Reply by Gerold Schrutz November 19, 20042004-11-19
"Mogens Heller Jensen" <mogens@mookid.dk> schrieb im Newsbeitrag 
news:419b7fec$0$13763$ba624c82@nntp03.dk.telia.net...
> Hi group! > > I'm an engineering student and I'm trying to implement an IIR filter. It's > a 6th order Chebyshev LP filter. > > We perform L2 norm scaling of the coefficients. > > To simulate the implementation environment we programmed the whole thing > in Matlab, rounding the products in the appropriate places. > > Everything works fine, but as the final task in this assignment we are > supposed to check whether noise-feedback would be an advantage. > > Only thing is: we got a brief explanation about noise-feedback, but were > supposed to figure things out by ourselves. > > Now come the questions: > > 1)Is it true that noise-feedback is merely adding -y[n-1] (subtracting the > delayed signal, that is)? > > We did this, and we just got what seems to be the same signal, but with > much lower amplitude. > > 2) Under what conditions is noise-feedback a good idea? And when isn't it? > > > Hope you are willing to help! > > Kind regards > Mogens
Hallo Mogens, 1) No. If you do so, at least you have to adapt your coefficients accordingly. It seems to me you didn't do that (... much lower amplitude). 2) Its always a good idea, especially for a fixed point realisation with limited accuracy. Rhonda Wilson wrote a good paper in the JAES 1993, Vol 41, numer 9. Hope it helps Gerold
Reply by Mogens Heller Jensen November 17, 20042004-11-17
Hi group!

I'm an engineering student and I'm trying to implement an IIR filter. 
It's a 6th order Chebyshev LP filter.

We perform L2 norm scaling of the coefficients.

To simulate the implementation environment we programmed the whole thing 
in Matlab, rounding the products in the appropriate places.

Everything works fine, but as the final task in this assignment we are 
supposed to check whether noise-feedback would be an advantage.

Only thing is: we got a brief explanation about noise-feedback, but were 
supposed to figure things out by ourselves.

Now come the questions:

1)Is it true that noise-feedback is merely adding -y[n-1] (subtracting 
the delayed signal, that is)?

We did this, and we just got what seems to be the same signal, but with 
much lower amplitude.

2) Under what conditions is noise-feedback a good idea? And when isn't it?


Hope you are willing to help!

Kind regards
Mogens