I was able to run the following flowgraph:
This runs at 4Msps complex, doing 131072-bin FFTs, giving a roughly 30Hz
resolution in the FFT bins. The de-interleaver
block “dispatches” out to 4 different FFT blocks, and I ran my tests
on a dual core Pentium D 925 with 667-Mhz DDR2
memory. I used the default “TPB” scheduler.
It handled the flow graph just fine without giving any USRP overruns,
until I upped it to 8Msps, at which point, I was getting
a lot of uOuO.
This was part of the followup I promised to do a few days ago. My
replacement mobo for my quad-core system arrives on
Monday, so I’ll re-try this experiment at higher bandwidths to see
what happens. Unfortunately, I can’t get FFT sizes greater
than 131072, without provoking shmget() errors.
This is actually a fairly encouraging result, assuming I didn’t botch
anything. I tested the results using my 21cm receiving setup,
and my 21cm signal source, and things appear to be showing up
correctly in the FFT outputs recorded in the output file.
Principal Investigator, Shirleys Bay Radio Astronomy Consortium