IIR quantization

In the matlab filter design filter box, when I click the Quantization
button after design a IIR, the freq response seems very different. I
cannot find the place where you can configure how many bits for the
system?

Where do you specifiy that you are using a 16bit or 32bit system?

thanks
Andrew

Andrew Xiang wrote:

> In the matlab filter design filter box, when I click the
> Quantization button after design a IIR, the freq response seems
> very different. I cannot find the place where you can configure
> how many bits for the system?
> 
> Where do you specifiy that you are using a 16bit or 32bit system?
> 
> thanks
> Andrew

Hi Andrew,

1) read the doc: "Working with Quantizers" in the Filter Design Tool 
Box Section of the Matlab Help. It will tell you all about it.
You should not only read it, but you should even have a profound 
understanding how quantizers work and influence the result if you 
use them.
"A Quick Guide to Quanized Filters Properties" (also a section in 
the Matlab help) will lead you directly to what you search.

2)  If you call "fdatool" in a command window, the 
"Filter Design & Analysis Tool" opens. 
Hold your mouse over the icons on the toolbar, and their meaning 
will appear. One is named: "Set Quantization Parameters".
If you activate "Turn quantization on", you see all quantization 
parameters which you can set/modify.

HTH,
Bernhard

-- 
before sending to the above email-address:
replace deadspam.com by foerstergroup.de

ahgu@yahoo.com (Andrew Xiang) writes:

> In the matlab filter design filter box, when I click the Quantization
> button after design a IIR, the freq response seems very different. I
> cannot find the place where you can configure how many bits for the
> system?
> 
> Where do you specifiy that you are using a 16bit or 32bit system?

If you are making an IIR filter for fixed point math, I *strongly*
suggest using a lattice implementation.

-- 
Johan KULLSTAM <kullstj-nn@comcast.net> sysengr

Johan Kullstam wrote:
> 
> If you are making an IIR filter for fixed point math, I *strongly*
> suggest using a lattice implementation.
> 

It depends. Lattice structure has generally larger quantization noise
artifacts then cascaded direct form biquads.

Vladimir Vassilevsky, Ph.D.

DSP and Mixed Signal Design Consultant

http://www.abvolt.com