Re: [Discuss-gnuradio] Higher Accuracy Metadata Header

[Paul Garver]

Hi Michael,

I’d be interested in your timing results once you fix the issue. We’ve done something similar in [1] at 25MSPS with a mean difference between B200s of around a sample (~40ns). I presume the B200 minis are on different computers due to the sample rate. Another issue could be in how you are calculating the time delay. We transmit pulses on the PPS and use a tool in gr-analysis (gr_fileman) [2] to chunk up the files from multiple B200s based on the time tags in the metadata header. Then, you can just to cross-correlation, take the magnitude of the DFT, pick the peak, and interpolate (or upsample). That’s what we do, anyways. I would be very much interested to hear your results.

PWG

[1] http://lists.ettus.com/pipermail/usrp-users_lists.ettus.com/2016-March/018979.html

[2] https://github.com/garverp/gr-analysis

From: Michael Skaggs <address@hidden>

Subject: Re: [Discuss-gnuradio] Higher Accuracy Metadata Header?

Date: April 25, 2016 at 10:09:16 AM EDT

To: Marcus Müller <address@hidden>

Cc: address@hidden

Hey Marcus,

In GRC, I have a USRP Source block with "Sync" set to "Unknown PPS" and "Timing Source" set to "External". I assumed the device would be setting time based on the external PPS using these parameters. The synchronize(S0) LED goes solid (after about half a second of recording, maybe less) and the other LED (S1) is blinking with the PPS.

Thanks,
Michael

On Fri, Apr 22, 2016 at 11:56 AM, Marcus Müller <address@hidden> wrote:

Are you sure you're setting the device time correctly? 2.5ms offset is definitely much much worse than what to devices that should have the same device time should exhibit. So:
How are you setting the device time?

Best regards,
Marcus

On 04/21/2016 08:59 PM, Michael Skaggs wrote:

Aha! You were correct. I was parsing the header incorrectly. However, even after parsing, and the more accurate values (now accurate to 10e-6), it appears the recorded RF data sets are still are offset from each other by an amount of nearly 2.5e-3 seconds. Any idea why this would be? It's strange, especially considering that the are synchronized to the same PPS, and I think the time is with reference to the pulse.

Michael

On Thu, Apr 21, 2016 at 11:57 AM, Marcus D. Leech <address@hidden> wrote:

On 04/21/2016 11:41 AM, Michael Skaggs wrote:

I'm trying to time/sample synchronize RF recordings with two B200minis. I am using the detached Metadata File Sink in GRC. Both recordings are at 30MSps and both B200mini boards are synchronized to the same 1PPS signal.

My issue is this, when I extract the data from the Metadata header file, the "rx_time" value is only accurate to 10e-4 seconds (0.0001s). Which, with a recording at 30MSps, this will only give me a sample alignment accuracy to 10e-4(s)*30(MS/s) = 30,000 samples.

If I'm attempting to align the two recordings by samples or time, this is not nearly accurate enough. Is there a way that I can get more accuracy out of my metadata header or a way that I can synchronize the recordings of these B200minis?

Thanks,
Michael

The precision of the timestamps from UHD should have a precision of whatever the master-clock is on the device--how are you interpreting
  the rx_time tag?  It's two parts--a uint64 with the full-seconds portion, and a double-precision float for the fractional part.



_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

From: Marcus Müller <address@hidden>

Subject: Re: [Discuss-gnuradio] Higher Accuracy Metadata Header?

Date: April 25, 2016 at 10:31:24 AM EDT

To: Michael Skaggs <address@hidden>

Cc: address@hidden

Hi Michael, 
I think I see a bug in gr-uhd there!

So point is that the python code generated by this is (example):

[...]
        ##################################################
        # Blocks
        ##################################################
        self.uhd_usrp_sink_0 = uhd.usrp_sink(
                ",".join(("", "")),
                uhd.stream_args(
                        cpu_format="fc32",
                        channels=range(1),
                ),
        )
        self.uhd_usrp_sink_0.set_time_source("external", 0)
        self.uhd_usrp_sink_0.set_time_unknown_pps(uhd.time_spec())
        self.uhd_usrp_sink_0.set_samp_rate(4e6)
        self.uhd_usrp_sink_0.set_center_freq(1.5e9, 0)
[...]

Now, what GRC generates here means that you first set your time, then you set your sampling rate – and on B2xx, that might have the side effect of changing your master clock rate. The MCR is the speed at which the internal time counter is incremented.

You can avoid that auto-MCR behaviour by explicitely setting the master clock rate first – could you add "master_clock_rate=30e3" to your device Address or device Arguments?

Best regards,
Marcus

On 04/25/2016 04:09 PM, Michael Skaggs wrote:

Hey Marcus,

In GRC, I have a USRP Source block with "Sync" set to "Unknown PPS" and "Timing Source" set to "External". I assumed the device would be setting time based on the external PPS using these parameters. The synchronize(S0) LED goes solid (after about half a second of recording, maybe less) and the other LED (S1) is blinking with the PPS.

Thanks,
Michael

On Fri, Apr 22, 2016 at 11:56 AM, Marcus Müller <address@hidden> wrote:

Are you sure you're setting the device time correctly? 2.5ms offset is definitely much much worse than what to devices that should have the same device time should exhibit. So:
How are you setting the device time?

Best regards,
Marcus

On 04/21/2016 08:59 PM, Michael Skaggs wrote:

Aha! You were correct. I was parsing the header incorrectly. However, even after parsing, and the more accurate values (now accurate to 10e-6), it appears the recorded RF data sets are still are offset from each other by an amount of nearly 2.5e-3 seconds. Any idea why this would be? It's strange, especially considering that the are synchronized to the same PPS, and I think the time is with reference to the pulse.

Michael

On Thu, Apr 21, 2016 at 11:57 AM, Marcus D. Leech <address@hidden> wrote:

On 04/21/2016 11:41 AM, Michael Skaggs wrote:

I'm trying to time/sample synchronize RF recordings with two B200minis. I am using the detached Metadata File Sink in GRC. Both recordings are at 30MSps and both B200mini boards are synchronized to the same 1PPS signal.

My issue is this, when I extract the data from the Metadata header file, the "rx_time" value is only accurate to 10e-4 seconds (0.0001s). Which, with a recording at 30MSps, this will only give me a sample alignment accuracy to 10e-4(s)*30(MS/s) = 30,000 samples.

If I'm attempting to align the two recordings by samples or time, this is not nearly accurate enough. Is there a way that I can get more accuracy out of my metadata header or a way that I can synchronize the recordings of these B200minis?

Thanks,
Michael

The precision of the timestamps from UHD should have a precision of whatever the master-clock is on the device--how are you interpreting
  the rx_time tag?  It's two parts--a uint64 with the full-seconds portion, and a double-precision float for the fractional part.



_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio