Comparing two HP Z3805A GPSDO 10 MHz frequencies for lab use

[HighVoltage]

Over a longer period of time, I wanted a 10 MHz GPSDO as a standard for the lab. But I was not sure, which one to get and our friend and eevblog member "ZUCCA" offered to send me his HP Z3805A in the mail for testing.

I started with a cheap GPS Antenna but could not get Z3805A to go in to "GPS-LOCK"
Then I tried a much more expensive GPS Timing Antenna and also could not get the Z3805A in to "GPS-LOCK"
Suspecting that something was wrong with the unit from ZUCCA, I ended up, ordering the very same Z3805A unit from the same seller in China. It arrived in a few days and I tried to get this one to go in to "GPS-LOCK" but it did not lock as well.

Then I ordered a new antenna. This time a Trimble Miniature 5V GPS Antenna P/N: 56237-00
This one seems to be much higher quality, including the lab quality SMA connector.
With this Antenna, I got the unit from Zucca in to "GPS-LOCK"
Then I ordered the same Antenna again for my unit and now both are in GPS Lock for a few days already.
But you have to have patients, it takes anywhere from 1 to 4 hours, before the unit will show the GPS-Lock light ON.

Now the question was, how stable is the signal from these two Z3805A units.
I hooked up my two counters and a scope

Placing the counters in RATIO mode "Frequency 1 / Frequency 2" should give me a good reliable reading, even if my counters are not perfectly calibrated.
And these are the results:
Counter 53131A shows: 1.000,000,00 and does not move at all
Counter 53230A shows: 1.000,000,000,001,9 but the last three digits are moving slightly

On the scope, I am measuring the Phase shift between the two waveforms
From yesterday to today, the phase shift was about 30 degree.
Which means, if this phase shift would stay constant, it needs 12 days for a full 360 degree cycle.
So, I think it is safe to assume, that both units are in perfect GPS lock and we can rely on the 10 MHz output.

So, before I am sending this one unit back to Zucca, does anyone here in the forum has some suggestions of what to measure and how to compare the two units for stability? I don't think it happens too often, that we see two GPSDO's next to each other.

[davorin]

I don't even have one...yet....my Jupiter-T modules are still waiting to be used...external GPS antenna is already in place (o;

So did you feed each counter with one of the GPSDO outputs? Or both counters from same reference clock?

Not sure...but maybe a HP 5372A/5371A will show more?

[HighVoltage]

Hello Davorin,
No, for this test I did not feed the output of the GPSDO to the counters as a reference.
This was on purpose, since I wanted to compare the two GPSDO's independently.
Both counters have the ultra high stability oven controlled references built in and should be spot on, when it comes to ratio measurements, since both channels are based on the same internal reference clock.

But it might be interesting to do the test you proposed, since I think I can rely on the outputs of the GPSDO's

[davorin]

How much did you pay for the GPSDO?

As mentioned I'm planning an own one based on Jupiter-T GPS modules with an OCXO and PLL as it has 1kHz and 1Hz Outputs only...
but as side effect I could read out and display exact date and time in a front panel (o;

But not sure if the result will be as accurate as the Z3850A.


How accurate is a GPS clock anyways....Long, Long time ago I saw a documentation where they compared two atomic clocks...synchronised them both...took one in a plane and flew around. When they landed the atomic clock in the plane was running behind the atomic clock at ground.

[Zucca]

How much did you pay for the GPSDO?

https://www.eevblog.com/forum/buysellwanted/ebay-hp-z3805a-gps-frequencytime-receiver-10-mhz-1pps/

How accurate is a GPS clock anyways....

Google knows always more than me...
http://gpsinformation.net/main/gpstime.htm

[HighVoltage]

I think it will be near impossible for me to measure the accuracy of the GPS timing. This probably requires an atomic clock.

This is why it is so great to have two GPSDO's at the same time.
Both have their own antenna and run independent of each other.
And seeing only a 30 degree phase shift in one day tells me that they are both locked perfectly.
Also, the Z3805A has two oven controlled oscillators internally.

May be someone here in the forum can suggest another way of measuring the differences between the two units with toolks that I have at hand.

Interestingly, one of the units is about 1 degree C hotter on the outside than the other one.
May be I should open both and have a look inside.

The cost was about $ 275 delivered to Germany by DHL Express.

