A look at my Symmetricom GPSDO / 10MHz reference (OCXO + Furuno receiver)

[uncle_bob]

Hi

If *frequency* accuracy is of interest, then ADEV (especially at 1 second) is probably not a good measure on a GPSDO. At 1 second, you are looking at the free running performance of the OCXO. You get great numbers, but the accuracy might only be as good as the OCXO's. You need to wait long enough that the GPS section has had time to get the OCXO settled down in order for the accuracy of the GPS system to transfer in a meaningful way to the OCXO.

A simple experiment:

Yank the antenna on the GPSDO and look at the ADEV. The numbers inside 10 seconds look no different than the numbers with the antenna connected. The OCXO *could* be off anywhere.

Leave the antenna off and watch the ADEV plot of the OCXO. As the unit stabilizes (over days / weeks), the curve past 10 seconds flattens out. The longer it is on, (in a stable environment) the flatter it is likely to get. That may continue out past 1,000 seconds or it may start to turn up a bit. The flatter that curve is, the better the short term frequency of the OCXO will be compared to the long term.

Yes, it would be nice if the ADEV curve dropped off as 1/sqrt(tau). That's not what happens in an OCXO. The physics don't work that way.

Bob

[NivagSwerdna]

Not sure how I would get ADEV?

But I think that's exactly my question... At startup you effectively get the OCXO performance and then the OCXO becomes gradually disciplined by the GPS PPS signal.  So when am I confident it is at least accurate to 1mHz in 10MHz.  Which I presume is 1e-10?

Looking at the DIAG:LOOP numbers they don't seem to change much over several days. 

Code: [Select]

-------------------------------------------------------------------------------
4/10/2016 21:42:57
-------------------------------------------------------------------------------

   LOS      MEAS       NCO STATUS WEIGHT PBUC FBUC DBUC LBUC IBUC G  M TC
-------------------------------------------------------------------------------
 0:  1  0.000e+00  0.000e+00 0x07AC   0   700  700    0    0    2  2  4 29
 1:  1  0.000e+00  0.000e+00 0x068C   0   700  700    0    0    0  2 14 226
GPS: 0 -2.082e-09 -6.762e-09 0x0000   1  ---- 1000 1000 ---- ----  3  6 31
-------------------------------------------------------------------------------
freq cor  = -6.761820e-09
phase cor = 2.000000e-11
gps phase = 3.333333e-09
temp cor  = -2.301801e-11
Command Complete

-------------------------------------------------------------------------------
4/17/2016 20:03:32
-------------------------------------------------------------------------------

   LOS      MEAS       NCO STATUS WEIGHT PBUC FBUC DBUC LBUC IBUC G  M TC
-------------------------------------------------------------------------------
 0:  1  0.000e+00  0.000e+00 0x07AC   0   700  700    0    0    2  2  4 29
 1:  1  0.000e+00  0.000e+00 0x068C   0   700  700    0    0    0  2 14 226
GPS: 0 -7.138e-09 -6.458e-09 0x0000   1  ---- 1000 1000 ---- ----  3  6 31
-------------------------------------------------------------------------------
freq cor  = -6.458316e-09
phase cor = -2.000000e-11
gps phase = 1.033333e-08
temp cor  = -3.138324e-11
Command Complete


and now...

Code: [Select]

-------------------------------------------------------------------------------
4/20/2016 08:25:29
-------------------------------------------------------------------------------

   LOS      MEAS       NCO STATUS WEIGHT PBUC FBUC DBUC LBUC IBUC G  M TC
-------------------------------------------------------------------------------
 0:  1  0.000e+00  0.000e+00 0x07AC   0   700  700    0    0    2  2  4 29
 1:  1  0.000e+00  0.000e+00 0x068C   0   700  700    0    0    0  2 14 226
GPS: 0 -8.632e-09 -6.277e-09 0x0000   1  ---- 1000 1000 ---- ----  3  6 31
-------------------------------------------------------------------------------
freq cor  = -6.276511e-09
phase cor = -2.000000e-11
gps phase = 1.533333e-08
temp cor  = -1.841605e-11
Command Complete


[NivagSwerdna]

I find this page http://www.gb3hm.co.uk/gps/UCCM.htm quite interesting... seems to be the same unit.  Can anyone interpret those graphs a bit for me?

