Re: [Discuss-gnuradio] rational_resampling question

[Eric Blossom]

On Fri, Feb 09, 2007 at 01:26:30PM -0800, Hans Glitsch wrote:
> Hello,
> 
> If I want to use the rational resampler without filtering, is it valid do 
> this?
> 
> taps = [1]
> self.resampler = gr.rational_resampler_base_scc( 3, 2, taps )

If by valid you mean, "will it compute what I ask it", the answer is
yes.  However, I don't think you're going to like the answer you get.

Why do you want to remove the filtering?
FWIW, it's a polyphase implementation and runs at the low rate.

See gnuradio-core/src/lib/filter/gr_rational_resampler_base_scc.cc
for the implementation.

Also, if you're trying to solve a performance problem, I _strongly_
suggest that you use floats or complex, not shorts.  
Trust me, they're _much_ faster.

Using *_scc you've effectively moved the short to float conversion
into the inner loop of the dot product.  This results in terrible
performance on most machines since the conversion is relatively slow
and you're performing it O(N^2) times.

If you're using the USRP as the front end, I suggest that you adjust
to doing most everything at complex baseband.  It simplifies many
things, and we've got very fast SIMD filtering kernels for floats and
complexes, not to mention fft-based filtering too.

Eric