Nyquist’s theorem only covers sampling (turning a “continuous” signal (which exists at all points in time) into a “discrete” function (which only exists at cert

Nyquist’s theorem only covers sampling (turning a “continuous” signal (which exists at all points in time) into a “discrete” function (which only exists at certain points in time)). Each of the Nyquist samples is a real-valued number which can take an infinite number of values. To get from Nyquist samples to something we can actually represent digitally we need to do a second step: quantisation. This is where we take the real- valued samples and quantise each to one of a fixed set of values. The bit-depth determines the number of values we can quantise to. The higher the bit depth the more values we can quantise to and so the less error introduced by quantisation. You can think of quantisation error as introducing a noise term which gets added onto your signal. Higher bit depth lets you either reduce the amplitude of quantisation error or increase the range of signals you can represent (this is called dynamic range). Once you get to about 16 bits the quantisation noise is inaudible in even the quietest listening environments while at the same time the dynamic range is high enough to cause instant hearing damage. So, in a similar way to Nyquist telling us that more than 44kHz sample rate is pointless, dynamic range tells us that more than 16 bit depth is pointless (for playback. Things get a little more complicated for recording and editing). Edit: as linked by another commenter, Monty from Xiph explaining it better than me: https://web.archive.org/web/20160304013821/http://people.xiph.org/~xiphmont/demo/neil-young.html https://web.archive.org/web/20160304013821/http://people.xip...