My unit is currently saying...

Code: [Select]

UCCM-P > SYSTem:STATus?
-------------------------------------------------------------------------------
090-03861-03   serial number W560935175   firmware ver 1.0.0.2-01     LINK mode
-------------------------------------------------------------------------------
Reference Status __________________________   Reference Outputs _______________
   Ref 8KHz 0: [LOS]
                                              TFOM     2             FFOM     0
                                              UCCM-P Status[ACTIVE]

>> GPS: [phase:  1.27e-08]
ACQUISITION ................................................ [ GPS 1PPS Valid ]
Tracking: 4 ____   Not Tracking: 5 ________   Time ____________________________
PRN  El  Az  C/N   PRN  El  Az                GPS      20:17:52     17 APR 2016
  7  12 329   39    10  16 153                GPS      Synchronized to UTC
  8  17 275   35    16  76 255                ANT DLY  +5.000E-08
 20  23  50   44    18  34 117                Position ________________________
 21  59  71   35    26  59 168                MODE     Hold
                    27  48 279
                                              LAT      N  51:25:23.472
                                              LON      W   0:17:26.922
                                              HGT                +2.70 m  (MSL)




ELEV MASK  5 deg
-------------------------------------------------------------------------------

Command Complete

Maybe ACQUISITION isn't a disciplined state?

[edpalmer42]

If *frequency* accuracy is of interest, then ADEV (especially at 1 second) is probably not a good measure on a GPSDO. At 1 second, you are looking at the free running performance of the OCXO. You get great numbers, but the accuracy might only be as good as the OCXO's. You need to wait long enough that the GPS section has had time to get the OCXO settled down in order for the accuracy of the GPS system to transfer in a meaningful way to the OCXO.

Hi Bob,

The ADEV graph is still useful to show where and how the OCXO and GPS performance curves merge.  Whether ADEV is necessary isn't clear to me since the main purpose of ADEV is to filter out OCXO parameters like aging that confuse normal mathematical variance calculations.  The GPSDO cancels those out.

The assumption here is that the GPSDO being measured has had enough operating time for the GPSDO to correct any OCXO frequency error.  i.e. the OCXO is locked to GPS.  The OP wanted to know if that's all that's necessary.  The answer is no, but the OP will have to decide whether the differences are significant.

At 1 second, ADEV shows you the noise aka jitter performance of the OCXO which is mostly independent of GPS.  The frequency situation may not be as well determined, but in any practical GPSDO, the OCXO isn't going to be bouncing off the walls.  The locked status prevents a frequency offset as averaged over the length of the discipling loop's time constant.  But OCXO frequency changes over periods less than the time constant would still show up as poor ADEV.  If there are no frequency changes, there can't be a frequency offset.

A simple experiment:

Yank the antenna on the GPSDO and look at the ADEV. The numbers inside 10 seconds look no different than the numbers with the antenna connected. The OCXO *could* be off anywhere.

Okay, using 10 seconds as an example of the loop time constant.  Can I revise your statement to say "If you freeze the disciplining loop, the numbers inside 10 seconds look no different than the numbers with the loop active."?  I'm sure you know that some smart GPSDOs can compensate for OCXO aging during holdover conditions.  I'm playing to the audience here.    Yes, the OCXO *could* be off anywhere, but if it was doing that, you would have been warned by the poor short-term ADEV while it was in the locked condition.  So, in practice, that won't occur.

Leave the antenna off and watch the ADEV plot of the OCXO. As the unit stabilizes (over days / weeks), the curve past 10 seconds flattens out. The longer it is on, (in a stable environment) the flatter it is likely to get. That may continue out past 1,000 seconds or it may start to turn up a bit. The flatter that curve is, the better the short term frequency of the OCXO will be compared to the long term.

Bob, I totally agree with you on this point.  This is standard OCXO behaviour.  They like to run continuously and if you let them they will continue to improve - typically beyond their factory specifications.  Sometimes, with top-grade OCXOs, they can approach Rb standards for long-term aging.  So, as your OCXO improves, you'd like to increase your time constant to make the best use of the OCXO's low noise performance while still using GPS to cancel out the aging.  This is why the OP might want to run his GPSDO continuously - so the OCXO reaches its maximum potential.  Whether that matters is up to him.

