist is sufficiently strong that intermodulation with the images reaches audible levels. That's not going to happen with music. There is thus no problem. If it was a problem, intermodulation between perfectly legal frequencies near fs/2 would already be making a dog's breakfast of things.
Several times stronger passband ripple and about 20dB poorer stopband attenuation. BTW, the screenshot you showed looks like the
