What short term frequency stability to expect from cheap TCXO

[erikka]

I did some searching on the forum but I could not find the info I was looking for so here is my question.
When measuring the output frequency of a cheap 10MHz TCXO there seems to be a 3Hz FM modulation with a deviation of about 0.1Hz.  (10e-8 short term variation)
This was measured using a 10 digit per second frequency counter and a gate time of 0.1 second
When using a gate time of 1 second the FM modulation is no longer visible and the stability of the 10MHz output is within 0.06Hz (below 10e-9 short term variation)
The speed of the frequency variations is the roughly same as the correction speed of the TCXO when subjected to an external temperature variation.
This was tested using 0.1 second gate time and touching the TCXO causing a large jump away from and to 10MHz.

Could it be the FM modulation is caused by temperature swing caused by the on/off switching of the heater by the temperature control in the TCXO?

[ogden]

You forgot to provide key piece of informaiton - make, model and (pointer to) datasheed of your cheap TCXO.

[erikka]

Not sure it will be of great help but here it is: https://datasheet.lcsc.com/lcsc/2103291206_Yangxing-Tech-OP252030MEDA4SI-u_C382499.pdf

[bdunham7]

Could it be the FM modulation is caused by temperature swing caused by the on/off switching of the heater by the temperature control in the TCXO?

If it is a TXCO, there's no heater.  If it is an OCXO, there is and perhaps that is the issue.

Edit: Now that I've seen the data sheet, I don't even see that it is TC, and since it uses a PLL, perhaps the frequency deviation is the PLL operation. 

[AndrewBCN]

Could it be the FM modulation is caused by temperature swing caused by the on/off switching of the heater by the temperature control in the TCXO?

If it is a TXCO, there's no heater.  If it is an OCXO, there is and perhaps that is the issue.

No, it's not an OCXO, and it's not a TCXO either. It's a crystal oscillator + one time programmable PLL multiplier in a single package. I don't see any temperature compensation mentioned in the datasheet.

Basically the short/medium/long term stability should be the same as any average crystal oscillator, and its frequency will vary as a function of ambient temperature, supply voltage, vibration, aging, etc.

[erikka]

If it would be a XO plus pll one would expect random short term frequency variations. These I did observe using a true XO. However the frequency variations are too much periodic. So suppose the data sheet is not of this device and it is indeed a TCXO in this small form factor. Could that explain the short term periodic frequency variations? Or am I chasing a ghost?

[bdunham7]

it is indeed a TCXO in this small form factor. Could that explain the short term periodic frequency variations? Or am I chasing a ghost?

Even if it were a TXCO, there's still no heater.  Whether you are chasing a ghost or not depends on how sure you are that the issue isn't with your frequency counter.  If your FC is much better than that and the periodic variations are well within the datasheet specifications, the 'why' becomes more of an academic issue, since the device would be deemed to be working properly by the specs.  If I had to guess as to 'why', I'd say it was just the operation of the PLL which may have a bit of lag or hysteresis.

[ogden]

Not sure it will be of great help but here it is: https://datasheet.lcsc.com/lcsc/2103291206_Yangxing-Tech-OP252030MEDA4SI-u_C382499.pdf

Provided info is great help indeed. Keyword "programmable" means PLL which means "lowest grade" of TCXO. No wonder you see artefacts of regulation in it's output.

[edit] Wait... I can't find any mention of TCXO in datasheet. Am I blind?

[erikka]

My mistake.
This is the correct datasheet
https://www.kds.info/wp-content/uploads/2015/11/dsb321sdn-1-d_pdf_en-1.pdf

I could not find the spec of the short term variations but if the measured 0.1Hz variance makes sense its clear what limit there is and I do not have to debug other parameters like supply instability

[AndrewBCN]

My mistake.
This is the correct datasheet
https://www.kds.info/wp-content/uploads/2015/11/dsb321sdn-1-d_pdf_en-1.pdf

I could not find the spec of the short term variations but if the measured 0.1Hz variance makes sense its clear what limit there is and I do not have to debug other parameters like supply instability

Yes, for a commercial-grade TCXO I believe a short term stability (over 1s) of 0.1Hz makes sense (as expected it's much better than a simple XO but much worse than an OCXO).

From the (new) datasheet, the sensitivity vs supply voltage is ±0.2×10E-6 Hz max. with Vcc ±5%. Any 7805 can do much better than +/-5%, so in principle I don't think you have to worry about your power supply (unless it really has a problem).

[erikka]

After some more testing I get these results.
Stability of the TCXO with gate time of 0.1 seconds over a 20 seconds period: 10e-8
So almost no drift and only the quick fluctuations
Stability of the TCXO with gate time of 1 second over a 30 seconds period, again no drift and much lower fluctuations: 10e-9
The difference between 0.1 second and 1 second is a factor 10 which  makes sense.
To validate the measurement method and to avoid I'm actually measuring the fluctuations of the frequency counter I measured the short term stability (30 seconds) of a good OCXO with a 1 second gate time and it's better than 10e-11
Now I need to learn how to create an Allan deviation chart from frequency measurements so the experts can give feedback after one quick look.

[fourfathom]

This varies hugely with the type of TCXO.  Here's a Fox TCXO (10 MHz) as the temperature slowly changes from 37C to 39C, then back down to 35C: The small 0.1 Hz steps are my measurement resolution, the approx 1 Hz steps are the TCXO digital correction (plot is frequency vs time):

The TCXO stays within its specifications, but the jumps may not be acceptable (they weren't for my application).  Even in a stable environment the frequency will appear to randomly jump +/- 1 Hz.

