MichaelBenjamin wrote:add dc of say 500 000 and hear if it makes a difference.
Adding a large fixed number will reduce the resolution, as you would be confining the original +1/-1 to fewer bits in the mantissa of the float number. The bigger the number you add, the lower the resolution of the smaller part that you are interested in.
As a simple analogy, imagine I have a calculator that can only show 8 digits. The decimal point can be shown at any place across the screen - so it is a sinple form of 'floating point' maths.
No problem, the decimal place moved further left, and we can still see the whole result.
Something is lost with these examples because there are no more digits left or the decimal point couldn't move any further.
Float numbers work very much like this - there are 23 bits representing the 'digits' (matissa) and 8 bits representing the 'position of the decimal point' (exponent), plus a sign bit for +/-.
Scaling (multiplying) your input wouldn't alter the precision (same number of digits), but might allow a feedback loop to go on longer before the numbers got too small (run out of decimal places). But in practice those numbers are so small, that a soundcard could never represent them anyway.The Wiki article about IEEE754
has a few useful explanations and tables about the limitations of float numbers.