EEVblog #457 - Oscillator Calibration Followup

[EEVblog]

[nitro2k01]

Too good for Australia!

You could check the integrity of the counter by feeding the same rubidium signal into both the ref in and the counter input and see if it's bang on down to the last digit. You would expect it to be of course. You could also try this with different length coaxes to introduce a slight delay and see if that makes the least significant digit wobble.

[EEVblog]

You could check the integrity of the counter by feeding the same rubidium signal into both the ref in and the counter input and see if it's bang on down to the last digit.

You don't need a rubidium to do that, any oscillator will work.
And yes, of course it'll be spot on, even the cheapies.

[c4757p]

You could check the integrity of the counter by feeding the same rubidium signal into both the ref in and the counter input and see if it's bang on down to the last digit. You would expect it to be of course. You could also try this with different length coaxes to introduce a slight delay and see if that makes the least significant digit wobble.

I wouldn't consider a function counter that can't even count its own reference "lower quality" or "out of spec", I'd consider it defective crap. How hard is counting??

[nitro2k01]

You don't need a rubidium to do that, any oscillator will work.

I'm not so sure. I would assume the frequency counter has its own high frequency oscillator which is supposed to be locked to the reference, but might lag ever so slightly to changes in the reference signal. If this is the case, it could be show the wrong value in least significant digit for an update here and there. The oscillator would probably have to be pretty crap for that to happen, but might be something to explore. You could even generate a 10 MHz signal which is deliberately modulated maybe by a modulator of 1 Hz and feed into both sides of the unit see if it ever shows anything but 10.000000000 MHz.

[EEVblog]

You don't need a rubidium to do that, any oscillator will work.

I'm not so sure. I would assume the frequency counter has its own high frequency oscillator which is supposed to be locked to the reference

No, that's not how counters work. The external reference becomes the direct clock for the counting logic. if you feed that same signal to the input, at worst you'll get +/1 count due to how the clocking is designed.

[nitro2k01]

Ok.

[mikes]

No, that's not how (most) counters work. The external reference becomes the direct clock for the counting logic. if you feed that same signal to the input, at worst you'll get +/1 count due to how the clocking is designed.

"Counters" (at least good ones) don't work that way, either. If they simply counted cycles per second, they'd always take a full second for each reading to get 1 Hz resolution. Some may measure the period, and report the reciprocal. They'll more often have some sort of interpolator, which allows them to measure fractions of cycles, with and use shorter gate times.

Phase noise and/or input comparator jitter and/or choice of gate time make it unlikely that you'll see an exact measure, even when self measuring the timebase.

[kxenos]

No, that's not how (most) counters work. The external reference becomes the direct clock for the counting logic. if you feed that same signal to the input, at worst you'll get +/1 count due to how the clocking is designed.

"Counters" (at least good ones) don't work that way, either. If they simply counted cycles per second, they'd always take a full second for each reading to get 1 Hz resolution. Some may measure the period, and report the reciprocal. They'll more often have some sort of interpolator, which allows them to measure fractions of cycles, with and use shorter gate times.

Phase noise and/or input comparator jitter and/or choice of gate time make it unlikely that you'll see an exact measure, even when self measuring the timebase.

They don't count per second. They count per ref. clock cycle. Phase noise and jitter are random effects and wouldn't change the result. Delay also wouldn't count. Imagine hitting your hand on your table and counting the taps you hear.

[Ferroto]

You should do a comparison to the rubidium standard you got off ebay.

[mikes]

They don't count per second. They count per ref. clock cycle. Phase noise and jitter are random effects and wouldn't change the result.

No. The counter at hand has a 10 MHz ref clock. If it counted "per ref. clock cycle," any frequency less than 10 MHz would count as 0 or 1. Only the very old, or el cheapos, rely on simply counting during the gate time. Most modern "counters" might more properly called frequency meters.

While calibrating the time base in the video, the counter appeared to be using a 100 ms gate time, yet had a resolution (not accuracy) of .1 Hz. With a 10 MHz input, that's only 7 digits if simply counting (1,000,000 cycles), yet it produces a resolution of 9 digits. That's because it's doing interpolation of the signal, not simply counting. Interpolation commonly involves counts within the gate time, adjusted with a measurement of how long it takes for the next "count" to occur. And that's not measured using ref clock cycles - it's often a ramp interpolator. Interpolation is also done at the start of the gate, since the signal cannot be assumed to be synchronous with the ref clock.