Here's an Abracon 10 MHz TCXO with a smoother correction characteristic -- it looks like a more traditional linear compensation design, but I don't actually know what's inside it.  This plot is frequency vs temperature, over several cycles.  This TCXO also stays within spec, but has much less random-looking variation.  You can see the "retrace" effect on the frequency.  Even OCXO's exhibit some degree of retrace:


Finally, here are some measurements I made on a SiTime MEMs TCXO (10 MHz), both freq vs time, and freq vs temperature.  I was using a "reptile incubator" as a thermal test chamber, and you can see effect on frequency when the chamber fans were cycling on and off (I later modified the chamber to fix this problem and give me better control of the temperature):

This oscillator is a micromachined silicon resonator.  It also stays in spec, shows "retrace", and has a very clean output, FM-wise.

[AndrewBCN]

...
Now I need to learn how to create an Allan deviation chart from frequency measurements so the experts can give feedback after one quick look.

Erik, I suggest you take a look at Lars' DIY GPSDO PDF documentation, he explains how to connect the USB serial output of the Arduino to TimeLab and generate those nice plots of ADEV and MDEV vs Tau.

TimeLab download: http://www.miles.io/timelab/beta.htm

And two documents linked from the TimeLab page:

1. https://www.nist.gov/publications/handbook-frequency-stability-analysis

2. http://www.wriley.com/Techniques.pdf

[erikka]

Here are the measurement using timelab
Measuring an OCXO with 1s gate time gave the OCXO-OCXO graph.
Measuring an OCXO with 0.1s gate time gave the OCXO-fast graph.
Measuring a TCXO with 1 s gate time gave the TCXO-OCXO graph.
Measuring a TCXO with 0.1 s gate time gave the TCO-fast graph.

My concerns:
- Should the 1 second gate time and the 0.1 second gate time graphs connect? (assuming the 12 digits/second of the counter should be better that the deviation with 0.1 second gate time?
- What is causing the fast oscillations on the OCXO-OCXO graph?

[thinkfat]

What was your test circuit? The datasheet says nominal load is 10k // 10pF. In my experience, not loading an XO correctly can cause increased phase noise.

[erikka]

The TCXO drives a PLL that generates the 10MHz.
The OCXO has a buffer amplifier and outputs 10MHz directly
The output loading of both is probably with less the 10pF and higher then 10k. Would that be a problem?
Can phase noise have such a repetitive pattern?

[Terry Bites]

We don't know what your application is.  What drift can you live with, thats the question? Find the part with the spec you need.
If you only take measurements on one device you won't know what the spread of characteristics will be.

You can always drop the manufacturer an email but they might not want to commit themselves! KDS quote 1ppm (1Hz per MHz) per year ageing and 2.5ppm over temperature. It will be far less than this on a short term basis and low temperature fluctuations.

[erikka]

Drift is not my current concern. The only concern is the frequency instability within one second seen with a gate time of 0.1 second. Is this a measurement problem? Can it be ignored?

[thinkfat]

The TCXO drives a PLL that generates the 10MHz.
The OCXO has a buffer amplifier and outputs 10MHz directly
The output loading of both is probably with less the 10pF and higher then 10k. Would that be a problem?
Can phase noise have such a repetitive pattern?

Well, I have a Taitien OCXO that has a very distinct FM noise that gets much reduced if I terminate it with its nominal 50 Ohms load. So, its sure possible.

[fourfathom]

To verify that your measurement system isn't at fault, just measure a different oscillator.  That's one reason I have so many measurements.  I even built a simple one-transistor oscillator using a "junk box" 10 MHz crystal, with no temperature compensation, just to see if I measured the same oddball behavior as my jumping-around Fox TCXO exhibited.  The junk-box oscillator wasn't very temperature stable but was otherwise clean.  It also showed "retrace", as well as a typical AT-cut frequency/temperature curve.

[AndrewBCN]

Drift is not my current concern. The only concern is the frequency instability within one second seen with a gate time of 0.1 second. Is this a measurement problem? Can it be ignored?

An OCXO in a low-noise test circuit (meaning both the power supply and the control voltage are low-noise, and you have a proper 0.1µF decoupling next to the OCXO) with a fixed load of around 1k/30pF should be very stable in the short term (anything less than 1000s).

So if you find you are measuring a noticeable frequency instability, it could well be a measurement problem.

As usual, to measure a clock's ADEV/MDEV, you need a better clock. So the question is, what are you measuring the frequency with?

[erikka]

As reference I'm using an Accubeat AR60A Rubidium clock. Frequency measurements used as input to timelab are done with a Picotest U6200A locked to the AR60A

[thinkfat]

This AR60A is a quite compact model. How sure are you about its short term stability? Have you tried finding the noise floor of your setup, for example, by measuring a very good OCXO?

[erikka]

What I did was to compare the AR60A with an OXCO using a 1 second gate time and they agree over a 100 seconds period within 2e-11. Just on the edge of the 12 digits per second resolution of the U6200A. So for short term stability with a 1 second gate time both seem ok. My concern is when I use a 0.1 second gate time and the frequency measurements start to have fluctuations that are not like noise but a real FM modulation, invisible when using a 1 second gate time. Maybe I should use a direct to DC radio and check if I can see any very low frequency output from the mixer