Re: [Discuss-gnuradio] Coherent de-dispersion

[Eric Blossom]

On Thu, Jul 01, 2010 at 11:35:13AM -0400, Marcus D. Leech wrote:
> On 07/01/2010 11:31 AM, Tom Rondeau wrote:
> >> Actually, my frequency-domain coefficient generator is based loosely on
> >> some earlier Swinburne
> >>  code, and then I use the Gnu Radio FFT filter block, after turning the
> >> frequency-domain
> >>  coefficients into time-domain ones (for some bizarre reason, the Gnu
> >> Radio FFT filter takes
> >>  its coefficients in time-domain, rather than frequency domain).
> >>     
> > (I should probably change the subject, but this is just a quick
> > interlude and hopefully won't spawn a long conversation.)
> >
> > Do you really think that's bizarre? I always think of filter taps
> > being naturally expressed in the time domain. But maybe that's just a
> > limitation of my background. In the constructor of the fft filter
> > block we do the FFT to convert them. Since all of our algorithms
> > return the taps in the time domain, it makes sense to me to handle
> > both filters the same.
> >
> >
> > Tom
> >   
> Oh, I agree that it's in the tradition of Gnu Radio filters.  It's just
> that after you've finished reading
>   paper after paper about FFT filters for de-dispersion, with everything
> being expressed in the
>   frequency domain, it's a bit jolting to find that the FFT filter takes
> "taps" in the time-domain.
> 
> Perhaps "bizarre" was too stong a term :-)

It was done that way so that you could swap the FFT based filter
for the dotproduct based filter with no other changes.  They take the
same constructor args.

Please feel free to build another block that takes the freq domain
taps if you like, or wrap the existing one in a hier block that does
the conversion of the taps for you.

Eric