[edpalmer42]

A better way to compare two frequency references is to connect one unit to Channel A, the second unit to Channel B, and then measure the time interval from A to B.  Collect the data on a computer via GPIB or RS-232 and analyze the data offline.  This removes the averaging effect of the gate time in the frequency measurement and allows the counter to make best use of its resolution.

The results of this should show some wandering back and forth of the phase due to differences between units, antennas and antenna locations, etc.  Because you're comparing two of the same model, the differences would be much smaller than if you compared two different models.

Ed

[SoundTech-LG]

I use my Stanford Research FS700 modified for External Reference 10 MHz input (since it normally uses an internal OCXO). Using one GPSDO for the FS700 reference, I can then compare my other GPSDO for Phase, and Frequency over time, using the FS700 Phase, and Frequency Comparator Display. Good down to approximately +/- 7 E-15. It also possible to use, and compare divided down frequencies as low as .1 Hz These have been available from time to time cheap on ebay, but recently outrageously priced as high as $1000. Not sure why. 

[HighVoltage]

Thanks Ed, I will try such a test.
The 53230A counter has USB logging, so this will be quick to setup.

Right now I did a test were one of the GPSDO's is the reference input to the counter and then I am measuring the other GPSDO on the input channel of that counter.

Kind of amazing results:

53131A: 10.000,000,000,0 (the last digit changes)
53230A: 10.000,000,000,010,7 (the last 3 digits change)

[HighVoltage]

I use my Stanford Research FS700 modified for External Reference 10 MHz input (since it normally uses an internal OCXO). Using one GPSDO for the FS700 reference, I can then compare my other GPSDO for Phase, and Frequency over time, using the FS700 Phase, and Frequency Comparator Display. Good down to approximately +/- 7 E-15. It also possible to use, and compare divided down frequencies as low as .1 Hz These have been available from time to time cheap on ebay, but recently outrageously priced as high as $1000. Not sure why. 

For me, this precision timing is only a hobby right now and I have no real world applications except for enjoyment.
What are you using a precision of 7E-15 for?

That FS7000 looks really interesting.
But that price really is a little steep.

[awallin]

So, before I am sending this one unit back to Zucca, does anyone here in the forum has some suggestions of what to measure and how to compare the two units for stability? I don't think it happens too often, that we see two GPSDO's next to each other.

What edpalmer42 said: put the counter (e.g. 53230A) in time-interval mode, have one GPSDO produce a "start" pulse, the other "stop".
For this you might want to divide down to 1Hz (aka 1-PPS) or slightly faster pulses, using e.g. PICDIV
http://www.leapsecond.com/pic/picdiv.htm

The good counters (53230A or SR620) are limited to about ADEV(tau)= 2e-11/tau(s)  so you can't measure below that.
If the GPSDO has a good OCXO inside it will start out below the noise level of the counter. So below tau=10...100s you will see counter white phase noise, after that the OCXO, and depending on the time-constant of the GPSDO-loop the ADEV should turn down at high tau.

[edpalmer42]

What edpalmer42 said: put the counter (e.g. 53230A) in time-interval mode, have one GPSDO produce a "start" pulse, the other "stop".
For this you might want to divide down to 1Hz (aka 1-PPS) or slightly faster pulses, using e.g. PICDIV
http://www.leapsecond.com/pic/picdiv.htm

Dividing the signal down is almost mandatory if you're comparing two OCXOs.  Since they're not running at the identical frequency, there will be phase wraps which must be removed before the data can be analyzed.  In this case, dividing the signal is optional because the two GPSDOs are running at the identical frequency.  The GPSDOs probably have a built-in 1 PPS output so either output could be used.

Ed

[jpb]

I'm fascinated by GPSDOs and am trying to design and build my own, at least I was until moving a couple of months ago and (temporarily) losing my electronics lab. Time nuts and other source are a great source of discussion on the measurements of oscillators and Allan deviation - e.g. see
 http://www.leapsecond.com/pages/gpsdo/

The thing with GPSDOs is that they are all a compromise between the short term accuracy of the OCXO and the long term accuracy (but short term jitter) of the GPS so it is an interesting control problem and people can get very carried away with seeking perfection.

Even doing the measurements is difficult - I've done some playing around with mixers which is the easiest way to get down to the very low noise floor needed.

[German_EE]

To compare two signals at the same frequency you just need an oscilloscope.

