Hi Stephan, some prelims first: Can you manually sync the devices using set_time_unknown_pps() before running the flow graph? I would like to rule out VITA time issues.
Also, your way of setting the time should work; it's the same code as if you were using tx_time tags. Although, that's something worth testing: Can you inject tx_time tags onto the tagged streams with increasing time stamps? I.e., generate a tagged stream every 1 second, and increase the time stamp in 1s increments between bursts, instead of using your modification. This should rule out any simple bugs. Some background: Assuming the VITA clocks in the two USRPs are in sync, and any time is written into the metadata, one of two things happens: 1) The packet arrives on time, i.e. before the VITA clock on the device reaches the time in the timestamp, and then the packets are transmitted at the same time, or 2) The packet arrives late -- in that case, UHD will eventually print an 'L', which I assume it doesn't in your case. Cheers, Martin On 17.11.2015 07:56, Ludwig Stephan (CR/AEH4) wrote: > Hello list, > > I have a really tricky question that we have been working on during the last > few weeks. I am asking for any advice how to handle this issue: > > In short: Samples input into the USRP sink do not appear at both hardware > outputs at the same time, but are set off by several sampling periods. > Meanwhile, we got the system synchronized when using tagged streams, but if > we use a message strobe as a periodic trigger, sync is lost again. > > The long story: > We are trying to synchronize samples (not carrier phase) in time at a > transmitter, which consists of 2 N210 (SBX daughter board, w/o GPSDO) > connected by a MIMO cable. We managed to synchronize the samples for the > transmission of a "packet_len"-tagged stream, which does not work > out-of-the-box, by adapting the source code of > https://github.com/gnuradio/gnuradio/blob/master/gr-uhd/lib/usrp_sink_impl.cc#L476 > We changed the USRP_sink_impl.cpp file in such a way that is reads the > current URSP time right before transferring the data to UHD/USRP and attach > this time plus a buffer time as a tx_time tag. > > We inserted the following lines after line 476 > _metadata.time_spec = get_time_now() + ::uhd::time_spec_t(0.15); > _metadata.has_time_spec = true; > , where 0.15 [ms] is the time buffer which shall care for the transmission > latency, in order to have the hardware clock have a time smaller the tag's > value. > > Question #1: Do you see any negative effects by this change? > > This works for a simple transmission with tagged streams. > > But: As soon as we switch to a transmission, which is triggered by a periodic > message strobe, the time alignment is lost. The setup is: > message strobe (with PDU content) -> PDU to tagged stream (packet_len) -> > unpack K Bits (convert 1 Byte to 8 Bits) -> tagged stream multiply length tag > (by 8) -> UChart to Float -> Float to Complex (as real part) -> both data > inputs of USRP sink > > By measuring both antenna RF output signals (modulated at 500 MHz carrier > frequency) with an oscilloscope, we determine a timing offset of some 12µs > for long (<500) packet lengthes, which is pretty much constant; but which > signal is ahead of the other seems to be random: one time antenna 1 is 12µs > ahead of antenna 2, the other time it is vice versa. > > The timing offset depends on the the packet_len,if it is chosen small > (1..500); the 12µs seem to be an asymptote. > Even more strange is that this delay is constant during a run of the model, > but changes (randomly?) from execution of the model to execution. > > The timing offset is independent from the chosen sampling frequency (at least > for large packets). > Since we intend to use 10 MHz of sampling frequency, 12µs offset results in a > shift of several samples > > Today, there has been raised a similar but different issue by Will. > Basically, I am asking the same questions but from a different background. > Since I am not sure whether our topics have the same cause, I ask in a > separate thread. > > We have an open ticket with NI/Ettus support, but they asked me to also post > this issue here, since there exists some overlap. > > Question #2: Do you have any hint to help us solve this issue? > > I noticed similar questions, which did not help us so far: > http://lists.ettus.com/pipermail/usrp-users_lists.ettus.com/2014-June/009842.html > http://lists.ettus.com/pipermail/usrp-users_lists.ettus.com/2014-April/009274.html, > which might be a hint to a work-around. > > Any help is appreciated > > Best regards, > > Stephan Ludwig > > Robert Bosch GmbH > Communication Technology (CR/AEH4) > Renningen > 70465 Stuttgart > GERMANY > www.bosch.com > > Tel. +49(711)811-8809 > Fax +49(711)811-5187845 > Mobile +49(172)5630639 > stephan.ludw...@de.bosch.com > > Registered Office: Stuttgart, Registration Court: Amtsgericht Stuttgart, HRB > 14000; > Chairman of the Supervisory Board: Franz Fehrenbach; Managing Directors: Dr. > Volkmar Denner, > Dr. Stefan Asenkerschbaumer, Dr. Rolf Bulander, Dr. Stefan Hartung, Dr. > Markus Heyn, Dr. Dirk Hoheisel, > Christoph Kübel, Uwe Raschke, Dr. Werner Struth, Peter Tyroller > > > > _______________________________________________ > Discuss-gnuradio mailing list > Discuss-gnuradio@gnu.org > https://lists.gnu.org/mailman/listinfo/discuss-gnuradio > _______________________________________________ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