Interpolation depends on accurately locating a point on the waveform - something which is affected by both phase noise (movement of that point in the time domain) and input comparator jitter (measuring a sine wave will be less accurate than measuring a square wave, because of the need to accurately locate a point on the relatively shallow slope).

[kxenos]

The technique of how the measurements are being done is irrelevant since it probably is linear to 10 or 100mHz. The timebase for all measurements is 10MHz. Remember, we are talking about Nitro's question that was if there will be any difference if you feed 10MHz in the input with a 10MHz source.

[EEVblog]

I just shot a quick video demoing this.
One in the queue already, so it'll be the video after that.

[grumpydoc]

There's a good description of frequency meter measurement techniques here

[EEVblog]

Dave, can you stop saying "I shot a quick video" please. Most of us have sussed that it is complete BS. You came this close (holds fingers up like Maxwell Smart, "missed me by this much") and then couldn't resist adding a bit more.
I know you are saying  you shot a video quickly without your usual extensive preparation and research  but you say it like you're claiming it is a "short" video without the waffle.

No, when I say at the start it'll be a "quick" video, it means I have no intention of looking at the thing/subject in-depth, it doesn't necessarily mean it will be short, concise, or lack waffle.

[free_electron]

Hook up the scopes 10mhz input to the rubidium, then hook up the counters 10mhz to the rubidium as well. Make sure to use equal length coaxes.
Take two coaxes with different lentgh now and hook up to both channels of the counter or two channels of the scope. You can now show accurate flight delay in coaxes. The counter can measure phase difference.

[jahonen]

Here is a HP appnote describing how to determine the stability of an oscillator compared to a reference oscillator using a phase difference method:

http://www.hpmemory.org/an/pdf/an_77-2.pdf

Apparently one could also misuse a VNA by feeding it with external reference and using it as a phase detector with a very long time span, as done by another Finnish guy:

http://www.kolumbus.fi/michael.fletcher/frequency_error_using_vna.htm

Regards,
Janne

[free_electron]

Dang it, i knew i posted a teardown of mine a while ago... Here it is

https://www.eevblog.com/forum/testgear/hpagilent-53181a-teardown/msg191212/#msg191212

[EEVblog]

Dang it, i knew i posted a teardown of mine a while ago... Here it is

https://www.eevblog.com/forum/testgear/hpagilent-53181a-teardown/msg191212/#msg191212

I think that's a 53131A and not a 53181A 

[grumpydoc]

I use the technique tha Dave demonstrated to check my FE-5680A against my GPSDO, although observing the frequency difference takes a little longer - the last time I checked the two against each other it took more than an hour for the phase difference to go from 0^o to 360^o. That's about 3 parts in 10¹¹.

[JackOfVA]

Some perhaps interesting plots attached.

First shows stability of a Racal 1992 frequency counter time base with the Option 4E oscillator - highest stability oven offered by Racal at the time the counter was manufactured - compared against an HP 5065A Rubidium standard. If trying to set a standard counter time base oscillator to zero is difficult, setting a high stability oscillator is difficult squared or cubed, as you have to wait hours between adjustments to see the effect of the tweak.

Second shows a comparison between an HP 5065A Rb standard and an HP 5017A Cesium standard. Difference in frequency is 6 x 10^-13. (Not the same Rb standard used to generate the first plot.)

Third plot shows the Rb standard used in plot 1 compared with an HP Z3816A GPS disciplined oscillator. Difference is 1.1 x 10^-12 - GPS oscillator shows some spikes where the receiver lost signal for a brief period and hence the results may not be as reliable as the other two plots.

The plots all show quantization due to the finite resolution of the counter.

Data taken via the 1992's GPIB port, with a Prologix GPIB-USB adapter and collection software I wrote. Plotted with Origin, a scientific/engineering data analysis and plotting program.

[free_electron]

I think that's a 53131A and not a 53181A 

fainbrartbainfrart brainfart

[mikes]

Here's a Trimble Thunderbolt (GPSDO) vs. an Efratom FRS Rb. Measured using an HP 5370A Time Interval Counter. Ambient temperature is also plotted, you can see its effect. It changed about 250 ns in 2 days, a difference of about 1.5e-12 (about 35 millionths of a second per year).

[ddavidebor]

Mmh, i never think that temperature is so important for a rubidium clock

[EV]

Mmh, i never think that temperature is so important for a rubidium clock

Probably more important for Efratom FRS Rb!?

[mikes]

Mmh, i never think that temperature is so important for a rubidium clock

Don't assume it's all in the Rb - both will shift with temperature change. The GSPDO will correct itself, but there's a time constant. Rbs are adjusted by changing a magnetic field (the C-field). So, any temperature sensitivity in the circuitry which controls that will affect the frequency.

The FRS does have provision for temperature correction, and I've got it set pretty close. Note that the changes are 2nd order, depending more on the rate of change than the actual temperature.

[rf-loop]

Mmh, i never think that temperature is so important for a rubidium clock

Probably more important for Efratom FRS Rb!?

Thunderbolt is sensitive for temp! (lot of) also it is not very frequency stable becouse in default parameters are optimized for time. So, it use quite big freq chances for run time error as small as possible.

Efratom FRS is sensitive for many things, also temperature.
Same for LPRO etc. (and some individual unit may drift lot of more than other.
Same for FEI
All these frequencies are walking around  as enviroment change.
What more... GPS  .. all times up and down.. but long time mean is ok. (this is why example 1pps discipline time constant need be long enough.

But then need ask how accurate it need be.

[JackOfVA]

Mmh, i never think that temperature is so important for a rubidium clock

Don't assume it's all in the Rb - both will shift with temperature change. The GSPDO will correct itself, but there's a time constant. Rbs are adjusted by changing a magnetic field (the C-field). So, any temperature sensitivity in the circuitry which controls that will affect the frequency.

The FRS does have provision for temperature correction, and I've got it set pretty close. Note that the changes are 2nd order, depending more on the rate of change than the actual temperature.

Looking at your plot, the time slope is not inconsistently correlated with the temperature slope. Generally, declining temperature corresponds to declining time but the sharp temperature snaps show opposite signs as the sharp drop in temperature corresponds to an increase in time.

That suggests several possibilities, including one that says the temperature change is not just inside the shop but rather reflects outside conditions and that the fast temperature excursions reflect storm fronts or diurnal shift and that causes a small change in atmospheric conditions, such as the refraction index which in turn causes a small but still significant shift in the accuracy of the GPS  timing signal and that the abrupt shifts in time represent corrections by the GPS reference, not the  Rb.

400 ns time shift corresponds to a path length change of ~ 400 ft or 130 meters. That seems within the realm of a change in refractive index due to atmosphere moisture content or temperature.

[mikes]

That suggests several possibilities, including one that says the temperature change is not just inside the shop but rather reflects outside conditions and that the fast temperature excursions reflect storm fronts or diurnal shift and that causes a small change in atmospheric conditions, such as the refraction index which in turn causes a small but still significant shift in the accuracy of the GPS  timing signal and that the abrupt shifts in time represent corrections by the GPS reference, not the  Rb.

Nah. It's just the furnace coming on.

[sergey]

Quick question about tools to be best used for calibrations. Is there are difference between plastic and ceramic screwdrivers and if so which affects less on variable capacitors and LC filters?

[SeanB]

The ceramic are slightly better for some, as they can have a lower dielectric constant. However they are very brittle, and break easily. Best is to have a PTFE one with ceramic tips, lowest loss and most likely to have a constant detuning effect on the circuit. longer ones are better as well to reduce the stray capacitance from you affecting the circuit.

[sergey]

The ceramic are slightly better for some, as they can have a lower dielectric constant. However they are very brittle, and break easily. Best is to have a PTFE one with ceramic tips, lowest loss and most likely to have a constant detuning effect on the circuit. longer ones are better as well to reduce the stray capacitance from you affecting the circuit.

Nice to know. Thanks!

[PA0PBZ]

I noticed Dave and others say 'dielectric', but isn't it just that you increase the plate size on one side of the capacitor? I know that the screw of these trimmers is connected to one side of the cap, and isn't dielectric supposed to be between the plates?

[JackOfVA]

I noticed Dave and others say 'dielectric', but isn't it just that you increase the plate size on one side of the capacitor? I know that the screw of these trimmers is connected to one side of the cap, and isn't dielectric supposed to be between the plates?

There's also a small stray capacitance from the adjustment screw and associated plate to the enclosure and ground and the rest of the circuit. It's only a tiny amount, but a small fraction of a pF is enough to shift the oscillator a few Hz. Hence anything that alters the stray capacitance will also shift the oscillator frequency.

