USRP2 dropping data

GNU Radio
1

Hi,

setup…
-2x USRP2s + GPS 10MHz reference + 1PPS
-Single GigE port, with USRPs on GigE switch
-Zotac H55-itx MB + i3 processor, so quite high end.
-VRT branch code. I also use PPS via DCD line to trigger when to load
time on next PPS, enable real time scheduling, ref to SMA, pps to SMA,
config_mimo(MC_WE_LOCK_TO_SMA) and use similar code to
rx_streaming_samples (but 2x USRPs).
-decimation set to 10
-feeding in same off tune carrier into RFX400s to get a sine wave, just
to observe sync.
And yet, after sampling 30 seconds of data to file from both USRPs and
the code reports 0 underruns, 0 lost packets, no S’s, I often get
dropouts/missed packets.
As I am receiving only, I gather from various posts that a single eth
port + switch should be Ok. The data captured always starts off in sync,
it just goes wrong sometime during the capture. The number of bytes
captured is normally the same for both channels too.

Hope you can help please,

2

After much ado…

The VRT branch I have always returns zero for rx_missing() and
rx_overruns(), with a FIXME comment. I notice this is now fixed in the
latest git.

I also found no problems of lost data using Ubuntu 9.10 on an older
Zotac motherboard, it wasn’t until I moved to the H55-itx, and had to
get the latest e1000 ethernet driver, that the missing data issue
appeared. I then had the problem using the later driver on the older
motherboard, which confused me. I wil try the later git code…

Checking the fractional timestamps of the received packets I found that
the receive samples handler always reports the same nitems (365 in my
case with a 10MHz sample rate = 3650), but the fractional count itself
sometimes updates (with the new intel ethernet driver) by a multiple of
this value (eg 3x), but because I’m using nitems to fwrite the data to
disk, data was being skipped. I’ve not had a chance to check the latest
git, but hopefully this is fixed.

Yes, config_mimo isn’t there, I was trying to go by memory, which isn’t
what it used to be, or ever was…:-o

Yes UHD looks the way to go, but loks to be another learning curve and
rewrite of code I’ve done already.

Cheers

----- Original Message ----
From: Per Z. [email protected]
To: David E. [email protected]
Sent: Mon, 17 May, 2010 20:30:12
Subject: Re: [Discuss-gnuradio] USRP2 dropping data

David E. wrote:

As I am receiving only, I gather from various posts that a single eth port + switch should be Ok. The data captured always starts off in sync, it just goes wrong sometime during the capture. The number of bytes captured is normally the same for both channels too.

Hope you can help please,

Discuss-gnuradio mailing list
[email protected]
Discuss-gnuradio Info Page

I have used VRT and everything has worked so far. However, I didn’t do
“config_mimo(MC_WE_LOCK_TO_SMA)” I did
“c.ref_source=usrp2::clock_config_t::REF_SMA;” and
“u2->config_clock(c);”. Is “config_mimo” really part of the host VRT
code ?

In any case, you and me should start to use UHD instead …