in theory 48khz allows for a 24khz bandwidth of which they say humans only have about 20khz range of hearing. so the extra 4k would be for good measure but the reality is all of this makes a number of assumptions that are not true in a real system.
recognizing nyquist theory is one thing but if you think for a second about how that sampled waveform will look you see that 2 samples per cycle is really not enough even at a minimum. take a 1hz sine wave and sample it 2 times per cycle. it is entirely possible that both of your samples fall at EXACTLY the two zero crossings that occur in every cycle once on the way up and again on the way down and your reconstructed output becomes a flat line. by this it becomes pretty clear to see that 2 samples per cycle really is not enough at all to reconstruct your sine wave. nyquist theory assumes a number of ideal conditions, none of which hold in a real sampling system. it merely represents an absolute bottom line minimum spec requirement if you want to represent ANY of the information at a certain frequency but not necessarily all of it in all circumstances. infact given 2 samples per wave taken at any point in the cycle i can draw for you an infinite number of possible near sine waves of different amplitudes that pass through those points. lets assume the amplitude of the signal matters. by raising the sample rate beyond the nyquist rate you quickly begin to remove the singularities and the multitude of waveforms that fit the samples for a given frequency and you allow breathing room for the realities of the limitations of a real sampling system, such as non-ideal pre-ADC bandpass and post-DAC reconstruction filtering.
if we assume 20khz is a reasonable limitation of human hearing without regard for the fact that our amp, speakers, etc further attenuate the response at these edge of spec frequencies
then 48khz is really still an absolute minimum. in my opinion 96khz gives plenty of breathing room to counter all of the additive realities of a true system from end to end.
Watch this video. I am beginning to think it should be required viewing before one can post about digital audio.
Not singling you out personally, but your post contains several fallacies which are repeated millions of times. Watch and understand this 23 minute video, and you will get why they are false. Not false just in theory, but in actuality. Two or more samples is enough to fully reconstruct the signal, and one and only one waveform fits any possible combination of samples as long as no frequencies exceed half the sample rate. Yes, filters are imperfect, and reconstruction falls just a bit short of perfect theory. But most the of the important factors have been dealt with. We can get something like 95% or more of what is predicted by theory and put the last few percent of inaccuracies in a place where humans do not hear them. Effectively, audibly very, very close to 100% fidelity to humans in actual use.
You can go to 96 khz if you just really, really, really want to be sure. There certainly seems close to zero point, and beyond 96khz there is zero point. It has never been shown in a credible repeatable test that people hear 96 khz vs 48 or even 44 khz. Null results so far.
Edited by esldude - 6/10/14 at 10:30am