I’m gonna put it to you that DSD and PCM are both capable of capturing and conveying essentially perfect sound. And instead of answering “why does DSD sound better” I want to answer as if you asked “why do DSD and PCM seem to sound different?”
DSD and PCM differ in two ways: sampling frequency and encode/decode complexity.
The sampling frequency matters from a low-pass filtering perspective. Every filter degrades the audio to some degree through phase shifting, pre-ringing etc. The steeper the filter the worse the impact. Constraining CD audio to 44.1kHz sampling required the use of “brick wall” filters somewhere during recording/production. DSD, even at its lowest rate, has another six octaves of headroom so it can use a very gentle filter during A-to-D conversion and minimise those impacts. But of course PCM can be operated at higher rates too, so with care the advantage of DSD there can be caught up.
The encode/decode complexity is a physical challenge. Creating a perfect 24-bit ladder DAC requires 99.99999% precision in the resistance values of your encode/decode circuit. That’s not just me mashing the keyboard, that’s the actual order of magnitude you get when you calculate 1 - 1/(2^23). 16 bit DACs are 256 times less demanding but still, look at the effort and expense involved even in doing that in the current crop of discrete R2R DACs. Imperfections in these resistor values lead to non-linearity in the D/A conversion and that degrades the “sound” of PCM in an audible way.
Contrast that to the DS DAC which just needs two precision voltage rails (one positive, one negative) and one ultra-low-jitter clock to convert binary data into an analogue electrical signal.
For that reason, the vast majority of ADC and DAC devices which produce or consume PCM don’t actually use pure PCM techniques internally. They use a much smaller number of bits (from 1 to about 6 at most) and much higher sample rates and sigma-delta modulation with noise shaping… much like DSD.
DSD’s major downside, of course, is ultrasonic noise.
So the lines in practice are really blurry. Are you hearing the effect of brick-wall filtering in your CD audio? Are you hearing non-linearities in your R2R DAC even when decoding 24/192 PCM? Does the ultrasonic energy of pure DSD cause intermodulation distortion somewhere in your playback chain? Does your off-the-shelf DAC chip’s internal processing treat DSD and PCM differently such that converting one to the other in your PC would make you think that “DSD sounds better than PCM”?
Ultimately, I think that DXD (aka 24/352.8kHz PCM) is the format with the least compromise for recording, processing and distribution (file size and transport compatibility aside). And I think that the DS DAC – which converts DXD to DSD256 on the fly – showed us the very best architecture for D-to-A conversion at lounge room scales (with the caveat that our amps and speakers need to be OK with the ultrasonic noise that makes it through the low-pass filter).