1) Connect the first oscillator to Channel A

2) Connect the second oscillator to Channel B

3) Trigger off Channel A. Waveform B will drift either left or right very slowly so you just need to see how long it takes for one cycle to drift by to measure in parts per trillion.

1 10 MHz cycle in:
1 second 100 ppb
10 seconds 10 ppb
1 min 40 seconds 1 ppb
16 minutes 40 seconds 100 ppt
2 hours 46 minutes 40 seconds 10 ppt
1 day 3 hours 46 minutes 40 seconds 1 part per trillion

[TheSteve]

With two GPSDO's I don't think you will ever see a full cycle drift - at least I haven't with mine. They do slowly move away from each other and then back again but the biggest difference I have seen is 30 nanoseconds.

[HighVoltage]

To compare two signals at the same frequency you just need an oscilloscope.

That is what I did, see above in the original post.
The drift was about 30 degree phase shift in 1 day

So, a very simple calculation come down to:

30 degree phase shift in one day
This means 12 days for a complete 360 degree cycle
12 days = 12 x 24 x 60 x 60 = 1.036,800 seconds
1/ 1.036,800 = 9,645E-7 Hz drift

As long as the drift is linear and stays constant over time,this simple calculation is probably correct.
But I do not know, if it might drift backwards after a couple of days or so.
I am taking measurements over night from the difference between two pulses.

Thank you all for some great suggestions and links.

[TheSteve]

Would your drift number really be drift though, I would have thought it was something closer to the maximum error/deviation. If the GPSDO is constantly being synchronized to the 1 PPS should it not eventually average out to be exactly 10 MHz? This assumes the GPSDO firmware is perfect or close to it, perhaps none are.

[acbern]

To compare two signals at the same frequency you just need an oscilloscope.

1/ 1.036,800 = 9,645E-7 Hz drift

well, it is actually much more precise than 9.6E-7, because you need to relate this to 10MHz. Would be bad GPSDOs with only that accuracy betwen each other.

[VK5RC]

The 'accuracy " you are looking for can be many different types,  if you are looking for very short term accuracy,  ie one cycle,  I think you  can be better off switching off the gps disciplining as it can induce phase errors as it' corrects"  the crystal,   if you measure over hours it will be better to leave the disciplining on.
The oscilloscope method works well over longer time frames,  but I doubt it would show 'jitter " ie cycle to cycle variation,  which may be critical especially if the reference frequency is to be multiplied many times perhaps for a microwave application.

[HighVoltage]

1 10 MHz cycle in:
1 second 100 ppb
10 seconds 10 ppb
1 min 40 seconds 1 ppb
16 minutes 40 seconds 100 ppt
2 hours 46 minutes 40 seconds 10 ppt
1 day 3 hours 46 minutes 40 seconds 1 part per trillion

Based on your simple calculation, I have in theory the following with a phase shift of 30 degree per day.
1 complete (360 degree) cycle in 12 days
12 days = 1036800 seconds

So, in other words it is about 0.0001 ppm or 0.1 ppt

This accuracy between two GPSDOs is kind of amazing to me!
I will see, if I can get a curve of the time shift variations over a few days.
 

[DimitriP]

On the scope, I am measuring the Phase shift between the two waveforms
From yesterday to today, the phase shift was about 30 degree.
Which means, if this phase shift would stay constant, it needs 12 days for a full 360 degree cycle.

If you run your test in the first 24 hours of having powered up, let them "cook" for a few days and re check the phase shift.

[HighVoltage]

If you run your test in the first 24 hours of having powered up, let them "cook" for a few days and re check the phase shift.

Yes, good idea, may be it will even stabilize more.
Although 0.0001 ppm is already unbelievable stable relationship between the two GPSDOs

I found a great tool in the Agilent 53230A, that can measure the phase shift directly
It shows 211.9 degrees right now.
Down to 15 digits !!!
Now, who needs that 15 digit precision for a phase shift?



 

[Zucca]

Down to 15 digits !!!
Now, who needs that precision for a phase shift?

It is not incredibly useful, but incredibly cool.

[SoundTech-LG]

Indeed, the coolness... too many clocks, not enough time.

Time Interval Measurement.  More good stuff in here: http://ko4bb.com/dokuwiki/doku.php?id=start

[DimitriP]

I found a great tool in the Agilent 53230A, that can measure the phase shift directly
It shows 211.9 degrees right now.

The 53230A has a statistics screen you can be having all kinds of comparative fun with ...