[USRP-users] E310 Rx Timeout with higher sampling rates
Marcus D. Leech
mleech at ripnet.com
Thu Apr 30 16:24:11 EDT 2015
On 04/30/2015 03:38 PM, Voss, Nathan J wrote:
> I am running in native mode.
> Running the 'rx_samples_to_file' application, I am able to get up to about 10Msps on the E310 before overflow issues arise. Definitely better than what I'm seeing in my own application, although I'm still surprised that it is capping out at only 10MHz when the advertised (analog) bandwidth is 56MHz. I'm also not seeing any 'Timeout' error codes from the sample application.
> For the record, my command line is:
>> rx_samples_to_file --args="master_clock_rate=56e6" --ant "RX2" --freq 315000000 --null --spb 512 --gain 25.0 --progress --type float
> I am running the stock image that comes shipped with the E310. I have seen previous posts regarding updating the SD image to the 'Dizzy Beta' image. Would that potentially have anything to do with this issue? I've double checked my code for any obvious problems, and have confirmed that my 'rx_metadata' is set with no time_spec (which could cause delay before samples start) and also made sure that the stream is successfully started in continuous mode before I call recv().
> Thanks again for your help.
The "Dizzy" images may have marginally better performance, not sure what
the differences are.
But getting 10Msps through the ARM CPU is, to be honest,
near-miraculous. The E310 has a dual-core ARM cpu, clocked at 700MHz
or so. If you
thought of it as somewhere around what a ho-hum desktop x86 machine
was 3-4 years ago, that would be about right.
You'll have to optimize your own application, use standard profiling
tools to find out where your hotspots are.
But also, it is intended that high-rate processing on the E310 happen on
the FPGA, not on the Zynq CPU cores. That's why E310 has a larger
FPGA, and supports RFNoC. With current technology, there's no way a
couple of ARM CPUs are going to be able to "keep up" at 50Msps, even
for relatively-trivial stuff. Even on current-desktop hardware,
just recording samples to an SSD at 50Msps is *challenging*.
More information about the USRP-users