Yes, it would be nice if the ADEV curve dropped off as 1/sqrt(tau). That's not what happens in an OCXO. The physics don't work that way.

Bob, you totally lost me on that point.  Did I imply something about 1/sqrt(tau) ? 

Ed

[edpalmer42]

Not sure how I would get ADEV?

But I think that's exactly my question... At startup you effectively get the OCXO performance and then the OCXO becomes gradually disciplined by the GPS PPS signal.  So when am I confident it is at least accurate to 1mHz in 10MHz.  Which I presume is 1e-10?

Looking at the DIAG:LOOP numbers they don't seem to change much over several days. 

You get ADEV by sending the data from your counter to a computer and then giving it to an ADEV program.  A program like Timelab combines those and lets you use either RS-232 or GPIB to collect the data and generates the graph automatically - if your counter is supported.

If your goal is 1 mHz accuracy, it shouldn't take very long.  Unfortunately, we don't have specs for these units.  However, the HP/Symmetricom Z38xx units use the same terminology and commands (mostly).  The Z3801A specifies a 10 MHz error of < 1e-9 averaged over 1 day (See note below).  They say that the spec is reached when the FFOM is <= 2.  GPS has improved since that manual was written (2000) so the performance of even an old GPSDO will be significantly better than spec.  But without measuring it, I can't say how much better or when you'd reach your 1 mHz goal.  I suspect that if you wait for an FFOM of 0 you'd be fine.

If you've had your unit running for a week or two, the loop parameters might have settled almost as much as they're going to.  Some GPSDOs remember the OCXO's behaviour even over a power cycle, so you might not be able to see a lot of variation even if you cycled the power.

Ed

Note:  I just looked at the specs for an HP 58540A Time & Frequency Receiver.  It's in the same family as the Z38xx units.  The manual, also dated in 2000, states that the 10 MHz accuracy is <= 1e-11 averaged over one day.  I don't understand why there would be so much difference in specs for the two units.  I doubt that there's much real difference between the Z3801A and the 58540A.

[uncle_bob]

Not sure how I would get ADEV?

But I think that's exactly my question... At startup you effectively get the OCXO performance and then the OCXO becomes gradually disciplined by the GPS PPS signal.  So when am I confident it is at least accurate to 1mHz in 10MHz.  Which I presume is 1e-10?

Looking at the DIAG:LOOP numbers they don't seem to change much over several days. 

You get ADEV by sending the data from your counter to a computer and then giving it to an ADEV program.  A program like Timelab combines those and lets you use either RS-232 or GPIB to collect the data and generates the graph automatically - if your counter is supported.

If your goal is 1 mHz accuracy, it shouldn't take very long.  Unfortunately, we don't have specs for these units.  However, the HP/Symmetricom Z38xx units use the same terminology and commands (mostly).  The Z3801A specifies a 10 MHz error of < 1e-9 averaged over 1 day (See note below).  They say that the spec is reached when the FFOM is <= 2.  GPS has improved since that manual was written (2000) so the performance of even an old GPSDO will be significantly better than spec.  But without measuring it, I can't say how much better or when you'd reach your 1 mHz goal.  I suspect that if you wait for an FFOM of 0 you'd be fine.

If you've had your unit running for a week or two, the loop parameters might have settled almost as much as they're going to.  Some GPSDOs remember the OCXO's behaviour even over a power cycle, so you might not be able to see a lot of variation even if you cycled the power.

Ed

Note:  I just looked at the specs for an HP 58540A Time & Frequency Receiver.  It's in the same family as the Z38xx units.  The manual, also dated in 2000, states that the 10 MHz accuracy is <= 1e-11 averaged over one day.  I don't understand why there would be so much difference in specs for the two units.  I doubt that there's much real difference between the Z3801A and the 58540A.

Hi

Here's the issue:

The "averaged over a day" number simply indicates that the phase lines up to some degree over 24 hours. The "frequency" being used is really frequency as delivered by GPS time. For practical use, it is not the number you think it is. The OCXO could swing +/- 1 ppm and hit that spec.

