On 20/07/2023 02:04, [email protected] wrote:
Thank you Marcus for the feedback. Indeed I realized just after sending the
post that I had omitted to use the external 10 MHz reference for clocking the
FPGA. Although I had suspicions on the result, I have waited to return to the
lab and run the experiment again with the correct settings before replying.
Looking at the script, it looks like you've set the *time* source to
"external", but are still relying on the internal
clock for the actual reference clock that is used to drive everything
internally--including the ADCs, FPGA
DSP bits, etc.
Corrected now, although I believe for the short measurement duration (3 seconds
for each run) the TCXO should be close enough to the White Rabbit grand master
reference to drift by a few samples at most (a 1 ppm frequency fluctuation would
lead to 3 us fluctuation after 3 s at most, or 15 samples at the end of the
record
or 5 samples after 1 s when timestamping the first 1-PPS rising edge).
The "1PPS" on USRPs is used ONLY as a 1-time synchronization point for
resetting the timestamp clocks to a known value
*at a known time* (the 1PPS pulse). That timestamp register "ticks
over" based on either the internal reference clock,
or an external one, but you haven't requested external reference clock.
As I understand using the internal reference instead of the external means I
loose
the alignement of the 1PPS edge with the clock driving the FPGA and hence might
be off by +/-1 sample or +/-200 ns when sampling at 5 MS/s. Here I observe again
4 ms fluctuations of the 1PPS edge detection (bottom chart) overnight, the full
experiment duration being 11 hours here.
Another thing I fail to understand is that when I look at the timestamp (stat
-c %y) of
the file last access time (top chart) I do get the sawtooth shape of the timer
control by
NTP and I do get the sawtooth amplitude as inverse of the linux kernel timer
rate so I'd
say that the record starts at the right time with respect to the 1-PPS rising
edge (otherwise
the file timestamp would be randomly distributed, which is not the case, see
inset)
but I fail to understand how the 1PPS edge detection in the IQ stream (bottom
chart) can
fluctuate so much if the IQ file last storage time is consistent.
Thanks.
_______________________________________________
USRP-users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
_______________________________________________
USRP-users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
I don't see a streaming time specified in the generated code that you
supplied.
Try setting a "Start Time" in your UHD source block of perhaps "2.0" --
this will arrange for streaming to commence 2 seconds
after it resets the timestamp clock. Otherwise, streaming is started
at some random time.
Also, rather than looking for the PPS edge in the data, the "File Meta
Sink" includes meta-data, like timestamps-as-seen-by-the-radio.
_______________________________________________
USRP-users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
