Sean -
Sean: "With digital, these "samples" are taken at very specific but limited points along the data path. One is not taking in ALL of the data, but simply "sampling" it and basing their deductions on averages and trends."
No, that's not correct. The Shannon sampling theorem says that your digital samples at 44.1khz or any other data rate will give perfect reconstruction of the analog waveform as long as there are no frequencies present above the Nyquist frequency (which is equal to 1/2 the sampling rate). The reason that oversampling is needed with a 44.1khz signal is not because of poor sampling of the waveform shape***, but rather because of the need to avoid sharp (brickwall) filters in the analog domain. With an oversampled signal, it is possible to use a gentle analog filter and do the steep brickwall cutoff at 22.05 khz in the digital domain, where it is easier to do well.
As far as the difference between oversampling and upsampling, they are the same thing but with different implementation. The traditional 'oversampling', which usually means increasing the sample rate by 8x through interpolation, is done in the d/a chip of the player. The quality of digital filters available in dac ic's is usually not high. When 'upsampling' is used, part of the 8x oversampling (typically 2x or 4x) is done in the processor chip before the data is sent to the d/a chip, and the final 4x or 2x will be done in the d/a chip to complete the total 8x oversampling. The difference here is that the quality of the oversampling filter implemented in the processor can be much higher than those used in the d/a chip.
*** Following your train of thought, if the 44.1khz data samples are not adequately capturing fast transients or peaks, it is necessarily because those peaks include frequencies above Nyquist. Such frequencies can only be recovered with higher basic sample rates (dvd-a, sacd), not by the upsampling of the redbook data. (There are some fine points about jitter, noise and data precision, but I doubt that's what you are arguing.)
Sean: "With digital, these "samples" are taken at very specific but limited points along the data path. One is not taking in ALL of the data, but simply "sampling" it and basing their deductions on averages and trends."
No, that's not correct. The Shannon sampling theorem says that your digital samples at 44.1khz or any other data rate will give perfect reconstruction of the analog waveform as long as there are no frequencies present above the Nyquist frequency (which is equal to 1/2 the sampling rate). The reason that oversampling is needed with a 44.1khz signal is not because of poor sampling of the waveform shape***, but rather because of the need to avoid sharp (brickwall) filters in the analog domain. With an oversampled signal, it is possible to use a gentle analog filter and do the steep brickwall cutoff at 22.05 khz in the digital domain, where it is easier to do well.
As far as the difference between oversampling and upsampling, they are the same thing but with different implementation. The traditional 'oversampling', which usually means increasing the sample rate by 8x through interpolation, is done in the d/a chip of the player. The quality of digital filters available in dac ic's is usually not high. When 'upsampling' is used, part of the 8x oversampling (typically 2x or 4x) is done in the processor chip before the data is sent to the d/a chip, and the final 4x or 2x will be done in the d/a chip to complete the total 8x oversampling. The difference here is that the quality of the oversampling filter implemented in the processor can be much higher than those used in the d/a chip.
*** Following your train of thought, if the 44.1khz data samples are not adequately capturing fast transients or peaks, it is necessarily because those peaks include frequencies above Nyquist. Such frequencies can only be recovered with higher basic sample rates (dvd-a, sacd), not by the upsampling of the redbook data. (There are some fine points about jitter, noise and data precision, but I doubt that's what you are arguing.)