[time-nuts] Re: CTI OSC5A2B02 OXCO testing

[Reginald Beardsley]

 A bit of background. I've got an absurd amount of HPAK kit. I bought a pair of used OXCOs on new PCBs via ebay or Amazon, I forget which. I set up a pair of splitters and fed the GPSDO to the 5386A counter and a DSO, did the same with one OXCO output and fed the input to the 5386A and the DSO along with a second OXCO just feeding the DSO.

I was able to set both units to the 0.1 ppb limit of the 5386A and made them up in SMA-F and BNC-F versions in small plastic cases.

The OXCOs have a 0.02 ppb/s stability spec which is presumably what you could expect from it in a GPSDO setting. The 10 year aging rate is stated as 0.4 ppb/yr.

I bought 10 bare OXCOs for $29.30 delivered. 

General plan:

---------------------------------------------------------------------------------------------------------------
Wire up all 10 to use a common PSU

Feed an LM399 reference voltage to voltage followers that drive a separate divider for each OXCO

Place a thermistor and resistor on each OXCO for use as a heating element and sensor to warm each unit above the threshold of the internal heater so the latter does not turn on and there are no temperature gradients across the device.

Set up a 10 port SMA relay tree to allow switching a phase comparator fed by a GPSDO and the switch output. *** The nature of this phase comparator is what I want to determine **

Place in a thermally controlled chamber and log the references for an extended period.
----------------------------------------------------------------------------------------------------------------------

At the moment my best option appears to be a nanoVNA H4 with the tinyPFA FW loaded. However, an HP equivalent would be very desirable. I'm an analog, RF oriented guy with a bit of a metrology bent and have no idea what HP made to measure such things. I've got 4 racks of HP and Tek gear. I would hardly notice more. That is my preferred solution. I have one of Leo Bodnar's dual channel GPSDOs.

Some years ago I got mixed up in basis pursuit to solve the heat equation. The answer is an infinite sum of exponentials. When I realized I was solving problems with 50,000 unknowns and fewer than 100 equations I became really fascinated as to how this could be possible. Which led to Donoho and Candes work in 2004-2008 and 3 years of reading math papers.

The aging behavior of crystals is well documented. Numerical experiment has shown that given measured curves I can predict aging to within a few percent for as long as I have history.

The aim of project is to characterize these OXCOs for 6-12 months. Then incorporate N of them in a single unit with an MCU that applies a voltage change to adjust for the ~0.001 ppb/day aging. And which periodically resynchronizes all N OXCOs along with applying any ambient temperature corrections required. Success is 0.1 ppb/yr for a 4 OXCO system. Possibly not achievable, but a worthy goal.

Why? Well, because I can and I've never seen any mention of doing it. Will I do exactly what I described? I hope not. I joined the list to get advice on how to do better.

Reg




     On Saturday, June 3, 2023 at 12:41:33 AM CDT, Forrest Christian (List Account) <lists at packetflux.com> wrote:  
 
 The way I would approach it as a beginner is with a 10mhz to 1hz picdiv and a timestamping counter like a tapr ticc.  Both of
Feed the ticc's reference clock with your 10mhz reference.   Then run the 10mhz out of the ocxo into the picdiv which will result in 1hz output.   Then run this 1hz into one of the ticc channels. 
That way you'll get a high accuracy timestamp of each 10,000,000th pulse.   If they are exactly 1 second apart you're right on 10mhz.   This last statement is oversimplified and ignores all sorts of potential errors such as accuracy of the clock source and other sources of measurement noise but is good enough to get you started down the time nuts path. 
Note that by logging each timestamp you can then feed that into timelab or another other similar tool and get some detailed statistical analysis of the ocxo over various time periods.  
One other method which may use hardware you already have is to feed your 10 mhz reference clock into one channel on a 2 or more channel oscilloscope and the ocxo output into the second channel.   Trigger off of the first channel and watch how much the signal on the second channel drifts.  If they are both on the same frequency there will be no relative drift.   If they are different they will drift and the rate will increase the further apart they are.   There is math you can do to determine how off if you can time how long it takes a single cycle to drift.  Note that it doesn't take much of a difference for this method to not be usable as you quickly get drift rates which exceed what a human can interpret on the screen.  
Note that all of the above is just to get you started.   Like I implied there is a lot of details I left out or oversimplified, but that should get you started.    
On Fri, Jun 2, 2023, 9:55 PM Reginald Beardsley via time-nuts <time-nuts at lists.febo.com> wrote:

Hi,

I've tried to avoid time-nutting for a long time, but I have lost the battle.

I bought 10 bare CTI modules on ebay for $2.93 each which I wish to test over a long period of time.  I can set up 2-4 GHz SMA relay switching, GPSDO reference etc.  Question is, what to read the frequency with.  My 5386A with a GPSDO isn't precise enough.

I don't know anything about it yet, but nanoPFA  FW on a nanoVNA  H4 looks attractive.  Are there other instruments/methods to consider?  I have a crazy mid 90's lab, so a bit more is not an issue other than where to put it.

All I'm looking for is a pebble tossed in the right direction.  Current plan is an LM399 reference with emitter followers driving Vref on the OXCOs, very stiff PSU and a fairly stable temperature.

Thanks,
Reg
_______________________________________________
time-nuts mailing list -- time-nuts at lists.febo.com
To unsubscribe send an email to time-nuts-leave at lists.febo.com