GPSDO brands?

[Housedad]

Which GPSDO units are the better or worse units? Feature sets? Symmetricon, Trimble, Oscilloquartz or any others?

[LapTop006]

Most of the major ones are fine, although one thing to look out for is newer models have newer GPS cores, which can be needed (week rollover issues just for a start).

[awallin]

check this one out:
http://www.jackson-labs.com/index.php/products/fury

datasheet has both a phase-noise plot and a long-term ADEV plot - look for those in other products also 

[edpalmer42]

If you google for <gpsdo "allan deviation"> you can find info on some of the usual suspects.

Here's some tests I've run on various units.  Normally, Standard Deviation doesn't work well on oscillators, but since these are locked to GPS, I think it's valid.

Code: [Select]

GPS Devices -- Measure & analyze the period of the 1 PPS Output

Device ............... Std Dev (ns).... Range (max-min)(ns) ... Device ...... Notes

Navsync CW12 ......... 4 - 5 .......... 20 - 25 ............... GPS Rcvr .... 1,7
Motorola UT+ ......... 40 - 55 ........ 95 - 110 .............. GPS Rcvr .... 2,7
Rockwell Jupiter ..... 10 ............  50 ...................  GPS Rcvr .... 3,7
Motorola M12M ........ 10 - 15 ........ 40 - 60 ............... GPS Rcvr .... 7

Trimble Thunderbolt .. 0.4 - 0.5 ...... 2 - 4 ................. GPSDO ....... 6,8
HP Z3801A ............ 0.1 - 0.2 ...... < 1 ................... GPSDO ....... 6
HP Z3817A / CW12 ..... < 0.1 .......... < 1 ................... GPSDO ....... 4,6
Jackson Labs GPSTCXO . 0.3 - 0.4 ...... 2 - 3 ................. GPSDO ....... 6
NEC NWM-034241-201 ... 0.1 - 0.2 ...... < 2 ................... GPSDO ....... 5
Trimble UCCM ......... 0.09 - 0.11 .... < 1 ................... GPSDO ....... 5

Results are based on multiple runs of ~ 1000 measurements each.
Sawtooth correction has not been used for any of the GPS receivers.  Where supported, it would reduce the numbers substantially.
All units were connected to the same antenna system.

Notes

1.  Sawtooth correction not supported.
2.  Most 'range' results were in this group, but there were a few at 20 - 30.
3.  Only one test.
4.  Requires external 1 PPS input.  Equipped with E1938 oscillator.
5.  Measurement made with Fluke PM6681.
6.  Measurement made with HP 5370B.
7.  Measurement made with HP 5372A.
8.  Standard parameters.
9.  Optimized parameters. (Not yet used in any tests)

[tautech]

So you didn't get enough good info from the other thread you started:
https://www.eevblog.com/forum/testgear/what-frequency-standard-is-sufficient/

 

[edpalmer42]

So you didn't get enough good info from the other thread you started:
https://www.eevblog.com/forum/testgear/what-frequency-standard-is-sufficient/

 

I know that this question is intended for Housedad, but I'll answer it anyway.  That thread is more about which type of standard is sufficient although there was discussion about a couple of specific GPSDOs.  I think that a separate thread is appropriate.  I've posted my data before, but I'd like to see some similar measurements for some of the newer models.  I have no intention of trying to buy one of everything!

[texaspyro]

but I'd like to see some similar measurements for some of the newer models.  I have no intention of trying to buy one of everything!

Well, uhhh,   I pretty much have at least one of each.  I'm in the process of testing them all in the same config.  Using a TAPR TICC clocked by a HP-5071A cesium beam oscillator.   PPS goes to channel A and the 10 MHz divided down to 1PPS goes to channel B.  GPSDOs are allowed to warm up for at least 24 hours.

[edpalmer42]

but I'd like to see some similar measurements for some of the newer models.  I have no intention of trying to buy one of everything!

Well, uhhh,   I pretty much have at least one of each.  I'm in the process of testing them all in the same config.  Using a TAPR TICC clocked by a HP-5071A cesium beam oscillator.   PPS goes to channel A and the 10 MHz divided down to 1PPS goes to channel B.  GPSDOs are allowed to warm up for at least 24 hours.

