I’m working on a multi-channel radiometer, based on USRP2 with the
dual-DDC feature.
I’m trying to come up with channelizing structure that won’t overwhelm
my CPU–I’m using
a 6-core Phenom II 1055T, with 4GB of 1333MHz memory.
I need to be able to carve-off 4 channels, with widths between 100KHz
and 500Khz, anywhere within
25MHz of the center frequency.
I’ve tried a single 25MHz input channel, then carving it up with 4 FIR
filters. That produces massive
‘O’, no matter how sloppy I make the channel filters.
I’ve tried a pair of 12.5MHz channels (using the dual-DDC feature that’s
relatively new). That produces
massive ‘O’.
Since ultimately, I’m mostly interesting in channel power estimates, I
might just FFT it, and pick off
the bin(s) that correspond to my channels.
Is one or two polyphase filter banks going to be cheaper? And how do I
compute the taps? There
doesn’t appear to be a helper function for that in GRC 
–
Principal Investigator
Shirleys Bay Radio Astronomy Consortium
On Thu, Jun 23, 2011 at 8:32 PM, Marcus D. Leech [email protected]
wrote:
the bin(s) that correspond to my channels.
Discuss-gnuradio mailing list
[email protected]
Discuss-gnuradio Info Page
Hi Marcus,
Check out a pfb channelizer!
gnuradio-core/src/python/gnuradio/blks2impl/pfb_channelizer.py is a nice
python wrapper for the class and gr_pfb_channelizer_ccf is the C++ name.
Also if you need to carve out a a shifted part of the spectrum, use the
gr_freq_xlating_fir_filter_*** block.
These are polyphase implementations so they should be computationally
efficient. Lastly, if you need even more efficiency you can see if they
are
doing the fft implementation of filtering (not sure if they are vs
standard
computation). As you know, there is the difference between NlogN and N^2
–Colby
On 06/24/2011 01:14 AM, Colby B. wrote:
Also if you need to carve out a a shifted part of the spectrum, use
the gr_freq_xlating_fir_filter_*** block.
These are polyphase implementations so they should be computationally
efficient. Lastly, if you need even more efficiency you can see if
they are doing the fft implementation of filtering (not sure if they
are vs standard computation). As you know, there is the difference
between NlogN and N^2
–Colby
Right, so how do I set the filter taps for the PFB channelizer? Is it
just a low-pass filter, or something
else?
–
Principal Investigator
Shirleys Bay Radio Astronomy Consortium
On Jun 24, 2011, at 8:22 AM, “Marcus D. Leech” [email protected]
wrote:
C++ name.
–Colby
Right, so how do I set the filter taps for the PFB channelizer? Is it
just a low-pass filter, or something
else?
–
Principal Investigator
Shirleys Bay Radio Astronomy Consortium
http://www.sbrac.org
It’s a low pass filter where the bandwidth is the width (or less) of the
channel, W, and the sample rate is that of the input signal, which is
WxN for the N channels. By default, the channelized will output channels
of bandwidth W at sample rate W.
Tom