My understanding is:
16 and 24 bits were chosen for processing architecture reasons, not for their initial SNR's. Original audio DSP's as we think of them these days were the Motorola 56K's - 24 bit processors. So you could process 16bit data, and still have 8 bits headroom, just in case you overflowed in your calculations etc. Registers within the device could also hold the extra overflow data.
Of course, Processors then moved to 32bit - (ADI Sharc, TI VC33 & C62x, C67x) Originally some of these processors were fixed point. So suddenly, you can improve your SNR (as ADC technology allowed us to), and still keep your mid-calculation data in the registers.
So, your data formats are pretty much set in stone - fixed bit depths which also allow for computational overflow. Your also right on your next point - even with a 120bit ADC - there are going to be 3 or 4 bits of noise.
Also, for getting more than 90dB in practice -- I think MCS is toying with some of TI's 118dB converters... It'll be interesting to see the results of his work. 100dB+ is a very attainable target these days.
I have one question for you guys in return - who do have studio's full of high end stuff. Can you really hear the difference between a 118dB converter and a 124dB converter? If the answer is "yes" - then how much is your decision biased by the frontend circuitry? How much of that is based on marketing?
Just food for thought.
Cheers
R
16 and 24 bits were chosen for processing architecture reasons, not for their initial SNR's. Original audio DSP's as we think of them these days were the Motorola 56K's - 24 bit processors. So you could process 16bit data, and still have 8 bits headroom, just in case you overflowed in your calculations etc. Registers within the device could also hold the extra overflow data.
Of course, Processors then moved to 32bit - (ADI Sharc, TI VC33 & C62x, C67x) Originally some of these processors were fixed point. So suddenly, you can improve your SNR (as ADC technology allowed us to), and still keep your mid-calculation data in the registers.
So, your data formats are pretty much set in stone - fixed bit depths which also allow for computational overflow. Your also right on your next point - even with a 120bit ADC - there are going to be 3 or 4 bits of noise.
Also, for getting more than 90dB in practice -- I think MCS is toying with some of TI's 118dB converters... It'll be interesting to see the results of his work. 100dB+ is a very attainable target these days.
I have one question for you guys in return - who do have studio's full of high end stuff. Can you really hear the difference between a 118dB converter and a 124dB converter? If the answer is "yes" - then how much is your decision biased by the frontend circuitry? How much of that is based on marketing?
Just food for thought.
Cheers
R