[SeanB]

Depending on the type the dielectric is either air, air and plastic, air and mica or air and air. The trimming operation alters the ratio between the first and the second, either it is all air, all the second or a combination of them. There is also capacitance between both and the trimming tool, the surrounding board, the case and pretty much the rest of the universe, and most have both a lousy tempco and drift with applied voltage and time. Heat a pin on one without otherwise touching it and the value will change. Big ones are not so bad ( 200pF to 1n) but anything below 50pF is going to drift a lot. The art is to get the value close enough with NPO ceramics then use the trimmer to get the final value, preferably as the bottom part of a capacitive divider with a larger trimmer value.

[max-bit]

Frequency meter, has an input 10 MHz, better connect the output frequency standard, the counter inputs.
I am at home I have a GPS receiver (Trimble Thunderbolt) display (that shows the status of the receiver) and four 10 MHz output.
You can get STABILITY about 1x10-11, the more I would not count
Cesium clocks gain of about 1x10-13, 1x10-15's best
The most accurate clock is made ??in Poland pulsar clock about 1x10-16
http://en.wikipedia.org/wiki/Pulsar_clock
Cool the frequency standard although a little big

[EEVblog]

Cool the frequency standard although a little big

Yes, I'm temped to gut the rack and put into a smaller custom box, maybe with an LCD status display or something.

[max-bit]

as a status display suggest:
http://www.m1dst.co.uk/2013/02/announcing-trimble-thunderbolt-monitor-1-0-netduino/
(so as not to create from scratch)
although I do not know whether it will be compatible with your device
It is designed for GPS Trimble (thunderbolt)

Probably not, then you will need to write code from scratch
Or mount frequency standard Trimble

[rf-loop]

Frequency meter, has an input 10 MHz, better connect the output frequency standard, the counter inputs.
I am at home I have a GPS receiver (Trimble Thunderbolt) display (that shows the status of the receiver) and four 10 MHz output.
You can get STABILITY about 1x10-11, the more I would not count
Cesium clocks gain of about 1x10-13, 1x10-15's best
The most accurate clock is made ??in Poland pulsar clock about 1x10-16
http://en.wikipedia.org/wiki/Pulsar_clock
Cool the frequency standard although a little big

And you have some data about Thunderbolt stability?
(I have followed many Thunderbolts, HP Z3801's (what is far over TB stability) Good OCXO's and some Rb's. )

Please do real mesurement so that take continuously exaple 100 second averages from TB and and other source (example from other TB or other more short time stabile source)  then compare these and show me statistics, example 1k pairs compared.  After then tell me about what you mean with this:

You can get STABILITY about 1x10-11

  How big peak-peak differencies you find in this test example between  two TB and what you believe about stability after this imagined test. (I have not exatly this but I can quess with my other experience that you do not claim then even 1x10-10 or do you trust even 1x10-9 if take short time (example 1-100s)  freq average.
Also it depends how you have set your TB. How accurate your antenna position really is, how you have adjusted optimal parameters for your individual TB (example discipline). Or do you run with factory defaults. Also there is big differencies between TB's. Some have more bad OCXO than others... some may have really bad and some may have acceptable good individual OCXO from Vectron if have good luck. 

Long time average is good but is it stability?

What about if look what TB itself is thinking about itself, then, everybody have rigtht to his own religion withoud doubt ( smile ).

Then

in some publication:
"The widespread belief that pulsars are the best clocks in the universe is wrong, say physicists."
look attached image

Btw, If I want measure example frequancy. I hope I can do measurement in some minutes but I do not want wait very long time.

Poland pulsar clock about 1x10-16

But I do not want wait 100 year integration period. It is not so handy.

And it is good to think it is Allan variance... it do not mean...etc..

[EEVblog]

I had no idea what this "Tunderbolt" thing was.
Google helped:
http://www.trimble.com/timing/lab-kit.aspx

[max-bit]

Specifications Trimble:
http://www.trimble.com/timing/thunderbolt-e.aspx
There is further
http://trl.trimble.com/docushare/dsweb/Get/Document-383329/022542-010B_Thunderbolt-E_DS_0807.pdf
10 MHz accuracy . . . . . . . . . . . . . . . . . . . . 1.16 x 10 - ¹² (one day average)
I gave the value of 1x10-11 (especially a little worse)
For our purposes can be measured with an accuracy of 1 Hz to 100 GHz  

[rf-loop]

On surplus markets these are very common models.
http://www.ebay.com/itm/Thunderbolt-PRECISION-GPS-10mhz-FREQUENCY-TIME-Standard-/180399458965?pt=LH_DefaultDomain_0&hash=item2a00a54a95#ht_8267wt_1056

Mostly they are salvaged from from old 911 systems.

Mostly they are good, and many times more than enough for  hobby use freq reference use.