The more important spec if you are interested in frequency is -- how close is it if I use a gate time of ____ ? Normal numbers are 1, 10, and 100 seconds. The gotcha is that there is no closed form solution to that measure. (Which is why we have ADEV). About all you can say is: If I want it to be correct 99% of the time, what is the limit? Consider that with thousands of samples in a day, with 1 or 10 second gates, having 1% out still is a lot of errors.

You don't see the 99% spec on the cell tower GPSDO's. It was not part of the master specification on the units. The frequency limit in the master spec was 1x10^-8 which is pretty silly in this context. Over the first day a typical unit will hit < 1 ppb on a 99% basis. That will evolve to about 0.1 ppb pretty quickly (day 2 or 3). It will progress into the 0.1 to 0.01 ppb range, but probably will not exceed 0.01 ppb unless you have a *really* good OCXO.

Bob

[edpalmer42]

Not sure how I would get ADEV?

But I think that's exactly my question... At startup you effectively get the OCXO performance and then the OCXO becomes gradually disciplined by the GPS PPS signal.  So when am I confident it is at least accurate to 1mHz in 10MHz.  Which I presume is 1e-10?

Looking at the DIAG:LOOP numbers they don't seem to change much over several days. 

You get ADEV by sending the data from your counter to a computer and then giving it to an ADEV program.  A program like Timelab combines those and lets you use either RS-232 or GPIB to collect the data and generates the graph automatically - if your counter is supported.

If your goal is 1 mHz accuracy, it shouldn't take very long.  Unfortunately, we don't have specs for these units.  However, the HP/Symmetricom Z38xx units use the same terminology and commands (mostly).  The Z3801A specifies a 10 MHz error of < 1e-9 averaged over 1 day (See note below).  They say that the spec is reached when the FFOM is <= 2.  GPS has improved since that manual was written (2000) so the performance of even an old GPSDO will be significantly better than spec.  But without measuring it, I can't say how much better or when you'd reach your 1 mHz goal.  I suspect that if you wait for an FFOM of 0 you'd be fine.

If you've had your unit running for a week or two, the loop parameters might have settled almost as much as they're going to.  Some GPSDOs remember the OCXO's behaviour even over a power cycle, so you might not be able to see a lot of variation even if you cycled the power.

Ed

Note:  I just looked at the specs for an HP 58540A Time & Frequency Receiver.  It's in the same family as the Z38xx units.  The manual, also dated in 2000, states that the 10 MHz accuracy is <= 1e-11 averaged over one day.  I don't understand why there would be so much difference in specs for the two units.  I doubt that there's much real difference between the Z3801A and the 58540A.

Hi

Here's the issue:

The "averaged over a day" number simply indicates that the phase lines up to some degree over 24 hours. The "frequency" being used is really frequency as delivered by GPS time. For practical use, it is not the number you think it is. The OCXO could swing +/- 1 ppm and hit that spec.

The more important spec if you are interested in frequency is -- how close is it if I use a gate time of ____ ? Normal numbers are 1, 10, and 100 seconds. The gotcha is that there is no closed form solution to that measure. (Which is why we have ADEV). About all you can say is: If I want it to be correct 99% of the time, what is the limit? Consider that with thousands of samples in a day, with 1 or 10 second gates, having 1% out still is a lot of errors.

You don't see the 99% spec on the cell tower GPSDO's. It was not part of the master specification on the units. The frequency limit in the master spec was 1x10^-8 which is pretty silly in this context. Over the first day a typical unit will hit < 1 ppb on a 99% basis. That will evolve to about 0.1 ppb pretty quickly (day 2 or 3). It will progress into the 0.1 to 0.01 ppb range, but probably will not exceed 0.01 ppb unless you have a *really* good OCXO.

Bob

Yes, the 'one day' figure by itself is open to a lot of interpretation.  I think the Z3801A manual is just a reissue of an older manual and so doesn't spec things very well.  The 58540A manual however, also specs the ADev as <= 5e-10 @ 1 sec. one hour after initial lock, so frequency swings of even 1 ppb are unlikely.  But you're correct that they don't give a 99% spec.

But that still doesn't help us with this Symmetricom unit.  I don't have one so I have no measurements to share.  Measuring a GPSDO can be a challenge.  Tau values of > ~1000 sec. aren't necessary because GPS sets the performance.  In any case, it's easy to make those measurements.  But measuring at short tau values at these levels requires some effort and equipment.

