GPS Timepulse Error (questions for timenuts)

[rpetitho]

I'm currently working on a GPSDO and had a question of how you interpret the spec on this GPS time pulse accuracy from the datasheet. It's from a ublox MAX-8 datasheet.

Accuracy of time pulse: 
RMS  30 ns
99% 60 ns

The datasheet position accuracy is 2.5 meters (CEP, 50%, 24 hours static, -130 dBm, > 6 SVs), which I attribute to about 9ns time error.  The unit is clocked at 26Mhz. So period is about 40ns. So I'll assume it will have quantization error +- 20ns. That gives me about 30ns error in-line with the RMS error. The other 30ns (to make 60ns) must be occasional internal artifacts or poor GPS satellite locations? Does this sound about right? And what about that last 1%?

The reason I'm asking is that I'm starting to program my GPSDO and my OCXO is single oven and still subject to some temp drift; it will be used in location with temp variation. To minimize error from temp drift I have to correct the OCXO at a faster rate, to minimize error from PPS jitter I have to correct the OCXO at a slower rate. I'm trying to best understand my GPS chip to find the right balance.

-Bob

[openloop]

While calibrating my 53131A from ublox pps signal I fed period readings into a spreadsheet.
Looking at several hours of data showed all kinds of things.
Like missed pulses, some unexplained glitches, shifts which I think happen when ublox switches between satellites, etc etc

My recommendation: if you want to know your source - collect lots of data.

[thinkfat]

Doing things "faster" in a GPSDO usually messes up the short term stability. I'd recommend isolating the OCXO (or the whole GPSDO) against temperature variations as good as possible. Doing temperature corrections accurately is not easy to achieve.

[Wim13]

Short time a GPS is bad, for the long time the GPS is excellent.

I have done some measurements, is the short (1- 30 minutes) i had about 100 nS variations. (30 meters)
Also saw some day and night shifts of  50 to 100 nSec. Has to do with radio path lengh, and radio reflections

You can also test this with plotting you own location for a longer time.

Also depends on how many sats you can see at the samen time, more is better.

So locking a OCXO, to a GPS needs a good algoritm for the long term.

[dietert1]

There used to be special antennas for this application, and it also depends on antenna location. We have a small dome antenna on the roof, brand is Lucent, 26 dB gain.

Regards, Dieter

[MosherIV]

That gives me about 30ns error in-line with the RMS error. The other 30ns (to make 60ns) must be occasional internal artifacts or poor GPS satellite locations? Does this sound about right? And what about that last 1%?

Hi.
Yes, I agree with you about the 30ns RMS error.
No idea what they are trying to say with the 99% 60ns

As thinkfat said, try to make your design as thermally stable as you can.
When testing, put the whole thing in the final case with lid on, this will give you accurate thermal effects.
Normally, you would leave the ocxo on for at least an  hour to warm up and thermally stableize.

From my understanding, normally your would use averaging over long periods (hours) to compensate for the 1pps gitter. Thermal effects are relatively slow (multiple or even ten minutes), the Oven Controlled part is meant to increase this further to multiple tens of minutes (after the warm up time). So after an hour, you can start to work out the ocxo thermak drift but the gps jitter may not be averaged out with only 1 hour, it may take multiple hours to do that.

[Bad_Driver]

I try to illustrate this GPS 1PPS-jitter topic by my simple measurement last week.

You see attached a 4h measurement with my new Siglent SDS2000x+ of my 4 U-Blox GPS receivers.
Trigger is on the 1 pps of a Neo m8n, the others are 3x Neo m6n.
Display persistents is unlimited. The m8n has an outdoor antenna on the roof, the other 3 suffer from their simple ceramic antennas
on my bench.

You can see the gausian distribution of the time pulses. I ordered more outdoor antennas and will repeat this little experiment and keep you
informed. Hope this helps.

To improve a good OCXO by GPS you need a very long feedback time constant > 1000 sec. From my investigations it seems better to
avoid a permanent PLL feedback loop. It could be enough to do a manual adjustment every week or so.

[openloop]

Bad_Driver, very cool experiment!

You can see the gausian distribution of the time pulses.

Definitely not Gaussian (with capital G - respect for the greats!)

Histogram is truncated to +/- 250ns so mean and stdev numbers are bogus.

I think, that using some robust averaging (e.g. median) is a good strategy here.

[dietert1]

I think mean and stdev numbers are completely valid. The small distribution diagrams show that the statistics is represented very well. Nobody cares about some outliers, on the contrary. The whole idea of median is to suppress outliers.

Regards, Dieter

[thinkfat]

I try to illustrate this GPS 1PPS-jitter topic by my simple measurement last week.

You see attached a 4h measurement with my new Siglent SDS2000x+ of my 4 U-Blox GPS receivers.
Trigger is on the 1 pps of a Neo m8n, the others are 3x Neo m6n.
Display persistents is unlimited. The m8n has an outdoor antenna on the roof, the other 3 suffer from their simple ceramic antennas
on my bench.

You can see the gausian distribution of the time pulses. I ordered more outdoor antennas and will repeat this little experiment and keep you
informed. Hope this helps.

To improve a good OCXO by GPS you need a very long feedback time constant > 1000 sec. From my investigations it seems better to
avoid a permanent PLL feedback loop. It could be enough to do a manual adjustment every week or so.

It's not reasonable to assume an OCXO can be "better" than GPS over a week. Good OCXOs are "better" than the 1PPS up to maybe 1000 seconds, but not beyond. Eventually you'll see them drift.

[FriedLogic]

  Ublox receivers can usually usually report the quantization error, and correcting for that really can improve the shorter term stability and make the filtering easier.

  It will, however, still wander plenty - particularly with a poor or indoor antenna - since there are various other errors going on over various timescales.

[velik_kazakov]

One good project that may help here. It average 1pps signal for 1000 seconds and drive the OCXO. Very good and symple idea for me.
https://www.instructables.com/id/GPSDO-YT-10-Mhz-Lcd-2x16-With-LED/

[thinkfat]

  Ublox receivers can usually usually report the quantization error, and correcting for that really can improve the shorter term stability and make the filtering easier.

  It will, however, still wander plenty - particularly with a poor or indoor antenna - since there are various other errors going on over various timescales.

The quantization error correction gets you a +/- 10ns (typically better) error range with a ublox timing receiver (m8t, in my case). Without the correction, you can easily add another +/- 20ns, and some more if you use a general purpose GNSS without over-determined clock mode. I've attached a screenshot from my homemade GPSDO, the "TIC" graph shows the 1PPS error that goes into the PLL calculation, in nanoseconds. Time axis is in seconds, so the graphs show the most recent 25000 seconds.

[rpetitho]

Thank you all for the data. Here is my addition. I have my GPSDO built-up and been collecting data. I did an experiment today comparing outdoor vs indoor antenna location. The graph shows OXCO generated PPS vs GPS PPS (quantization error corrected) in blue. It also shows the voltage control. It is obvious that the antenna location will make a huge difference if trying to use a PLL type of control. Looks like I need to build up an outdoor antenna if I want a good setup .

Outdoor max location deviation: 3.4 meters
Outdoor STDEV: 0.62 meters

Indoor max location deviation: 31.6 meters
Indoor STDEV: 5.65 meters

Antenna was on top of an outdoor umbrella with pretty good view of the sky in the outdoor setup, and the antenna was sitting in my garage with the garage door open for the indoor setup.

[ Specified attachment is not available ]