Precise, low cost GPSDO and frequency counter? And TCVCXO self-calibration

[sabia]

Hi, all

I'm setting up a lab at home to develop some serious real-time pieces of kit (sports timer), and for that I will need an ultra-precise (<<1ppm) reference oscillator and a frequency counter.

For my counter I really like the AE20401. Unfortunately its self calibration is only to within 1ppm, using an external 10MHz source like a GPSDO. But it has an internal 20MHz oscillator. So I was thinking of buying a 20MHz GPSDO and route the signal into the AE20401 instead of its regular oscillator. (Possibly with a buffer/level shifter in-between.) This should make it perfectly calibrated, always....right?

My timing system will use a TCVCXO with 0.14ppm variations that I will need to calibrate manually at first - in production I will perform self-calibration to a reference oscillator. These are 20.000 MHz, pullable oscillators, and the voltage to the adj pin will be fed from a SPI pot. I am thinking of routing the reference oscillator to a D-type and the on-board oscillator as the data. This should produce a signal with a frequency that is the absolute of

f_ref - f_onboard

so essentially a flat 0 or 1 if the two are exactly the same frequency. The CPU will measure the frequency at a few points of the SPI pot, and determine the ideal potter position (the pulling is reasonably linear).

I would love to hear from you eevbloggers out there - what you think of

1) the idea of using a 20 MHz GPSDO to turn the AE20401 into a perfectly calibrated frequency counter and
2) the idea of self-calibration with a D-type

Should it work, or am I about to spend money on things that won't work? Thanks!

[edavid]

I'm setting up a lab at home to develop some serious real-time pieces of kit (sports timer), and for that I will need an ultra-precise (<<1ppm) reference oscillator and a frequency counter.

1ppm is not difficult, but it still seems like overkill for a sports timer.

For my counter I really like the AE20401. Unfortunately its self calibration is only to within 1ppm, using an external 10MHz source like a GPSDO.

That seems like a poor choice, since you hardly need a GHz counter to develop a sports timer.  Instead, get a high resolution reciprocal counter like an HP 5335A or Philips PM6654.

But it has an internal 20MHz oscillator. So I was thinking of buying a 20MHz GPSDO and route the signal into the AE20401 instead of its regular oscillator. (Possibly with a buffer/level shifter in-between.) This should make it perfectly calibrated, always....right?

10MHz standard + simple doubler would be easier.

2) the idea of self-calibration with a D-type

Digital mixers have quite a bit of jitter.  It would be much better to use a GPS 1pps signal as a gate for calibration.

[awallin]

1) the idea of using a 20 MHz GPSDO to turn the AE20401 into a perfectly calibrated frequency counter and

GPS isn't perfect. TvB has measurements of different GPSDOs:
http://www.leapsecond.com/pages/gpsdo/

but if your frequency counter is only 8 digits or so, and the TCVCXO instability similar or worse, maybe it doesn't make a difference.

If the 10MHz REF input on your counter is any good you should use it and not bother with the 20MHz hack.
If you feed the same 10MHz from a passive splitter to both REF and CH1 input of your counter and collect gap-free data (hopefully it does Pi-counting) you can evaluate the noise-floor of the counter. If it would significantly improve by a 20MHz hack then the REF-input is really poorly designed...

2) the idea of self-calibration with a D-type

looking for a zero beat-frequency between REF and TCVCXO should work. How do you know if you are below or above? Maybe always approach from the same direction?

think real hard about your spec "<< 1ppm" and the TCVCXO you've chosen. Is it realistic to hold calibration within spec for any length of time and over temperature/voltage etc. variations? Good OCXO's age ~few ppb/day.

[edavid]

If the 10MHz REF input on your counter is any good you should use it and not bother with the 20MHz hack.

The counter he mentions has a 20MHz timebase, not 10MHz.

think real hard about your spec "<< 1ppm" and the TCVCXO you've chosen. Is it realistic to hold calibration within spec for any length of time and over temperature/voltage etc. variations? Good OCXO's age ~few ppb/day.

I think that spec is for his test equipment, not his TCVCXO based product.

[sabia]

Hi

Thanks for the replies, and sorry for not explaining better.

The GPSDO main purpose is to

* provide a (factory) calibration for the 0.14ppm on-board oscillator and
* ensure that the AE20401 frequency drift measurements are measured to a much better accuracy than the 0.14ppm oscillator that I want to measure the drift of

Keeping it all on 20MHz seems less of a hack to me than doubling a 10MHz, but then I'm not used to frequency doublers.

1ppm is not difficult, but it still seems like overkill for a sports timer.

For my counter I really like the AE20401. Unfortunately its self calibration is only to within 1ppm, using an external 10MHz source like a GPSDO.

That seems like a poor choice, since you hardly need a GHz counter to develop a sports timer.  Instead, get a high resolution reciprocal counter like an HP 5335A or Philips PM6654.

I have an old HP 5335A but there is no way to log frequency drift over time etc with that unless I get a GPIB PC interface, LabView etc. The AE20401 is around €100 and has logging software and USB interface. 

But it has an internal 20MHz oscillator. So I was thinking of buying a 20MHz GPSDO and route the signal into the AE20401 instead of its regular oscillator. (Possibly with a buffer/level shifter in-between.) This should make it perfectly calibrated, always....right?

10MHz standard + simple doubler would be easier.

2) the idea of self-calibration with a D-type

Digital mixers have quite a bit of jitter.  It would be much better to use a GPS 1pps signal as a gate for calibration.

