Inverting 16 bits vs 24 bits just gives noise - wrong procedure?

I'm trying to demonstrate the difference between a 16 bit version of a file and a 24 bit version.

If I take a 24 bit recording and mix it down to 16 bits with no dither, then invert that against the original 24 bit file, wouldn't that reveal the information in the 24 bit file that is missing in the 16 bit file?  All I get is noise at about =96dB (metered) when I do.

Am I doing something dumb (again)?

[wouldn't that reveal the information in the 24 bit file that is missing in the 16 bit file?]

If I take a 24 bit recording and mix it down to 16 bits with no dither, then invert that against the original 24 bit file, wouldn't that reveal the information in the 24 bit file that is missing in the 16 bit file?  All I get is noise at about =96dB (metered) when I do.

It would only reveal the difference if there was a difference. What I think you've got is the inherent noise at the -96dB level in the 24-bit file truncated abruptly at that point (16-bit no dither) - which, if you subtract one from the other would give you - noise!

So it's worth trying to record a 24 bit file that does have something more useful in it below the 16 bit level than noise?  Hmmm - I guess if I recorded something at very low level, or created something, then it should work.  I shall try.

Hum.  If I create a 24 bit sine wave at -100dB, then mix that to 16 bits and invert against the original, I get a sine wave (I have to normalise to hear it of course).  But doing the same at -50dB produces a buzzy kind of noise, because I guess you can't have the quiet part of a sine wave divorced from the rest.

So how could one demonstrate the difference between a 16 bit and 24 bit file of the same audio unless the level of the audio was entirely contained in the last 8 bits?

So how could one demonstrate the difference between a 16 bit and 24 bit file of the same audio unless the level of the audio was entirely contained in the last 8 bits?

With some difficulty! I think that it's easier to demonstrate that dither works as a means of extending the apparent bit depth than it is to come up with a good demonstration of any difference between 16 and 24-bit resolution - certainly on a real-world signal, at any rate.

It's easier to demonstrate visibly what the difference is - a sine wave at -80dB looks a damn sight better in 24-bit than it does in 16, although that can be a bit misleading. Generally it will sound a bit better with the reproduction level turned up than the 16-bit one will, simply because the BG noise level will be lower.

This is untested, so I might be full of cr*p (my DAW is busy doing other things right now).   If I wanted to hightlight the difference between 16 bit and 24 bit, I'd do something like your -50dB sine wave test, only a little more extreme.

I'd record some real-world material at a very low level (-60dB or so) in 24 bit.  Downsample a copy to 16 bit, and then normalize them both.

[edit]  and I would just play them back - not invert and add.  I think the difference would be more apparent just by listening to them.  -60dB might not be low enough to get the best impact.