“So you have defined oversampling (see WFTD oversampling). However, the description is a bit unclear as to what happens with all the other samples. Specifically, in regard to ADC oversampling, what is done with the 64 samples to derive the 1 ‘keeper’ sample?”
Okay, we’re digging deep here. Listen folks: the process of converting analog audio to digital data and back again is quite a bit more complex than the pedestrian (or even the typical pro audio guru’s) understanding of it. A full course in digital audio theory is beyond the scope of inSync, but here is a brief layperson’s look at this part.
To specifically answer your question, the process of decimation is used after the A/D converter to get rid of the ‘extra’ samples. Decimation effectively ‘throws away’ the extra data. Why sample it at a higher rate and then throw it away, you ask? There are many reasons. Some of the main ones are filters: more gentle anti-aliasing filters can be employed, which reduce problems like phase shift and other tonal ‘annoyances.’ You get a reduction of noise because the overall noise ‘power’ (we’re talking wattage here folks) is spread out over a wider range of frequencies, the upper range of which is thrown away, thus throwing the noise away too. For example, a system that oversamples at 96 kHz, but decimates down to 48 kHz throws away half of the total noise, resulting in 3 dB less noise power in the final result (heavy stuff, huh?). Further, the process of noise shaping can be employed after sampling and before decimation, which can shift a lot of noise (such as quantization noise) further up into a region that gets taken out during decimation.
In the D/A converter side of things oversampling is basically used to ‘sample’ each of the samples coming off the tape or CD at a higher rate. One 44.1 kHz sample may be ‘sampled’ by the converter many times (this isn’t technically accurate as to what really happens, but illustrates the point). The end result is that more gentle filters are employed and things sound better.