impact of noise in zero-padded region of filtered time trend data

On Wed, 17 Oct 2012 23:28:47 -0500, all4dsp wrote:

>>> It is a bit unconventional, and there are pitfalls to what you're
>>> doing
> 
>>> (mostly involving the fact that your sampled time sequence input and
>>> output aren't what a continuous-time filter either sees or spits out).
> 
>>> As long as your sampling rate is high enough compared to your filter
> (and
>>> signal) bandwidth there's a good chance that you're OK.
>>
>>I agree.
>>
>>-- glen
>>
>>
> I understand one wants the sampling rate higher than the signal
> bandwidth to prevent aliasing. But, what problem arrises as the sampling
> rate approaches the filter bandwidth? And, generally, what would be a
> good minimum ratio of sampling rate to filter bandwidth?

Wow.  Good question.  I just know it's something to be avoided.

Hmmm.

Come to think of it, if the filter is going to do anything significant to 
the signal, it'll have to have a bandwidth considerably (over 10 times) 
narrower than the sampling rate.  Otherwise the filter's frequency-domain 
response isn't going to have the chance to drop much toward zero.

-- 
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com