Ed

[TheSteve]

This is good stuff guys. I think there are quite a few of us that would like to understand the best way to compare GPSDO's to really determine how(if?) they are performing.

[uncle_bob]

Not sure how I would get ADEV?

But I think that's exactly my question... At startup you effectively get the OCXO performance and then the OCXO becomes gradually disciplined by the GPS PPS signal.  So when am I confident it is at least accurate to 1mHz in 10MHz.  Which I presume is 1e-10?

Looking at the DIAG:LOOP numbers they don't seem to change much over several days. 

You get ADEV by sending the data from your counter to a computer and then giving it to an ADEV program.  A program like Timelab combines those and lets you use either RS-232 or GPIB to collect the data and generates the graph automatically - if your counter is supported.

If your goal is 1 mHz accuracy, it shouldn't take very long.  Unfortunately, we don't have specs for these units.  However, the HP/Symmetricom Z38xx units use the same terminology and commands (mostly).  The Z3801A specifies a 10 MHz error of < 1e-9 averaged over 1 day (See note below).  They say that the spec is reached when the FFOM is <= 2.  GPS has improved since that manual was written (2000) so the performance of even an old GPSDO will be significantly better than spec.  But without measuring it, I can't say how much better or when you'd reach your 1 mHz goal.  I suspect that if you wait for an FFOM of 0 you'd be fine.

If you've had your unit running for a week or two, the loop parameters might have settled almost as much as they're going to.  Some GPSDOs remember the OCXO's behaviour even over a power cycle, so you might not be able to see a lot of variation even if you cycled the power.

Ed

Note:  I just looked at the specs for an HP 58540A Time & Frequency Receiver.  It's in the same family as the Z38xx units.  The manual, also dated in 2000, states that the 10 MHz accuracy is <= 1e-11 averaged over one day.  I don't understand why there would be so much difference in specs for the two units.  I doubt that there's much real difference between the Z3801A and the 58540A.

Hi

Here's the issue:

The "averaged over a day" number simply indicates that the phase lines up to some degree over 24 hours. The "frequency" being used is really frequency as delivered by GPS time. For practical use, it is not the number you think it is. The OCXO could swing +/- 1 ppm and hit that spec.

The more important spec if you are interested in frequency is -- how close is it if I use a gate time of ____ ? Normal numbers are 1, 10, and 100 seconds. The gotcha is that there is no closed form solution to that measure. (Which is why we have ADEV). About all you can say is: If I want it to be correct 99% of the time, what is the limit? Consider that with thousands of samples in a day, with 1 or 10 second gates, having 1% out still is a lot of errors.

You don't see the 99% spec on the cell tower GPSDO's. It was not part of the master specification on the units. The frequency limit in the master spec was 1x10^-8 which is pretty silly in this context. Over the first day a typical unit will hit < 1 ppb on a 99% basis. That will evolve to about 0.1 ppb pretty quickly (day 2 or 3). It will progress into the 0.1 to 0.01 ppb range, but probably will not exceed 0.01 ppb unless you have a *really* good OCXO.

Bob

Yes, the 'one day' figure by itself is open to a lot of interpretation.  I think the Z3801A manual is just a reissue of an older manual and so doesn't spec things very well.  The 58540A manual however, also specs the ADev as <= 5e-10 @ 1 sec. one hour after initial lock, so frequency swings of even 1 ppb are unlikely.  But you're correct that they don't give a 99% spec.

But that still doesn't help us with this Symmetricom unit.  I don't have one so I have no measurements to share.  Measuring a GPSDO can be a challenge.  Tau values of > ~1000 sec. aren't necessary because GPS sets the performance.  In any case, it's easy to make those measurements.  But measuring at short tau values at these levels requires some effort and equipment.

Ed

Hi

Unfortunately 1 second ADEV tells you only that the standard deviation of the frequency change second to second (after drift is removed) is some number. It tells you nothing about the average change at all. The system can be cycling 1 ppm up and down and still hit a 5 x 10^-10 ADEV. It will have to do it very slowly. That's exactly what GPSDO steering does (change things slowly). Is it likely (or even possible) that it swings a ppm? No, it's not and it doesn't do that. The point is simply that it *could* do that and a 1 second ADEV would not catch it.

