It doesnt matter if the values represented by the 16 bits are -32768 up to +32767, or 0 up to 65535. There are still 2^16 different levels, whatever offset you use to count them. The minimum possible amplitude is that of a signal that only takes values one unit apart. It doesn't matter where those adjacent values fall. They could be 0 and 1, or 777 and 778, -345 and -344, or if unsigned perhaps 42069 and 42070. The perpetrator of the 15 bit dynamic range nonsense in the referenced vinylengine thread was saying something about 0 being below the noise floor therefore the minimum signal has amplitude 2 bits. This is complete nonsense. First of all, what is the so called noise floor they refer to? It seems like they already assume that a one-bit amplitude signal is already overwhelmed by noise? What noise?? Secondly, as I said, even if it could possibly make any sense that the level called 0 is unuseable as a local minimum or maximum, (it makes no sense, but let's humor them and pretend it does ), any offset signal that varies between two consecutive codes both on the same side of zero does not include the (evil noisy) zero code. Thus we have an example of a meaningful minimum signal of one unit amplitude and of course the maximum is the signal with extremes of -32768 and +32767. Or in the unsigned case 0 and 65535. Either way signed or unsigned the ratio of max peak-to-peak to min peak-to-peak is 65535 = 2^16-1 = 10^(16 log_10(2))-1 = approximately 10^4.816. The base 10 logarithm of this number is 4.816, so stated or in decibels,
the ratio is 20*4.816 = 96.32 dB