@amirm "It is the job of the DAC to be immune to that noise, jitter" I designed the Cypress (RIP) FX2LP USB2 receiver block. It is the phy's job (part of the DAC unit, but not part of the DAC per-se) to be immune to jitter (the "receive eye"). These serial protocols are spec'd such that if each piece of the chain meet spec (as any USB2 cable that costs more than $0.25 will), then yes, bits are bits. If jitter is bad enough to cause reception errors, then the loss in sound quality will be hugely non-linear. Ie, as you slowly increase jitter, music will go from sounding perfect, to very suddenly, pops, drops and all other manner of badness. The music won't slowly and gradually sound worse. You're a scientist Amir. Please don't (even accidentally) contribute to Audio Bullshit(tm). I love everything you do, but you missed the mark on this one a little by hedging when no hedge was necessary.
Glad to see other designers joining our forum.
As to your comments, what you say is only true at subsystem level, not the entire device (DAC unit). Due to shared power rails (or at least ground) and capacitive, inductive and magnetic coupling, what goes on in the USB subsystem can indeed bleed into the DAC and show up all the way on its analog output! The DAC chip clock and reference circuits can easily be polluted for example, and generate 100% correlated jitter/modulation products with the data that was fed to the DAC.
Even when eliminating the DAC, we can see this effect. Here is an example of a couple of USB to S/PDIF converters. I fed them both digital streams comprised of the j-test signal over USB. This is the results:
The J-test signal has two nested squarewaves in the form of PCM samples. The larger signal runs at 12 kHz and the smaller one at 250 Hz. We clearly see a peak at 12 kHz for both USB to S/Pdif converters. We also see an array of sidebands likely created by the 250 Hz sub squarewave.
In both of these cases the data was extracted correctly from USB interface so that jitter is not material. What is material is that the incoming digital stream has bled into the S/PDIF outbound clock. This is not good as correlated jitter is more audible than random jitter (although in above example is not material due to low amplitude).
Here is an example of a DAC telegraphing the same interference now onto its analog output:
So what looks like isolated subsystems on paper, when merged into a practical DAC (unit) implementation is anything but. Audio, despite its low speed, has incredibly dynamic range with 16 to 24 bit samples. This puts huge requirement on auxiliary circuits around the DAC chip to be quiet.
A very well designed DAC has high immunity to such cross-coupling and hence my comment.
In addition to all of the above, unless the USB bus is galvanically isolated, the shared ground with the PC happily pulls the noise on that connection and feeds it through on the ground side of the RCA connector on the output of the DAC. This is why one hears the PC activity in some systems.
In summary, as long as we are dealing with analog output of a DAC, or timing variations on a digital interconnection which by definition is analog in nature,
bits are NOT bits. The system is mixed-signal and has to deal with digital+analog. Digital-only analysis does not suffice.