[Scilab-users] iir() generates unstable filters

[Matt Bruensteiner]

I've seen similar issues brought up in the archives, but I don't find any
resolution. 

I'm using iir() to generate a butterworth lpf. 

The resulting filter has poles outside the unit circle.  I am by no means a
DSP expert, but from what I understand, this indicates a non-stable filter.
When I apply this filter to a measured waveform using flts(), the result
blows up, with the output increasing exponenntially over time.

I tried both even and odd filter orders:

-->lpfilt = iir(4, 'lp', 'butt', [.0875 0], [0 0])
 lpfilt  =
 
                                       2            3            4  
    0.0030512 + 0.0122049z + 0.0183073z + 0.0122049z + 0.0030512z   
    -------------------------------------------------------------   
                                           2            3   4       
        0.2139357 - 1.3042473z + 2.6500338z - 2.5712226z + z        
 
-->roots(denom(lpfilt))
 ans  =
 
    1.0361719               
    0.6177887 + 0.5547720i  
    0.6177887 - 0.5547720i  
    0.2994733               
 
-->lpfilt5 = iir(5, 'lp', 'butt', [0.0875 0], [0 0])
 lpfilt5  =
 
                                       2            3            4           
5  
    0.0007227 + 0.0036136z + 0.0072273z + 0.0072273z + 0.0036136z +
0.0007227z   
   
--------------------------------------------------------------------------   
                                            2            3           4   5       
       - 0.1155084 + 1.2115289z - 3.1061603z + 4.3810649z - 3.226861z + z        
 
 
-->roots(denom(lpfilt5))
 ans  =
 
    1.1204851 + 0.4583414i  
    1.1204851 - 0.4583414i  
    0.4264616 + 0.6409903i  
    0.4264616 - 0.6409903i  
    0.1329677               

<end transcript>

When I do the equivalent thing in Octave, I get the same zeros but different
poles:

octave:2> [b, a] = butter(5, 0.0875 * 2)
b =

  7.2273e-004  3.6136e-003  7.2273e-003  7.2273e-003  3.6136e-003 
7.2273e-004

a =

   1.00000  -3.22686   4.38106  -3.07823   1.11122  -0.16406

(Note octave gives the coefficients in the reverse order from SciLab).

When I plug Octave's denominator back in to SciLab, the filter has the
expected features (LPF with -3 dB at f/Fs = 0.0875).

I am currently using SciLab 5.3.3 64-bit on Win 7. 
I know that SciLab 5.4 is available, but the updater doesn't play nice with
the network setup (proxies?) at my office.---I'll certainly upgrade if this
is a known bug that has been fixed in 5.4.

My question is:

* Is this a bug or do I simply not know what I'm doing with either SciLab or
DSP?
* If it's not a bug, how should I use SciLab to generate a stable filter?

Thanks,

Matt




--
View this message in context: http://mailinglists.scilab.org/iir-generates-unstable-filters-tp4026779.html
Sent from the Scilab users - Mailing Lists Archives mailing list archive at Nabble.com.