Re: Why is my Butterworth filter so noisy?

[Fredrik Lingvall]

Matthias Brennwald wrote:
> On 22.09.2008, at 15:26, Fredrik Lingvall wrote:
>
> > >
> > Try,
> >
> > freqz(b,a,1/(t(2)-t(1)))
> >
> > instead
>
> Ok, this will produce a beatuiful plot of the transfer function as it 
> should be. However, the noise will still be there in my filtered data.
>
> Matthias
The noise is due to the poor estimator of the transfer function that you 
have used. You don't have infinite precision in your data. If you take 
the uncertainties into account, say using a model like

y = h*x + e


where e is a "noise" term (i.e., the model misfit) and h is the impulse 
response of your filter, then you can derive a better (more robust) 
estimator for h.

/Fredrik