If you are going to evaluate a 1 second 99% spec at the 1x10^-11 level, you pretty much must have a Hydrogen Maser to compare it to. Anything else (including a high performance tube in a 5071) isn't going to do the job. There are a few other bits and pieces you will need, but the maser rules out most of us common folk.

Another approach is to compare two GPSDO's of different designs against each other. That can lead to all sorts of tangles. Yet another approach is to take the gate time out to > 100 seconds and a 5071 *might* do the job with some hand waving involved. Needless to say, you don't see many GPSDO's with a 1 second 99% confidence point at 1x10^-11 guaranteed by 100% testing.

Bob

[edpalmer42]

Unfortunately 1 second ADEV tells you only that the standard deviation of the frequency change second to second (after drift is removed) is some number. It tells you nothing about the average change at all. The system can be cycling 1 ppm up and down and still hit a 5 x 10^-10 ADEV. It will have to do it very slowly. That's exactly what GPSDO steering does (change things slowly). Is it likely (or even possible) that it swings a ppm? No, it's not and it doesn't do that. The point is simply that it *could* do that and a 1 second ADEV would not catch it.

If you are going to evaluate a 1 second 99% spec at the 1x10^-11 level, you pretty much must have a Hydrogen Maser to compare it to. Anything else (including a high performance tube in a 5071) isn't going to do the job. There are a few other bits and pieces you will need, but the maser rules out most of us common folk.

Another approach is to compare two GPSDO's of different designs against each other. That can lead to all sorts of tangles. Yet another approach is to take the gate time out to > 100 seconds and a 5071 *might* do the job with some hand waving involved. Needless to say, you don't see many GPSDO's with a 1 second 99% confidence point at 1x10^-11 guaranteed by 100% testing.

Bob

Luckily, we're just hobbyists so we don't have to worry about 99% of anything.    For that matter, we don't even have to produce definitive results, although they should be repeatable.  As you say, we don't have Hydrogen Masers.  I happen to have a Cs standard, but it's not a magic bullet either.  John Ackerman did some GPSDO comparisons that highlighted both good results and traps for young players.  http://www.febo.com/pages/gpsdo_comparison/ .  Beyond that, we can muddle around and see what happens.  I've wondered what the results would be if I used the 'three-cornered hat' technique to compare three GPSDOs.  Maybe I or someone else here will find that out.

It's all good!

Ed

[NivagSwerdna]

Simplistically, So if I wait 10^4 seconds (presumably after hold reached) then I should have got a good chunk of quality out of my box?  10,000s is about 3 hrs?

[edpalmer42]

Simplistically, So if I wait 10^4 seconds (presumably after hold reached) then I should have got a good chunk of quality out of my box?  10,000s is about 3 hrs?

Yes, but if it takes that long to reach an acceptable level (i.e. FFOM = 0), personally, I'd just let the thing run 24/7.  A 3 hour warmup time would drive me crazy.  I have a really old Rb standard that takes almost an hour to warm up.  By the time it's ready to go I'm practically gnawing on the furniture in frustration. 

Ed

[uncle_bob]

Simplistically, So if I wait 10^4 seconds (presumably after hold reached) then I should have got a good chunk of quality out of my box?  10,000s is about 3 hrs?

Hi

GPSDO's go through a series of filter adjustments as the OCXO warms up and the unit stabilizes. You only get through the first part of the process 24 hours. Running for at least three days is about the minimum for reasonable accuracy.

Bob

[edpalmer42]

Simplistically, So if I wait 10^4 seconds (presumably after hold reached) then I should have got a good chunk of quality out of my box?  10,000s is about 3 hrs?

Hi

GPSDO's go through a series of filter adjustments as the OCXO warms up and the unit stabilizes. You only get through the first part of the process 24 hours. Running for at least three days is about the minimum for reasonable accuracy.

Bob

Hi Bob,

Can you give an estimate of what 'reasonable accuracy' would be?  We're flying blind with this Symmetricom GPSDO.  We have no docs, no specs, and no clue what the design requirements were.  But if it was telco equipment, the start-up phase might have been very loosely defined since it was likely intended for 24/7 operation.  It appears to be from WiMax equipment.  And remember that NivagSwerdna is only looking for 1 mHz accuracy for the 10 MHz output.