That is certainly an option as well. Although I understood from another thread on here that the 1ppm duty cycle is extremely low (or maybe it was high?) - anyway the point being that the pulse to be captured needs to be coupled onto the device very well with a high bandwidth connection (i.e. no stray cap/inductance), as otherwise there will surely be a problem with repeatability between pulses? I'm thinking that the timer's 20MHz oscillator will be probed by an external board containing a D-type, and a cable from the 20MHz GPSDO. It will feed the mixed signal back to the timer's CPU for it to self-calibrate by adjusting SPI pot. Even if the 20MHz signal has poor slew rate it shouldn't matter as long as the D-type triggers on every pulse. If the 1pps pulse was used a poor slew rate would introduce larger errors, I believe. But I can try both methods

GPS isn't perfect. TvB has measurements of different GPSDOs:
http://www.leapsecond.com/pages/gpsdo/

Thanks, I'll check that out!

but if your frequency counter is only 8 digits or so, and the TCVCXO instability similar or worse, maybe it doesn't make a difference.

If the 10MHz REF input on your counter is any good you should use it and not bother with the 20MHz hack.

The AE20401 doesn't have a REF input, only a self-calibration. The best oscillator option it has has 1ppm, which is worse than what I'm trying to measure and calibrate. So because the counter has a 20MHz internal oscillator I figured I could kill a few birds with one stone.

If you feed the same 10MHz from a passive splitter to both REF and CH1 input of your counter and collect gap-free data (hopefully it does Pi-counting) you can evaluate the noise-floor of the counter. If it would significantly improve by a 20MHz hack then the REF-input is really poorly designed...

looking for a zero beat-frequency between REF and TCVCXO should work. How do you know if you are below or above? Maybe always approach from the same direction?

Yes, I was thinking to go from one direction.

think real hard about your spec "<< 1ppm" and the TCVCXO you've chosen. Is it realistic to hold calibration within spec for any length of time and over temperature/voltage etc. variations? Good OCXO's age ~few ppb/day.

That is why I want a GPSDO for absolute reference in the lab/factory, as there should be no ageing. The product (timer) will have a VCTCXO, which will of course age, and is allowed to drift/age a few ppm/year, and short term drift 1ppm.

[uncle_bob]

Hi

First and before you spend any money decide:

What level of precision do you need?

A stopwatch that is good to 1 ppm needs to run for several days before you can detect an error at that level in normal use.

What level of resolution do you need?

Generally resolution at 10X the precision spec is plenty good enough.

So, if something around 100 ppm is a rational precision spec, then resolution at the 10 ppm range is also reasonable. Both are guesses, put in your numbers as needed.

Your 5335 will work with TimeLab and be a much better counter than anything you can buy new for $100. The only thing you need is a (dirt cheap) GPIB interface. It goes into "talk only" mode and you log the data.

Bob

[sabia]

Hi

First and before you spend any money decide:

What level of precision do you need?

For calibration I need an oscillator with an absolute accuracy of 0.014ppm, and that will remain accurate to within that over years. I can't be bothered to send the reference to calibration every year. This number is based on 10x the accuracy of the TCVCXO that I want to self-calibrate.
For the actual timer product I have external requirements that dictate a short term accuracy of +- 1ppm. Ageing must be no more than +-3.0 ppm/year. Timing resolution must be 1/1000th second or better. I will be using 0.14ppm main timer accuracy.

A stopwatch that is good to 1 ppm needs to run for several days before you can detect an error at that level in normal use.

What level of resolution do you need?

Generally resolution at 10X the precision spec is plenty good enough.

My timing system will consist of many optical gates, which due to other constraints will need to be closely synchronized to the main timing unit. For cost reasons they will use 1ppm VCTCXOs. it is a time-of-day system, and the synchronization needs to be better than 200 us between all units. It's a bit hairier than your average stopwatch...

Your 5335 will work with TimeLab and be a much better counter than anything you can buy new for $100. The only thing you need is a (dirt cheap) GPIB interface. It goes into "talk only" mode and you log the data.

Bob

Great, thanks for the tip!

[uncle_bob]

Hi

First and before you spend any money decide:

What level of precision do you need?

For calibration I need an oscillator with an absolute accuracy of 0.014ppm, and that will remain accurate to within that over years. I can't be bothered to send the reference to calibration every year. This number is based on 10x the accuracy of the TCVCXO that I want to self-calibrate.
For the actual timer product I have external requirements that dictate a short term accuracy of +- 1ppm. Ageing must be no more than +-3.0 ppm/year. Timing resolution must be 1/1000th second or better. I will be using 0.14ppm main timer accuracy.

A stopwatch that is good to 1 ppm needs to run for several days before you can detect an error at that level in normal use.

What level of resolution do you need?

Generally resolution at 10X the precision spec is plenty good enough.

My timing system will consist of many optical gates, which due to other constraints will need to be closely synchronized to the main timing unit. For cost reasons they will use 1ppm VCTCXOs. it is a time-of-day system, and the synchronization needs to be better than 200 us between all units. It's a bit hairier than your average stopwatch...

Your 5335 will work with TimeLab and be a much better counter than anything you can buy new for $100. The only thing you need is a (dirt cheap) GPIB interface. It goes into "talk only" mode and you log the data.

Bob

Great, thanks for the tip!

Hi

Just get a cheap telecom Rb off of eBay and use that as a reference. It will run at least 10X better than your needs pretty much forever.

Bbo