On 22.10.2010 06:05, Mathieu Bouchard wrote:
    On Fri, 22 Oct 2010, Martin Schied wrote:
      
      
      I wanted to use a fairly high a < 1
        because then the phase for all frequencies is approximately 90°
        off like for the ideal a=1. Using slightly smaller factors and
        comparing input / output didn't satisfy my expectations. maybe
        that would't matter at all - let's experiment...
        
      
      
      How is the phase preservation an important thing for you ? I ask
      because I didn't really think of it as important... (and I still
      somewhat don't know when it's really important).
      
    
    I think it's not important for a speaker simulation either, but my
    assumption was that a correct behavior in phase would be a "sign" of
    a working integrator.
    
      
      that's indeed interesting. So the gain is
        defined for a constant signal having the same input and output
        samples (or in other words DC having no amplification) if I
        understood it correctly.
        
      
      
      It's defined for any signal. There are different equivalent ways
      to define the gain of a linear filter. In my head, I was thinking
      of an input signal containing a single 1 in a sea of zeroes... but
      it might be a bit easier to understand it using a constant input
      signal. Then for [sig~ 1], [rpole~ 1] will diverge (as much as the
      float32 number format can allow it to...) because 1/(1-1) is
      undefined (it's a division by zéro). But for [sig~ 1] again,
      [rpole~ 0.999] will output a constant 1000.
      
    
    that's the way I understood it. fine!
    
    Martin