Ed

[TheSteve]

Probably just as hard to find the specs of the equivalent Trimble units.

[uncle_bob]

Simplistically, So if I wait 10^4 seconds (presumably after hold reached) then I should have got a good chunk of quality out of my box?  10,000s is about 3 hrs?

Hi

GPSDO's go through a series of filter adjustments as the OCXO warms up and the unit stabilizes. You only get through the first part of the process 24 hours. Running for at least three days is about the minimum for reasonable accuracy.

Bob

Hi Bob,

Can you give an estimate of what 'reasonable accuracy' would be?  We're flying blind with this Symmetricom GPSDO.  We have no docs, no specs, and no clue what the design requirements were.  But if it was telco equipment, the start-up phase might have been very loosely defined since it was likely intended for 24/7 operation.  It appears to be from WiMax equipment.  And remember that NivagSwerdna is only looking for 1 mHz accuracy for the 10 MHz output.

Ed

Hi

Is the unit capable of doing position hold to a prior survey?

If so, is the unit previously surveyed in and that position saved?

Right there you have a process that will take longer than the 3 hours he's talking about. While it's in survey mode ... no bets at all about accuracy.

How good is the antenna location?

How many sat's are locked on at all times?

If the answer is "always > 6, usually more than 8" that's a long way from "usually 6" that some antenna locations give you. A good antenna with 360 degree view to the horizon will do a survey pretty quick (hours). An antenna that keeps going down to and below 4 sats could take days.

How long has the unit been power off?

An OCXO may move a few ppb in the first hour and not settle down for quite a while if it's been off power for weeks / months / years. It will do a lot better if it has been on power for months and was only turned off for a couple of minutes. The control loop has a tough time just tracking the nutty GPS wander. Throw in the OCXO moving like crazy and it's going to do worse.

Which version of the HP / Symmetricom firmware is in these units?

I have seen some of them that are moving in the 10 mHz for the first 24 hours after turn on. It's not a one sided sort of "always gets better" kind of thing. It's swinging back and forth in frequency to keep the time aligned. On those units, you can still see very distinct changes in the disciplining over the first couple of days.

How picky are you?

There have been posts made of people doing plots of these sort of units showing > 1 mHz (on 10 MHz) swings after over a month on power. They many not do it often, but it can "hang out" in that region for an hour or more.

Lots of variables. The only safe answer is to keep it on power for multiple days.

Bob

[NivagSwerdna]

Is the unit capable of doing position hold to a prior survey?

Yes.  Indeed I programmed my display to save the result of the survey to EEPROM and I can load this after a cold restart.  I haven't timed it exactly but after a cold restart, and after OCXO warmup the unit will go to FFOM=0 within approx. 10 minutes of having its prior position re-loaded.

I'm a bit sceptical about short term performance,... in particular I am not sure what parameters the unit stores between cold starts, if it remembers a few things I can see it could restore and get itself rapidly quite close but it doesn't seem to remember much (e.g. it forgets initial position)

If the PPS of the GPS is accurate to 100nS (which I seem to remember from a Fugano spec) then that is only 10e-7 so I can see it rapidly getting to 10e-7 once a GPS lock has been achieved (although obviously it would need to converge on that solution somehow).  If I am after 10e-9 then that definitely seems to imply some time... 10 mins is 600 seconds, maybe that is enough.

I thought I knew my racal-dana 1998 frequency counter was out... It was showing 9.999793 but then I left it and the GPSDO running and came back two days later and it reads..  9.999993

I guess I'll never be sure of the frequency of anything ever again!

[uncle_bob]

Hi

The simple answer it to let your GPSDO run continuously. That is the environment they were intended to be used in. The firmware in the unit was all targeted at an "always on" situation. The only specification that really matters on them is "holdover". That is: how long will time pulse output stay within some limit (say 1.5 us for 4 hours) after the antenna is unplugged. Even that has a bunch of "after three days on power" sort of qualifications on it.

Bottom line - it's just a happy coincidence that these gizmos put out accurate frequency. We all use them that way, but they were not designed with that as a primary goal. The specs don't talk about it and the firmware is not optimized to do it.

Bob

[TheSteve]

Bottom line - it's just a happy coincidence that these gizmos put out accurate frequency. We all use them that way, but they were not designed with that as a primary goal. The specs don't talk about it and the firmware is not optimized to do it.

