@solderdude has basically provided the answers - I/V conversion, filtering of out-of-band noise and aliasing, diff/SE conversion, combining multiple channels, and achieving the desired output level.
OP, if you want to know more about the challenges that the output stages tends to face, look at some datasheets for really
old parts (found e.g. on
Datasheet Archive). CS4303 is a good one, as it shows how much ultrasonic noise a high-order delta sigma modulator will spit out (and these are far from gone, as any CS4272 powered interface will attest - see e.g. Behringer UMC202/204HD, Focusrite Scarlett 2i2). It's about the worst case, as I don't think they ever went beyond 5th-order modulators, instead more modern designs tend to be up to 5 bits with various tricks to preserve linearity at 1-bit level (see e.g.
the CS43122 whitepaper for a typical early-2000s kind of design). The function of oversampling filters is briefly touched on as well, though if you want to know more about those, I might recommend DF1700 or SM5803. CS4328 is another good one to look at, the first DAC with all switched-capacitor lowpass filtering (discussed in
The Audio Critic, issue 21 - the TDA1547 had had first-order switched-capacitor filtering before but nothing to that extent). For the challenges of traditional resistor ladder DACs, see e.g. PCM56.
Give a man a hammer, and he'll find
new and innovative ways of hitting his thumb...
The "NOS" / "ultra slow" filter option is only intended for use at 384 or 768 kHz, combined with external upsampling to essentially replace the built-in digital filter. For example, you can use Foobar2000 with the SoX resampler plugin for some of the best upsampling money can buy (but actually doesn't have to since they're
free). In fact, I may be inclined to stack two instances of the resampler DSP, the first to e.g. 48 kHz to get rid of any ultrasonics (those needing to cater to their feline or canine audience may have to go to 96k
), and the second to 384k or 768k. Whatever aliases remain beyond 360 / 744 kHz would easily be taken care of by analog filtering, which is the entire point of using oversampling in the first place.
Looking at filters in time domain is rarely ever useful, unless you're Julian Dunn and trying to pin down pre-echo (which, however, generally is so small you wouldn't even see it at a normal size). Thankfully Monsieur Fourier is telling us that for any impulse response, there is an exactly equivalent set of frequency and phase responses. What is most prominently seen in impulse response relates to what's happening around filter cutoff, and that in turn generally is not the very most interesting area when we're talking 44.1 kHz and above.
I could still see some mid/side going on, but I/Q? Where is hearing supposed to be hiding an
SSB detector? As far as I'm aware the inner ear rather is sort of a frequency binning affair above the bass range (a bit like an FFT, but with variable bin size so frequency perception becomes roughly logarithmic). When it comes to hearing models, I'd recommend starting with Messrs. Zwicker, Fastl and Blauert for the time being.