Forums

What is the standard output of a low pass digital filter design program?

Started by Nasser M. Abbasi May 3, 2010
Hello,

I wrote a small program for class to design a low pass digital filter. it is 
an IIR filter designed using Butterworth.

We are given the specifications, and I have generated H(s) and H(z) (using 
impulse invariance, and using bilinear methods).

What is considered as a "standard" output to print as a result of such a 
design run? I am now printing the pole locations for Butterworth and for 
H(z). I also print the spectrums (mag and phase).

I wanted to print H(s), but need to decide the best way to do this.  I was 
thinking to print only the coefficients of the denominator polynomial and 
the numerator polynomial for both H(s) and H(z).   i.e. write H(s) = 
N(s)/D(s), and H(z)=N(z)/D(z), and make a table listing the power of s or z, 
and the coefficient of this term.

Printing H(s) and H(z) itself is hard to format correctly due to size change 
as order of the filter can change with each run, but printing the 
coefficients only is easier to format as I can make a table, the length of 
table can change, but I have more control of formatting its columns.

Would this be OK? Or is there another standard format I should be aware of 
for this sort of thing? Any examples I can look at? I've seen many examples 
googling around, but each seems to generate different output, and wanted to 
check if there is a standard one.

thanks
--Nasser



Nasser M. Abbasi wrote:
> Hello, > > I wrote a small program for class to design a low pass digital filter. it is > an IIR filter designed using Butterworth. > > We are given the specifications, and I have generated H(s) and H(z) (using > impulse invariance, and using bilinear methods). > > What is considered as a "standard" output to print as a result of such a > design run? I am now printing the pole locations for Butterworth and for > H(z). I also print the spectrums (mag and phase). > > I wanted to print H(s), but need to decide the best way to do this. I was > thinking to print only the coefficients of the denominator polynomial and > the numerator polynomial for both H(s) and H(z). i.e. write H(s) = > N(s)/D(s), and H(z)=N(z)/D(z), and make a table listing the power of s or z, > and the coefficient of this term. > > Printing H(s) and H(z) itself is hard to format correctly due to size change > as order of the filter can change with each run, but printing the > coefficients only is easier to format as I can make a table, the length of > table can change, but I have more control of formatting its columns. > > Would this be OK? Or is there another standard format I should be aware of > for this sort of thing? Any examples I can look at? I've seen many examples > googling around, but each seems to generate different output, and wanted to > check if there is a standard one.
This is where engineering stops being an applied science and becomes an art. There is no standard, and there probably shouldn't be. It is up to you to balance readability aesthetics and coding time to come up with the best solution for you at this point in time. Printing the filter coefficients is probably good. For the digital filter, printing the pole and zero pairs is probably better, because the pole positions get very sensitive to filter coefficients as the filter order gets high and (for precisely the same underlying mathematical reasons) when you implement the filter you'll want to split it into a cascade of 2nd-order filters. But as I said -- it's up to you. -- Tim Wescott Control system and signal processing consulting www.wescottdesign.com
On 4 Mai, 04:51, "Nasser M. Abbasi" <n...@12000.org> wrote:


> Would this be OK? Or is there another standard format I should be aware of > for this sort of thing? Any examples I can look at? I've seen many examples > googling around, but each seems to generate different output, and wanted to > check if there is a standard one.
As others already said, this as much about art as about anything else. I would prefer to keep internal data formats as generic as possible, and instead derive whatever representations of the system is required. When I wrote my filter design program I kept everything represented as biquads internally, and derived everything else from the biquads. It is far simpler to derive a frequency response from a set of biquads than the other way around. What one ends up with, is a filter design program that designs the set of biquads, and a set of functions for converting the biquads to whatever format one wants. If somebody wants a format you don't already can deliver, all you need to do is to write nother formatting function. Very convenient. Rune