I thought of you when I wrote that!    But for anyone else who's reading this, you don't need a cesium to make the measurements.  As long as your reference has negligible aging during a 1000 sec. run, that's good enough.  My counter was referenced to my Efratom FRT Rb house standard for my measurements.

[texaspyro]

My runs last over 18 hours and go to tau=20,000 seconds...

[edpalmer42]

You don't need to go to 20K sec. to characterize a GPSDO.  You know that if you average over a long enough time period, the frequency will be dead on.  You can measure Allan Deviation, but they all resemble each other pretty closely - particularly at large tau values where they all converge due to the impact of GPS disciplining.  The only place you'll see differences is in the central part of the graph where the OCXO and the disciplining parameters affect the graph.  For a good example, look at http://www.leapsecond.com/pages/gpsdo/ .

What's easier to measure and for most users more useful, is short term jitter.  All you have to do is measure the period of the 1 PPS and calculate the standard deviation and min-max range.  Do it a few times to catch outliers and you're done.  If you're lucky, your counter does all the calculations for you and you just write down the results.  Normally, you can't measure Standard Deviation of an oscillator because drift messes up the math.  That's why they came up with Allan Deviation in the first place.  However, if the oscillator is locked to GPS, there's no drift so the math should work.

Texas, I know that you know all this, I'm preaching to the other readers.

[texaspyro]

You don't need to go to 20K sec. to characterize a GPSDO.

Actually you should...  a lot of GPSDOs seem to want to occasionally go a bit bonkers.  If you are only looking at 1000 seconds of data, you can easily miss the shenanigans.  And to be really safe make sure you include the time around 00:00 UTC (and particularly around the time-of-week rollover)... evil spirits like to come out an play around the witching hour.

[edpalmer42]

You don't need to go to 20K sec. to characterize a GPSDO.

Actually you should...  a lot of GPSDOs seem to want to occasionally go a bit bonkers.  If you are only looking at 1000 seconds of data, you can easily miss the shenanigans.  And to be really safe make sure you include the time around 00:00 UTC (and particularly around the time-of-week rollover)... evil spirits like to come out an play around the witching hour.

I agree with you, but that's a never-ending chase.  There will always be weird 'corner cases' where strange things happen.  Are you going to include entry to and exit from holdover?  What about the next 1024 week rollover?  Yeah, I know that's getting silly, but you get my point.  Regarding 'evil spirits' and 'the witching hour', do we have to make special measurements on Halloween?   

IMHO, the 1000 second measurement, particularly if you do a few, will give you good confidence in the performance capabilities of a GPSDO under normal operating conditions.

After all, look at my list.  There aren't enough types or measurements to be definitive, but if you rank them by those numbers, do you disagree with the order?  Actually the Trimble UCCM is very interesting.  Look at how tight the results are!  I did 7 tests of 1000 measurements each.  The Std. Dev. readings are from 98 - 110 ps. and the min-max range values are from 0.58 - 0.86 ns.  Very impressive!

[MrW0lf]

I agree with you, but that's a never-ending chase.  There will always be weird 'corner cases' where strange things happen.

Q&A red alert [Pseudo]random glitch fanboy identified. You know what will happen if such attitudes will sneak up on growing generation and logic/code ends up in vehicle software, banking, etc? I know you know but just preaching the other readers...

[edpalmer42]

I agree with you, but that's a never-ending chase.  There will always be weird 'corner cases' where strange things happen.

Q&A red alert [Pseudo]random glitch fanboy identified. You know what will happen if such attitudes will sneak up on growing generation and logic/code ends up in vehicle software, banking, etc? I know you know but just preaching the other readers...

I hear you and yes, people designing professional equipment must worry about 'corner cases' and other odd situations, but seriously MrW0lf, we're not designing life-critical equipment here.  We're a bunch of hobbyists checking out some used equipment that we bought off fleabay for use in our home labs!

Can we possibly maintain a reasonable sense of proportion and common sense?

[Scratch.HTF]

Oscilloquartz is part of Swatch who makes world class timing equipment (under the Omega brand) for the Olympic Games; so therefore, Swatch applied their area of expertise to GPSDO units.

[awallin]

Oscilloquartz is part of Swatch ...

