Quote Hugh Robjohns:
Hence a practical 16 bit audio system gives a signal-noise ratio of 93dB. A theoretically perfect 24 bit system would be 141dB, but typical implemenations are currently around 19, 20 or 21 bit resolution giving SN ratios of 111, 117 or 123dB.
24 bit converters however have any explicit dither buried under many bits of self noise so arguably the numbers would be straight 6 times effective word length (Obviously that is coming from the wrong direction as in reality it is the noise performance that is specified, not the number of bits).
In practise I don't think anyone explicitly dithers 24 bit conversion, the thermal noise inherent to the electronics is more then good enough.
96 kHz sampling, 24 bit samples: 48 kHz max freq, 141 dB dynamic range.
In your dreams
One interesting observation is that SNR is specified over a stated bandwidth, so you COULD run a 16 bit converter at way above 48K and hide all the dither energy up above 20Khz, giving you a 16 bit converter with a noise floor in the audio band better then 16 bits....
This is in fact how most modern converters operate (but most are WAY shorter then 16 bits at the input stage).
One other thing I spotted up thread was an all to common misstatement of the sampling theorem.
The actual condition is NOT that the sample rate must be twice the bandwidth to be recorded, but that it must be GREATER then twice the bandwidth to be recorded.
I think this misunderstanding gives rise to a lot of confusion when considering say a signal at 22.05Khz with a sample rate of 44.1Khz (which actually fails the test at least if the intended bandwidth extends to DC).
Audiophiles use phono leads because they are unbalanced people!