earnmyturns
Headphoneus Supremus
I'm by no means an expert here, just working from my very basic understanding of signal processing... A PCM DAC receives a sequence of digital samples, and creates a continuous waveform from them. Each sample ideally represent an instantaneous measurement of the original analog waveform, with the samples regularly separated in time at the sample rate (such as 44.1k samples/second). What happens to the analog waveform between the sample points? That's the job of the reconstruction filter in the DAC. Now let's go to the ADC that created the samples. Each sample is an ideal point measurement of the captured waveform, but in physical reality, any such measurement is averaged (smeared) over a short time interval (think of the shutter speed in a camera for an analogy, and how a fast moving subject gets smeared if the shutter speed is too slow). So, the ADC needs a means of dealing with this smearing that avoids creating audible artifacts — stuff not on the original waveform — especially for fast transients — such as the leading edges of percussion or plucked string sounds. That's where some kind of digital filter comes in, such as the "inverse megaburrito" filter that @Baldr mentions.
You can see how the ADC and DAC filters are in some sense each other's inverse: the ADC filter figures out how to create instantaneous samples that reflect the full analog waveform in a desired way, while the DAC filter works to recreate the full waveform from the instantaneous samples.
For anyone who has heard of Nyquist's theorem all of this may seem redundant, but the theorem is not quite in force in the physical world because real music and real electronics are not strictly band-limited as the theorem requires.