OSA is, since 2014, part of ADVA
https://en.wikipedia.org/wiki/Oscilloquartz

[tomato]

You don't need to go to 20K sec. to characterize a GPSDO.  You know that if you average over a long enough time period, the frequency will be dead on.  You can measure Allan Deviation, but they all resemble each other pretty closely - particularly at large tau values where they all converge due to the impact of GPS disciplining.  The only place you'll see differences is in the central part of the graph where the OCXO and the disciplining parameters affect the graph.

You're making the assumption that the GPSDO is operating correctly.  The main reason for testing a device is to verify that it is operating correctly.

[edpalmer42]

You don't need to go to 20K sec. to characterize a GPSDO.  You know that if you average over a long enough time period, the frequency will be dead on.  You can measure Allan Deviation, but they all resemble each other pretty closely - particularly at large tau values where they all converge due to the impact of GPS disciplining.  The only place you'll see differences is in the central part of the graph where the OCXO and the disciplining parameters affect the graph.

You're making the assumption that the GPSDO is operating correctly.  The main reason for testing a device is to verify that it is operating correctly.

The assumption that probably should have been stated is that you've got lock, good satellite visibility and signal strength, completed survey, etc.  You can't measure ADev or Std. Dev.  until all the basics are covered.  These steps alone will weed out most faults.

Texas's argument is that even when the basics are covered, the 18 hour test is needed to test for glitches and since you're doing that, you might as well measure ADev while you're at it.  I don't disagree with that, but an 18 hour test is just as arbitrary as a 1K (~17 min.) test.  I'd rather do a few 1K runs on different days and times than one 20K test.  I'm sure that you could find a unit that might pass a 1K sec. test but fail an 18 hour test.  I'm also sure that you could find a unit that would pass an 18 hour test and ultimately fail.

Murphy rules!

[tomato]

The assumption that probably should have been stated is that you've got lock, good satellite visibility and signal strength, completed survey, etc.  You can't measure ADev or Std. Dev.  until all the basics are covered.  These steps alone will weed out most faults.

Most faults doesn't mean all faults.  That's why you do the tests.  (Especially since people on this forum seem to favor used equipment of unknown condition bought at bargain basement prices.)

... an 18 hour test is just as arbitrary as a 1K (~17 min.) test.  I'd rather do a few 1K runs on different days and times than one 20K test. 

It's not an either-or choice.  It's just as easy to make a few 18 hr. runs on different days, and cover all your bases. 

[edpalmer42]

The assumption that probably should have been stated is that you've got lock, good satellite visibility and signal strength, completed survey, etc.  You can't measure ADev or Std. Dev.  until all the basics are covered.  These steps alone will weed out most faults.

Most faults doesn't mean all faults.  That's why you do the tests.  (Especially since people on this forum seem to favor used equipment of unknown condition bought at bargain basement prices.)

... an 18 hour test is just as arbitrary as a 1K (~17 min.) test.  I'd rather do a few 1K runs on different days and times than one 20K test. 

It's not an either-or choice.  It's just as easy to make a few 18 hr. runs on different days, and cover all your bases.

Now you're just trolling.  You know as well as I do that this is a statistical rabbit hole that has no end.  'All faults'??  Good luck finding 'all faults'.  Every user has to decide what level of performance is appropriate for their situation.  But if someone is even thinking that they might need to do 'a few 18 hr. runs' to 'cover all your bases', they won't be buying their hardware on fleabay because their corporate lawyers won't let them.

It's also worth keeping in mind that regardless of how long you test, there will still be data glitches caused by any number of things that have nothing to do with the experiment.  In fact, it might be impossible to tell whether a glitch is due to internal or external factors.  That's why all of the ADev programs have an extensive suite of data editing tools and warn that you have to edit out the glitches before you can analyze the data.

This isn't a professional time-nut forum.  It's a bunch of hobbyists who think that GPSDOs and other time-nutty things might be interesting to play with.  Telling them that they have to spend many hours making a single set of measurements is neither helpful nor correct.  You might as well tell someone that's making a 555 led blinky that his test is a failure unless the frequency is correct to withing a milliHertz.

Ed

[texaspyro]

I originally posted this image in the wrong thread...

Here is an example of why you want to look at GPSDO data over extended periods of time.  A 1000 second run would miss that big honkin' burst of crapitude.

[edpalmer42]

I originally posted this image in the wrong thread...

