[music-dsp] SVF stereo operation
kaleja at regency.estarcion.com
Tue Oct 7 11:03:11 EDT 2003
On Mon, Oct 06, 2003 at 09:35:26PM -0700, Joshua Scholar wrote:
> Like in our mixing thread, constraints make simple problems subtle.
> I just realized that my suggestion (using constant amplitude sample noise)
> isn't the perfectly general solution I thought it was.
> I though making the noise a constant amplitude and adding it at every stagem
> would make creating the occasional denormal impossible by getting rid of all
> zeros, but it turns out that in a specific case, it can give you 50% zeros
> back! If a stage turns out do do exactly nothing, or just a subtraction,
> then adding the next bit of noise can exactly cancel out the noise you added
> at the previous stage, for some samples - giving you (horrors) zeros!
Nothing wrong with zeros, though. Because of the precision
limitations of floating point numbers, the operation:
denormal + tinynumber - tinynumber
is unlikely to result in a denormal. Denormals generally occur
as a result of recursive multiplications. Finally, if you do
occasionally generate a lone denormal, it's not that big a deal;
the problematic performance hit occurs when you get into long
streams of denormals.
More information about the music-dsp