But it need know its limits. Specially for short time accuracy.

[rf-loop]

Specifications Trimble:
http://www.trimble.com/timing/thunderbolt-e.aspx
There is further
http://trl.trimble.com/docushare/dsweb/Get/Document-383329/022542-010B_Thunderbolt-E_DS_0807.pdf
10 MHz accuracy . . . . . . . . . . . . . . . . . . . . 1.16 x 10 - ¹² (one day average)
I gave the value of 1x10-11 (especially a little worse)
For our purposes can be measured with an accuracy of 1 Hz to 100 GHz  

Yes of course one day average.  And if all other things are optimally. It is just "up to" value.
It is totally different case than short time stability!
How long time you use for example measure some frequency. 10s interval... 100s interval..   if you blindly trust trimble is sure better than 1x10-9 momentarily. It is religion.
Of course it may be. But this is other case and if you  have not other realiable reference you do not know if it is accurate or far away.  It is good to remember.

"there is always exactly accurate time if there is alone and only one clock." It include more wise than it looks like first.. 

[EV]

Here is some statistics with my Trimble Thunderbolt. In the attachment is 8 measurements of 36 samples with 10 minutes gate time. So one measurement takes 36 x 10 minutes (6 hours).

[alm]

People shouldn't be too religious about rubidium oscillators sourced from a random eBay source either. Working rubidium oscillators have a very stable frequency, but it's not a primary standard, so no guarantee of its accuracy unless you calibrate it to another source.

A GPSDO is referenced to very accurate (not just stable) frequency standards in GPS satellites, so there is some assurance of absolute accuracy. The short term stability is limited by the (usually) OCXO, however.

[max-bit]

For amateur uses a huge quality
The frequency is calibrated to GPS
A GPS satellites are built Cesium clocks
Of course, you should be aware of the pros and cons.

In Europe they were used patterns such as relying on the DCF77 signal
In Poland, in 80's, the standard frequency signal was transmitted from:
http://en.wikipedia.org/wiki/Warsaw_radio_mast

[max-bit]

http://www.ko4bb.com/Manuals/05)_GPS_Timing/Datum/LPRO/Datum_LPRO-101.pdf
hmmmm 

[max-bit]

[rf-loop]

Here is some statistics with my Trimble Thunderbolt. In the attachment is 8 measurements of 36 samples with 10 minutes gate time. So one measurement takes 36 x 10 minutes (6 hours).

Yest, this same kind of what I have get with littlebit different timing and setup some years ago.

But then short time frequency from Trimble Tb, example 10s or 100s, it is different case.

More long average and GPS DO reach of course more and more GPS system accuracy.

Btv, EV, you have good LPRO individual and perhaps also its temperature is quite stabile.
(some LPRO's are more and some less sensitive for temp.)

I have many long records from test where was reference from good individual HP Z3801 as reference for HP53131A and under test some LPRO.  53131A was connected to printer and result is only there. ( It was for find good and bad individuals from "lot" of LPRO's and also for fine adjust them back to factory limits  5x10-11 after they was used variable time in some systems. After they arrive these was drifted (or mis aligned) so that they was randomly inside +-1x10-9. and some of units also failed...  but most of them need only adjust some internal things as described in service manual.)

In this case I find that trimble can not use for calibrating becouse it takes too long time to wait averaging for enough reliable accuracy.  If have only one what need adjust and have time for it, no problem. Trimble can use. But if need adjust many nearly as work... no way.

Of course there is much what can do for make trimble better. But in this time I did not know all things for this.
And becouse this time I have two very good Z3801 individuals running for ref, it was better solution.



 

[max-bit]

I do not see the problem
The precision of 1 (0,1) ppb is sufficient for applications amateur
and so no quartz clock will not be better ....
And the rubidium clocks from ebay and so there are new so ... we do not know what generates

[olsenn]

I do not see the problem
The precision of 1 (0,1) ppb is sufficient for applications amateur
and so no quartz clock will not be better ....
And the rubidium clocks from ebay and so there are new so ... we do not know what generates

If you're working with digital electronics and don't want to decompile your firmware into asm and count clock cycled explicitly, having a counter synchronized to your MCU clock can be helpful. Sometimes this is impractical, in which case a stable and accurate frequency source comes in handy. If nothing else, it gives you an idea of just how bad your crystal is.

[EV]

Btv, EV, you have good LPRO individual and perhaps also its temperature is quite stabile.
(some LPRO's are more and some less sensitive for temp.)

