After getting the RME ADI-2 DAC, I have been speculating as to a technical explanation regarding my observation that the Marantz HD-DAC1 sounded better (or should I just say, "different") playing back USB Spotify Premium (upsampled to 192/24) than when playing back CDs through either optical or coax. My speculation does not have to do with the sample rate per se, or the difference between SPDIF -vs- USB. It has only to do how digital filters and sample rates interact.
With DACs like the Marantz that do not allow a user to choose a filter (and if necessary, a PEQ fix for the rolloff that some filters induce), the designers picks a filter for you. If they pick a slow filter, the 44.1 input will be rolled off. But with a 192k sample rate, the slow filter (if I understand the guys at RME correctly) will show very little roll off in the audible spectrum. If, on the other hand, they pick a fast filter, the frequency response for the 44.1 input will be flat but the output will have a different impulse response. Either way, the 44.1 output will be different than the upsampled 192k output. For example, if a super slow filters is chosen, the 44.1k signal will begin rolling off at 5k and will be down 3 dB at 20k. I am pretty sure this is AB/X audible for someone like Amir who is trained in discerning these differences. Whether the significant impulse response differences are detectable via AB/X I don't know. I just know that after playing around with these filters for CD playback, I definitely have my preferences.
Does this make sense to you guys that are well-versed in signal processing?
I would also suppose that it is possible for the DAC implementation to be smart and use a different filter for lower sample rates than for higher sample rates. But I don't know if this is typically done.
As an aside, RME actually does use two different filters by default. At or below a sample rate of 192k the users-configured filter is applied (with Short Delay Fast being the default.) Above 192k RME overides the user configuration and applies a Slow filter. I found this out in a discussion with the RME MC. I pointed out that they did not document this. He took note and said that they will eventually get around to documenting the behavior.