I wondered what was going on there.... 

Here is an example of why you want to look at GPSDO data over extended periods of time.  A 1000 second run would miss that big honkin' burst of crapitude.

Actually, it looks like it was about 45 minutes long.  That kind of stretches the definition of 'burst'!  But now what?  Was the cause internal or external?  Weird microwave oven or someone playing with a GPS blocker?  Bad solder joint or logic error?  Was it a one time event or will it repeat?  How long are you going to continue the test?  What have you proven?

Sure, a 1000 sec. test would have missed that noise, but that's not what it's trying to measure.  The 1000 sec. test assumes that the unit is operating properly during the test and attempts to document the performance during that 1000 sec. interval.  Even an ADEV test makes the assumption that everything is stable during the test.  If all hell breaks out you either edit it out of the data or delete the data and try again.  If it happens too often, you start looking for faults.

I'm measuring jitter rather than ADEV because jitter is a lot easier for a newbie to measure and understand.  ADEV isn't nearly as intuitive.  As I said earlier, I'd really like to see some more measurements so we can decide if measuring jitter actually makes sense as a way to characterize GPSDO performance at a hobbyist level.  Based on my measurements, it looks promising.

Ed

[rbm]

Here is an example of why you want to look at GPSDO data over extended periods of time.  A 1000 second run would miss that big honkin' burst of crapitude.

Texaspyro, In real world terms, what is the significance of this "burst of crapitude"?  Is this something that only affects the operation of a machine depending on the short term stability of the GPSDO signal?  Or is this something that adds error to the long term accuracy of the GPSDO?  I'm not very familiar with the concepts to work this out for myself.

[texaspyro]

Was the cause internal or external?  Weird microwave oven or someone playing with a GPS blocker?  Bad solder joint or logic error?  Was it a one time event or will it repeat?  How long are you going to continue the test?  What have you proven?

No idea what caused it.   I had Heather monitoring the GPSDO and nothing seemed to be amiss (like satellite dropouts) at the time.  A lot of GPSDOs seem to occasionally have similar behaviors...  something upsets their tracking loop and they go a bit wonky for a while.

What does it prove?   Well, it proves that GPSDO can go a bit wonky.

Disturbances like that will degrade the overall ADEVs of the GPSDO.  A 1000 second run would probably not catch such things.   A longer run will.   24 hours or so seems to be a reasonable length to catch the majority of such things.

BTW, to get the best performance from most GPSDOs try to set the satellite elevation and signal level masks to values that minimizes the changes in the tracked satellite constellation.  Most GPSDOs seem to add noise to the outputs whenever the constellation changes.

[tomato]

Now you're just trolling.

I happen to agree with Texaspyro's opinion on this subject, and have given reasons why.  That hardly makes me a troll.

You know as well as I do that this is a statistical rabbit hole that has no end.  'All faults'??  Good luck finding 'all faults'.  Every user has to decide what level of performance is appropriate for their situation.

An 18 hr. Allan Variance run of a GPSDO still does not guarantee one will catch 100% of faults, but it helps.  The best thing is that it costs nothing to extend an Allan Variance from 1,000s to 20,000s; you start your computer program and let it run.  Your computer doesn't care how long the data acquisition run lasts.

But if someone is even thinking that they might need to do 'a few 18 hr. runs' to 'cover all your bases', they won't be buying their hardware on fleabay because their corporate lawyers won't let them.

Corporate lawyers?  You're going to have to translate this, because I have no idea what it means.

This isn't a professional time-nut forum.  It's a bunch of hobbyists who think that GPSDOs and other time-nutty things might be interesting to play with.  Telling them that they have to spend many hours making a single set of measurements is neither helpful nor correct.  You might as well tell someone that's making a 555 led blinky that his test is a failure unless the frequency is correct to withing a milliHertz.

Nobody is telling anyone what they have to do.  I simply agree with Texaspyro's opinion that one should measure the Allan Variance of a GPSDO out to longer (20,000s) times.  I think this is especially important for hobbyists since many are buying GPSDOs of unknown operating condition.  I disagree that one only has to measure the Allan Variance out to 1,000s because "at large tau values ... they all converge due to the impact of GPS disciplining" because that assumes a priori that the GPSDO is operating correctly.