I have also two other not so good LPRO. One with 42 uHz std deviation and the other with 165 uHz std deviation. My two FE-5680A rubidium standards have std deviation of 73 uHz and 77uHz.

I use this Trimble thunderbolt only to adjust the rubidium standards to correct frequency. It is good for that purpose. The accuracy is good enough for me.

[NiHaoMike]

I noticed Dave and others say 'dielectric', but isn't it just that you increase the plate size on one side of the capacitor? I know that the screw of these trimmers is connected to one side of the cap, and isn't dielectric supposed to be between the plates?

There's also a small stray capacitance from the adjustment screw and associated plate to the enclosure and ground and the rest of the circuit. It's only a tiny amount, but a small fraction of a pF is enough to shift the oscillator a few Hz. Hence anything that alters the stray capacitance will also shift the oscillator frequency.

Why don't they design it so that the screw is connected to ground?

[SeanB]

The screw typically is grounded, but the approach of the trimmer to the other plate is still a capacitive operation.

[rf-loop]

I use this Trimble thunderbolt only to adjust the rubidium standards to correct frequency. It is good for that purpose. The accuracy is good enough for me.

Yes, for this purpose it is very good and simple (and of course it can use as reference directly if know its limits and this is enough for needs!

[Dr. Frank]

Hello Dave,
this device already will make a Time-Nut out of you, as soon as you install an antenna on the roof top of your house, and start to measure the stabilities of GPSDO and Rb standard.
Be careful, it's highly infectious!


In this thread, there's a whole lot of misinterpretation of the behavior and characterization  of Time Standards / clocks.

Stability of a clock is always referenced to the time scale, where this is measured, i.e. short term of 0.01, 0.1, 1, 10 sec (measuring time), mid term from 10 .. 1,000 sec and long term > 1000 sec. This is described / measured by the so called Allan Variance, a special statistic, that separates the different time areas of stability.
The simple variance calculation does not do the job correctly, as it will mix all the different time domains.

XTals in general in most cases have very good short term stability, i.e. 10E-9.. 10E-12.
Also the mediocre XTAL of Daves counter will have quite good short term stability, I suspect.
But its midterm and longterm stability is really bad, additionally, XTALs can be calibrated to 0.1 ppm resolution only, as demonstrated.

OCXO (running freely) have very good short term and midterm Allan variance stability, from 10E-11 down to 10E-13 for some special designs. They also can be trimmed much finer, < 10E-9, that means at least 100 times finer than the XTAL in the 54131A.
And they stay tuned on the order of of 10E-10 for weeks and months, depending on the design and vintage of the device.

OCXO were beaten only by Hydrogen Masers, which have 10E-13 .. 10E-16 from short to mid term (one does not need necessarily a Pulsar for that stability). They need only be calibrated once against a Cs clock and stay there for years.

Atomic standards as Rb, Cs have very bad short term stabilities, due to small interrogation times of the Cs beam, or the small path in the Rb bulb, and due to their measuring principles as such.
Their mid and especially long term stabilities are extremely high, 10E-11 .. 10E-13 can easily be achieved on small, commercial devices.

They normally contain also an OCXO, which is disciplined with > 100 s Time Constant, so those combinations have very good short term stability from the OCXO and even better mid/long term stability from the atomic standard.

Btw.: The american GPS system currently relies mainly on Rb based clocks: http://tycho.usno.navy.mil/gpscurr.html
The Cs clocks on board are either out of order (shorter operation time), and/or Rb has better midterm stability, and can easily be corrected by the ground stations.

GPSDO systems like Thunderbolt, or the one Dave has acquired, have a similar performance  like atomic standards, but due to the noisy transmission path, it takes longer to reach a similar  mid/long term stability.

<10E-11 is easily achieved for short to midterm, and for 10E-12 .. 10E-13 stability and uncertainty, it takes 1 to 10 days of averaging.
Therefore that GPSDO can be used to calibrate the PRS10 Rb standard within 1 day to about 10E-12.

Several of the Time-Nuts (e.g. John Miles) have Allan Variance diagrams and explanations online, for all the mentioned types of clock, so everybody can understand easily the concept of Clock Stability :
http://www.ke5fx.com/rb.htm
http://www.ke5fx.com/timelab/readme.htm

Conclusion: Daves Time Standard is really a very nice device, and I would not sell it.

Frank

[hairykiwi]

Dr. Frank - thanks for the really interesting background info in your post and the links.

There's a good description of frequency meter measurement techniques here

Cheers grumpydoc - also interesting reading.