[music-dsp] Truncation noise: Mathematics (for Nigel ;-)

[Payan, Remi]

Hi dear all,
As a follow up to the discussion we had with Nigel, I have found a paper 
from Jon DATTORRO, written for the 7th AES international conference.
It's on the front page of his homepage.
http://www.stanford.edu/~dattorro/

Part I details diffferent noise shapers techniques for IIR biquads, leading 
to noise cancellation in particular cases.
One can conclude from this that using 64x32=64 mpys for poles is 
the equivalent of a second order truncation noise canceller (page 3), 
when you set Ki (error feedback coefs) equals to bi (poles coefs).

BTW, what would be a good method for estimating the truncation noise ?
May be computing a reference output with Matlab and comparing the results ?

Here is what I have in mind:
1) Compute a reference output with Matlab, everything in double precision.
2) compute another reference output still with Matlab, but rounding the
coefficients 
to 32 bits representation. This would show the noise induced only by
coefficients 
truncation.
3) Run the filter. Comparing the output to (2) would isolate truncation
noise induced by 
truncation in calculation. Comparing to 1 would give the whole noise induced
by 
calculation plus coefficients rounding.

Is this correct ?

Remi

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://music.columbia.edu/pipermail/music-dsp/attachments/20011129/0d6f4ccb/attachment-0001.html