"Thanks Dan. I just wanted to make sure I was clear. My understanding is as follows: Assuming 16 bits as a final destination to be monitoring for, technically, if I put 16 bits out through a 24 bit conversion, I have 8 bits, 48 db, before I get down to losing real information digitally….”If you put 16 bits out through 24 bit conversion? I assume you mean if you start with 24 bits AD. Well, you seem to insist on using theoretical gear. There is no such thing as a real 24 bit AD. The last few bits are noise, and they do not count. In fact, very often a real mic- mic pre – AD does not even yield 16 bits. That is what matters! There is NO WAY you have 48 dB extra!!! In most cases you have 0dB to spare!!! Many if not most CD’s out there do not meet the 96dB noise floor. And than comes the DA, and the dither…
“Where I run into a problem is that the analog noise floor of my convertor, if it is very good, will probably be somewhere around -118, so I really only have 4 bits, 32db, before I get into my 16 bits of info. Is this correct?”If your final format destination is 16 bits, a DA with 17-18 bits performance will not be the limiting factor. The 16 bit format will be the limiting factor.
”My one additional question do we continue to hear low-level info behind the noise of the convertor, or is it completely lost? I've assumed this is analog noise, but is there a digital component to the noise that makes it an actual barrier instead of a mask?”No you do not! I saw a lot of such claims about hearing below the noise floor, and that is marketing BS. You do not hear under the noise floor unless you are willing to play marketing tricks with definitions. When we talk about 6dB/bit, or 96dB dynamic range for a 16 bit machine, we are using a definition that “lumps” everything into one number.
That 96dB number is the COMBINED NOISE ENERGY AT ALL THE AUDIO FREQUENCIES. If you look at an FFT plot, the noise level at each individual frequency is at about -136dB. So when you put say a -100dB 1KHz tone there, it is much higher than -136dB and you can see it on the FFT. You can also hear it if the volume is turned up. But it does not mean you can hear under the noise.
Think of a forest with 100 young trees, each one is 1 foot tall. The combined height is 100 feet, and it may serves to describe the general amount of wood. In fact, you can have 50 trees at 1/2 foot and 50 at 1.5 foot and have the same general number 100 feet combined.
Do you want to claim you are taller than a forest with 100 feet height?
So assuming we are talking about n bits, and also assuming the FFT noise floor is flat (equal length trees), the FFT noise will be about 43dB lower (for 22KHz audio bandwidth 20*log(sqrt(22050))=43dB) than what we normally refer to as noise (the number combining the noise at all the frequencies).
“The idea of quantization distortion and when it would set in isn't really something I understand and I was wondering at what point it would become an issue.”Absolutely, quantization distortion will become an issue and that is why I said you will need to have dither added. If you want to understand it better, there are some papers on my web
www.lavryengineering.com. Look at “Do you need 20 bits” (under the support section).
Regards
Dan Lavry