Measuring an OCXO - or not

[metrologist]

I managed to acquire a Wenzel large can OCXO. Unfortunately, it does not have a tuning pin, just an adjusting screw, so I probably will not be able to use it for a GPSDO. Although, I think moving supply voltage will move the frequency too, within limits...

I wanted to verify the aging so have a Trimple UCCM GPSDO referencing my counter. It takes a 10 seconds gate to read mHz. I am recording the reading every 24 hours, starting after a few days after adjusting close to 10MHz.

I used page 12 as a reference: http://freqelec.com/oscillators/understnding_osc_specs.pdf

Is this method going to give me any useful information? Looks like the aging is approaching 4E-10/day, using a log trendline projecting 2 periods.

[jpb]

The OCXO probably has better Alan Deviation over 10 seconds than the GPSDO so it seems a pity to take your measurements over just 10 seconds every 24 hours.
If you have a counter that can automatically take reading or allows a very long gate then I'd try taking average frequency over a day rather than 10 seconds, the GPSDO should be approaching 10^-12 or better averaged over a day and you should get a smoother measure of the drift as with luck temperature effects will average out over 24 hours for the OCXO.

Have a look at Bill Riley's page - it has some very interesting publications to down load and his software is now free as well:

http://www.wriley.com/

[metrologist]

The counter takes 10 seconds to make a single reading with this many digits enabled. I took a reading once a day at the same time for 10 days.

I thought I would have to take a reading for every 10 second gate, and average all 8640 readings over the 24 hour period. I don't have a GPIB interface yet, so I'm wondering if this is all useful?

[edpalmer42]

What kind of counter do you have?

The best way to measure Allan Deviation is to use a time interval counter.  Channel A goes to the GPSDO 1 PPS output.  Channel B goes to the DUT.  Measure the time from the start of the 1 PPS to the start of the next cycle from the DUT.  Depending on how good your DUT and counter are, you might need a high-performance sine to square converter.  Make sure that the counter isn't averaging multiple measurements.  Log the data via RS-232, GPIB, USB, whatever.  Either analyze the data later, or if you have software that supports your counter, you might be able to do it real-time.

Ed

[metrologist]

Counters are HP5385A and HP5314A. I don't think either one can do A to B counts here as I don't think the first does it and the second I think is limited in frequency.

[edpalmer42]

I agree that neither of those counters is good for Allan Deviation.  In general, ADev is a measurement that focuses on the short-term performance of an oscillator.  You need a good counter and need to take lots of measurements.  Maybe you should add a good counter to your list to Santa. 

But your aging measurement is useful by itself.  Don't try to convert it to Allan Deviation, just use it as aging.  In general, higher quality oscillators like to be kept running and the longer they run, the lower the aging goes - within limits.  Set up your oscillator with continuous power and let it cook.  See how good it gets.  The better the oscillator, the longer it will take to settle down to a constant aging rate.  Some oscillators specify 30, 60, or even 90 days of continuous operation before their aging spec is valid.

Although your data isn't really appropriate for Allan Deviation, realize that there's no need to do the calculations yourself.  The top software packages that do the calculations for you are Stable32, Timelab, and Plotter.  All are free.  Stable32 is the gold standard, but has a rather steep learning curve.  Plotter is good, but the author has passed away and there's no support.

Stable32:  https://ieee-uffc.org/frequency-control/frequency-control-software/stable32/
Timelab:  http://www.ke5fx.com/timelab/readme.htm
Plotter:  http://www.ulrich-bangert.de/html/downloads.html

All will accept data from a file.  Check the documentation for formats, etc.

If you really want to use this oscillator with a GPSDO, you should think about opening it up and adding the VCXO feature to it.  Or just buy an oscillator that had the feature already.

Ed

[metrologist]

It's just for grins now. I have several other high quality OCXO's that I will use for my Lars GPSDO build, this one is a bit of nostalgia - from friends in the industry.

Since I already set up Excel, I just need to plug in the daily frequency value and the charts update automagically. Continuing with the daily logs... This osc seems out of spec.

[edpalmer42]

It's just for grins now. I have several other high quality OCXO's that I will use for my Lars GPSDO build, this one is a bit of nostalgia - from friends in the industry.

Since I already set up Excel, I just need to plug in the daily frequency value and the charts update automagically. Continuing with the daily logs... This osc seems out of spec.

Really?  If I'm interpreting the second graph correctly, it looks like after 2 weeks aging is < 5e-10 per day.  That's a pretty good number.  What model is it?  Do you have the aging specs?

Ed

[metrologist]

It is a custom "to spec" similar to their Timekeeper series. I do not know what it's based from as the OCXO is probably 15 years old now, most of the time sitting in storage. Whatever they had from that era in this size.

http://www.wenzel.com/model/timekeeper/

I thought the spec was <2e-10, but their commercial offerings are <5e-10/day aging.

Also, wouldn't the red line in the first plot need to average under 5mHz to meet the 5e-10 spec? The log trendline projects 10 samples. That curve was going the other way in the beginning heading to 4mHz/day, and that is what I would casually expect. I think for the ADEV, I would need to have more many more (thousands?) samples per day?

Soon I will attempt a phase noise measurement...

Thanks for the comments 

[KE5FX]

It's just for grins now. I have several other high quality OCXO's that I will use for my Lars GPSDO build, this one is a bit of nostalgia - from friends in the industry.