Are you talking about this Symmetricom unit specifically or all GPSDO's in general?

[NivagSwerdna]

The simple answer it to let your GPSDO run continuously.

Looks like it's getting a Raspberry Pi and doubling as a NTP server then!

[uncle_bob]

Bottom line - it's just a happy coincidence that these gizmos put out accurate frequency. We all use them that way, but they were not designed with that as a primary goal. The specs don't talk about it and the firmware is not optimized to do it.

Are you talking about this Symmetricom unit specifically or all GPSDO's in general?

Hi

All of the GPSDO's that:

1) Are OCXO based
2) Were designed for wireless com use
3) Came from Trimble and HP / Symmetricom
4) Were sold in high enough volumes that you can *find* multiple examples

Have a *lot* in common with each other. That's even more true when you narrow it down to gear made by Trimble or just to gear made by HP / Symmetricom. You can look at the lockup profile on an early Trimble and see the same thing going on with their new stuff. The same is true of the way the early HP's chug through settings in the same way the later Symmetricom designs do. So yes, this is similarity and taking lotos of data on a few hundred GPSDO designs over the years.

Are they identical? Certainly not. They targeted different holdover specifications and have different OCXO's as a result. The early ones used a lot more components to get the job done than the later ones. Some have more extensive command sets that others. The early ones came out in big sturdy enclosures. The later stuff is barely more than a pc board with parts on it. Some specific examples if this model have an OCXO from supplier A and others have an OCXO from supplier B. One or the other likely does a little better.

All of these units require a good antenna location. They all need to (somehow) be put into position hold mode. They all work best when on power continuously. The cosmetic differences are bigger than the model to model variations when using them as a frequency standard in a normal "alway on" lab environment. They are lower cost, lower power, higher reliability, and longer life than any of the Rb's or Cs standards you commonly come across on eBay. Yes you have to shop a bit for the lower cost part. Yes you *can* find GPS disciplined Rb's on rare occasions.

Yes, that's a long and rambling answer to a fairly simple question.

Bob

[TheSteve]

OK, I just got the sense you were implying(well you said it) that it is only a happy coincidence they generate an accurate frequency. Considering several HP models say "Frequency reference" on them I'd like to think they were specifically designed with accurate frequency in mind. One of mine says "reference clocks" and doesn't supply any 1 PPS etc so the only thing it is doing is providing an accurate frequency output.

[uncle_bob]

OK, I just got the sense you were implying(well you said it) that it is only a happy coincidence they generate an accurate frequency. Considering several HP models say "Frequency reference" on them I'd like to think they were specifically designed with accurate frequency in mind. One of mine says "reference clocks" and doesn't supply any 1 PPS etc so the only thing it is doing is providing an accurate frequency output.

Hi

It is abundantly unclear if the HP "frequency reference" units actually are supposed to be different than the cell tower / wireless designs. A *lot* of what you see sold as the lab units are cell tower surplus that has been re-packaged in China in the pretty desktop enclosure. When you get one, a very real question to ask is "what i*is* this beast?". At least in my highly informal survey, 9 times out of 10, it's a cell tower unit with cell tower firmware in it.

That said, yes, you can target the firmware at frequency instead of time. You also can target something other than holdover performance. I have never seen a lab environment that actually used / needed / wanted / cared about holdover. Doing all the heavy lifting to get that done takes a bit of gear and quite a bit of time. For the investment, it's better to do the work on an Rb than on an OCXO. You will invest a lot and you might as well have the better result. The cost of the Rb is trivial compared to the other gear and your time. The downside of the Rb is that with a > 48 hour loop, they take a *long* time to test and tweak in. Getting to 10 or 20 time constants just for settling is a while. You also will spend some time on pressure and temperature compensation.

Bob

[kilobyte]

Hi,

I got my UCCM today and saw that U22 ist missing.

Can someone look what's the Part Marking on U22?
I think it could be some sort of a singlegate and/or logic.

If i have cleaned up the oily PCB i will take some highres photos because i didn't find any yet.

Regards
Kai

[Macbeth]

Can someone look what's the Part Marking on U22?

It's a 5 pin SOT marked Z32H. I reckon NC7SZ32