Originally Posted by George Mills
You can do 0.5bits though. The 24bit value is really 48bits internally (floating point I believe) they subtract 3dB (or even 1dB, what ever) and dither it. I don't know if dither is done before or after it's converted back to integer.

I have to disagree, the notion of half a bit is impossible.
1 bit is equivalent to 6dB (or x 2), 3dB isn't equivalent to 0.5 bits though.

If you had a full scale 24 bit signal;

FFFFF (Hex) = 16777215 (Dec) = 111111111111111111111111 (Bin)

-0.5dB is the same as x 0.944, so applying this to our full scale signal;

F1A9FB = 15837691 = 111100011010100111111011

The MSB is still used.

-3dB is the same as x 0.708, so applying this to our full scale signal;

B53F7C = 11878268 = 101101010011111101111100

The MSB is still used.

-6dB is the same as x 0.5, so applying this to our full scale signal;

7FFFFF = 8388607 = 011111111111111111111111

The MSB is lost, and as 011111111111111111111111 = 11111111111111111111111, it can be seen that the equivalent of 1 bit has been lost, or 24 bits has been reduced to 23 bits.

I'm sure you know this, but thought it might help explain it to those who didn't know.

Steve.