Continuous recording of frequency shifts

[tkamiya]

Hi....

I've seen this before but I only vaguely remember....  Can someone please point me to a source?

I have two frequency source.  I want to CONTINEOUSLY record frequency shifts one relative to the other.  I know the oscilloscope method to compare two but it doesn't record and it doesn't record back and forth movement unless I'm watching 24x7.

I have oscilloscopes, pen recorder, two universal counters with A/B input and math function.  A few more stuff but too many to list.  What I need is to measure two GPSDO output.  Compare them.  See if they move relative to each other.  Possibly more than 360 degrees.  Likely, they will move +/- in relative direction over time.

Help?

[jpb]

You aren't very specific over time scales. GPSDOs should have very low Alan Deviations over a period of hours to days but will deviate much more over a period of 10s to 100s of seconds.
If you want to get the relative phase every second then one approach is to divide down to 1 Hz and measure it as A/B on the two inputs of a counter but the counter needs to have a very good one-shot resolution. The difference may be less than 10^-11 which is only 10 psecs. Doing it every 10 seconds or so is more realistic.

There are lots of other approaches such as using an offset frequency and mixing both down to a very low frequency.

I recommend Bill Riley's site
http://www.wriley.com/#papers

But I see, having just gone on it, that the US Government shutdown means that the link to NIST is currently broken!

[awallin]

If they both have 10MHz outputs then two PICDIVs can be used to divide down to 1PPS and the time-interval between the 1PPSs recorded with e.g. a TICC time-interval-counter.

If your GPSDO has an OCXO it might be quiter than this type of 1PPS measurement setup ( 2e-11 @ 1s or so) - but only up to maybe tau=100s or so - after that the measurement system 'wins'.
Against a stable reference (H-maser) GPS-receivers usually show 'huge' diurnals (12 hour perioid) of 10ns or so - so you'd better buy a H-maser if youre serious about this time-nuttery 

[tkamiya]

Haha.....

Yes, I will purchase a Hydrogen Maser....  I wonder if Home Depot carries it?

I wasn't very specific because I didn't know what to be specific about.  Thank you for mentioning short time change can be poor on this setup.

Yes, the oscillator is Oven Compensated.  They are both GPSTM unit by Nortel.  I am using 10MHz output only.

[jpb]

A simpler approach, depending on your counter, is to use one as the counter reference and measure the other but you really want to do back-to-back measurements. My counter (Tek FCA3100) will do this with a resolution (1 shot) of 50 psecs but this means that the limitation is around 10 seconds or longer - probably 100 seconds or so for the measurement system not to affect the outcome significantly.

If the two GPSDOs are similar enough to be assumed identical statistically then you can assume (approximately) that each contribute an equal amount and reduce the values by 3dB.

[nctnico]

For this you'll need a time interval counter. You'll probably want to use the 10MHz outputs because the 1PPS outputs likely drift. If you have a modern one (like from Keysight) you can let it record to a USB stick and read the data when done. A time interval counter can do a few measurements per second but that should be enough to catch any frequency drift because this will be a slow process on a GPSDO anyway.

[metrologist]

I had two GPSDO on my DSO with infinite persistence. They would drift +/- 180 degrees every few hours during some routine, but otherwise stayed in phase for most of the time. I've also created a time lapse video for this kind of thing.

[tkamiya]

I had two GPSDO on my DSO with infinite persistence.

I really like this idea!  Thanks for sharing!

[nctnico]

I had two GPSDO on my DSO with infinite persistence. They would drift +/- 180 degrees every few hours during some routine, but otherwise stayed in phase for most of the time. I've also created a time lapse video for this kind of thing.

That won't do you much good. You know how much the frequencies have drifted but not how that is related in time. When using a time interval counter you can import the data into Excel or Octave and make a graph versus the time/date of the event. You can also run statistics on the data.

[joeqsmith]

If you don't care about the rollover,  I would just null the phase,  mix them, into an LNA and into a data logger.     

[tautech]

Be prepared, it could be a deep rabbit hole you're entering:
https://www.eevblog.com/forum/metrology/an-advanced-question-sampling-an-oscillators-signal-for-analysis/

 

[rhb]

Be prepared, it could be a deep rabbit hole you're entering:
https://www.eevblog.com/forum/metrology/an-advanced-question-sampling-an-oscillators-signal-for-analysis/

 

Very true.  But Dan has made substantial progress following a rather rough start.  I suggest sending him a PM asking him to comment in this thread.  He's a very nice guy and is being *very* thorough.  He's basically trying to develop standards lab level skills in the subject.

[Tomorokoshi]

Here are the results from some dabbling I did. It doesn't answer the question, but perhaps some of the techniques have value.

1. HP 203A dual-channel signal generator with the HP 53310A showing the Time Interval between the two 10 KHz signals. The change in time difference is due to one revolution of the Phase Lag goniometer control, from 0 to 360 degrees, or a time lag at 10 KHz of 0 to 100 us.

2. HP 8601A Generator set to 100 KHz. Notice the polarized cyclical drift with a period of 16.6 ms. The 60 Hz power line frequency is getting into the system. Needs better power supply filtering.

3. HP 241A Oscillator set to 100 KHz. Now there is a non-polarized, somewhat cyclical drift with a period of 8.3 ms. The 120 Hz rectified frequency is getting into the system. Needs better power supply filtering.

4. The Time Interval between the HP 8601A and the HP 241A before tuning in the base frequency.

5. The Time Interval between the HP 8601A and the HP 241A after attempting to tune in the base frequency. The jitter characteristics didn't change. The correlation is random due to the noisy characteristics of the two source signals.

6. HP 35665A source oscillator set to 100 KHz. Much more stable with no power line noise.

7.  The Time Interval between the HP 35665A and the HP 241A. Much cleaner due to the much more stable source.

Note: All references to "HP 35565A" should actually be "HP 35665A".

[cncjerry]

As mentioned, one way is to divide the 10Mhz output down to 1PPS using PICDIV chips.  If you have a simple PIC programmer you can knock them out quickly.  You need a front end to the PICDIV and I use the front end to the TADD-2 off the TAPR site.  You can skip the transformer input if your source can handle it but you still need a 50ohm load.  Then as someone mentioned, you need a TIC.  I also use their driver chip to build a signal for 50Ohm output.  I have a 5371 or two and a bunch of counters but the TAPR TIC is decent.  That will give you something less than 100ps of accuracy (as a broad term).  The PICDIV chips have very low jitter.  This setup should be about as good as you can get at home for measuring ADEV in the range of .1 to 10,000 seconds.  Use Timelab directly from the TIC or other counters.  Or Stable32.  I wrote a program that logs to MSAccess instead in case there is an outage over the length of the run.  There are other ways to collect the data but nothing is easier than Timelab reading two channels directly from a TIC. 

As someone mentioned, the 1PPS output of the units will jump around, most likely, with some type of correction, depending.  The 10Mhz outputs when divided down should be very stable depending on the disciplined oscillator in the unit.

Search the time-nuts list on leapsecond.com for more info.

Jerry