Since I already set up Excel, I just need to plug in the daily frequency value and the charts update automagically. Continuing with the daily logs... This osc seems out of spec.

Really?  If I'm interpreting the second graph correctly, it looks like after 2 weeks aging is < 5e-10 per day.  That's a pretty good number.  What model is it?  Do you have the aging specs?

Ed

He probably has a Timekeeper-class part, or the similar 'Small Fry'.  Some of those were made without an electrical tuning pin.

If designing a new GPSDO, it would be a good idea to use a DDS instead of going to the trouble of driving a varactor from a DAC.  In other words, don't tune the OCXO at all, but zero in on the desired output frequency digitally.  So the lack of an EFC pin isn't necessarily the end of the world.

Edit due to simulpost: yes, you'd want to run the OCXO for at least a week in order to get a good confidence interval on the 24-hour ADEV.  If it is a NOS part or one that's been in storage for several years, it may take weeks to settle down to its final long-term stability.  It will essentially be undergoing a rapid-aging process, making up for lost time.  You may also see a lot of short-term frequency jumps during that period.

[metrologist]

I plugged it in on Sept 1st or 2nd, let it settle in and it's been on since. The spec requires 30-day warmup. After a couple weeks, I started playing with the trimpot. It always had an upward drift, so I adjusted it lower an amount based on projected settling. I wasn't sure if I was seeing mechanical relaxing in the trimpot.

edit: it's been on 39 days and the average of the latest 26 daily samples is 7e-10.

I'm not visualizing the DDS approach. Clock a DDS with the OCXO and then make corrections to the DDS output based on feedback from a PPS input?

[KE5FX]

I'm not visualizing the DDS approach. Clock a DDS with the OCXO and then make corrections to the DDS output based on feedback from a PPS input?

Yep.

[iMo]

There are 48bit DDS chips available. With ~10MHz input the resolution is 35 nanoHertz. You will not get 10MHz output without using the PLL multiplier inside the DDS, however. With 4x10MHz you get 10MHz 14bit sine output with 0.14uHertz resolution.

[KE5FX]

There are 48bit DDS chips available. With ~10MHz input the resolution is 35 nanoHertz. You will not get 10MHz output, however.

Sure you can.  Just keep track of the residual frequency error over time and reprogram the DDS to cancel it out.  You can maintain as much precision as you want on the controller.

[iMo]

@KE5FX: do you mean a software DDS? Or an external DDS chip?

[KE5FX]

@KE5FX: do you mean a software DDS? Or an external DDS chip?

You can think of an external DDS chip as the last stage of a software DDS.  The software knows exactly how much tuning error exists at any given moment, and (with some effort) it can keep track of how long that error has existed.  Each time it reprograms the tuning word to steer the GPSDO output, it can add or subtract a few parts in 10^-10 (or whatever) to compensate.  As long as the error is so small that it's impossible or impractical for an external user to notice it before it's corrected, it's all good.

[jpb]

Leapsecond has an analysis + measurements on the FE405 which uses this approach
http://www.leapsecond.com/pages/fe405/
But the FE405 uses the DDS ouptut in a PLL with a further output crystal to smooth out most of the DDS noise, there are still some spurs visible though.

[iMo]

While reading the first patent - this is something I need to understand:

The essentially zero-beat feed-back signal from the second mixer 30 closes the phase-locked loop and locks the vcxo 38 to the desired output frequency.

[metrologist]

So, it appears I am not measuring phase noise of the oscillator with my equipment. There is so much signal and I have a 10dB pad on the RF Input, the noise floor of the spa is ~-100dBm. The largest spur was ~-70dBc, but I was thinking of using the noise marker feature. I'll look again tonight.

[metrologist]

Posting another set of weekly data. I adjusted the plots so the traces should line up... and also the trendlines on the top plot are now linear and lower is a power, both projecting 10 samples. I think I should change the lower plot to MADEV.

[metrologist]

My ADEV chart always seemed suspicious to me. I checked the calculations and the running sum of diff squared dropped the $ signs, so was not actually creating a running tally. I corrected this and the plot seems more like what feels right.

[jpb]

My ADEV chart always seemed suspicious to me. I checked the calculations and the running sum of diff squared dropped the $ signs, so was not actually creating a running tally. I corrected this and the plot seems more like what feels right.

If you want to check your plots, I'd recommend Bill Riley's program Stable 32 which is now free (it used to be quite expensive):
http://www.wriley.com/
EDIT : I see I've already mentioned this in a previous post but in a slightly different context.

[metrologist]

Thanks, again.  I did import the frequency measurements into TimeLab, but I'm not sure if I am setting the correct parameters of if this kind of measurement is valid for the data. The results are much different.  Also, in TL I imported straight linear data, as if the OCXO would drive exactly 5mHz/day, and I get a straight line with upward slope where in my Excel chart I get a horizontal line.

[metrologist]

And I try Stable32 using a two column text file time(s) <-> Frequency

[KE5FX]

Thanks, again.  I did import the frequency measurements into TimeLab, but I'm not sure if I am setting the correct parameters of if this kind of measurement is valid for the data. The results are much different.  Also, in TL I imported straight linear data, as if the OCXO would drive exactly 5mHz/day, and I get a straight line with upward slope where in my Excel chart I get a horizontal line.

Unless you are taking one sample per day, your sample interval is a bit long.   Can you post your data set as an attachment?