Re: Why is my Butterworth filter so noisy?

[Sergei Steshenko]

--- On Tue, 9/23/08, Julius Smith <address@hidden> wrote:

> From: Julius Smith <address@hidden>
> Subject: Re: Why is my Butterworth filter so noisy?
> To: address@hidden
> Cc: address@hidden, "Matthias Brennwald" <address@hidden>
> Date: Tuesday, September 23, 2008, 12:05 PM
> Another option is to design the Butterworth zeros and poles
> and then
> call zp2sos() to convert to (real) series second-order
> sections.
> 
> The "noise" in the original post is due to
> missing frequencies in the
> input signal, resulting in spurious numerical failures in
> the spectral
> ratio Y./X. 

[snip]

I often use the following quick and dirty solution: I add to input signal
Dirac pulse with amplitude, say, 1e-8 of max signal value.

If we are talking about real physical signals this error is negligible
- we can't measure signals with relative accuracy of 1e-8.

This Dirac pulse has evenly distributed spectrum, so holes in the real
spectrum become much less deep.

Regards,
  Sergei.