muthineni wrote in news:5065000000080000009C310000-
1005954886000@exchange.ni.com:
> I want to create a filter which has a particular shape. The pass band
> is -1 to 1 Hz. The input samples to this filter is COMPLEX WHITE
> GAUSSIAN NOISE at a sampling rate of 2400 samples per second. To be
> specific the SHAPE of the filter should be gaussian in the pass band.
> please note that "gaussian" in the last sentence refers merely to the
> shape and not the probability distribution function.
>
> Please help me in creating the filter and viewing the Power Spectral
> Density (with the proper scaling).
>
> Please also help me as to how i can change the pass band from (-1,1)
> to (-5,5) Hz.
>
> I am using Labview6i
>
Yes, you can design such a filter in Labview, but there are design
considerations you want to address before you start.
Your signal of interest (<5Hz) covers a very very small region in your
frequency spectrum because you are actually sampling too fast (yes, this
does come up every now and again), and filter design would be very
difficult. You would end up with very wide filters to get a sufficiently
fast dropoff. Thus, you need to revisit your design.
The first option would be to build a more modest filter, maybe 100 Hz or so,
then decimate down to 200 Hz and build your final filter from there.
If you really aren't interested in signals faster than 5 Hz, you should
actually build and ANALOG filter on your input before you sample, with a
cutoff of around 50 Hz, sample at about 100 or 120 Hz, then digitally filter
down to where you need it to be. I know this sounds like alot of work, but
it really is the way to get things done. In fact, if your noise signal has
components higher than around 1KHz, you should be prefiltering prior to
sampling anyway to avoid what we call aliasing, or disguising high frequency
content as low frequency content.
As for the digital filter design, FIR digital filters are nothing but
convolutions of a certain function, essentially weighted sums of the input
signal. Generally, these filters take the shape of sin(pi*x)/pi*x with a
bit of windowing to smooth out some of the ripples that are induced by
having less than infinite length in your filter. Pick up a book on digital
signal processing to help you out.
Good Luck,
Scott
