GPSDO issues

[Yrrah]

Hello,

In 2020 I purchased a GPSDO in the far east. Reason for that one was that it had an Oscilloquartz Star 4+ OEM board inside. Of good reputation. Just before my Z3801A had died. Indeed, upon arrival I immediately confirmed the OEM board was there. And some piggy back boards as well. The biggest boards, around the OCXO is marked: amoj1010 2020.04.28. No mention of BG7TBL who made many of these GPSDO's, despite the case looks very similar to his designs. I never really liked its 10 MHz output signal spectrum but for the purpose (ext clock for my HP5342A): so far it was good enough. I use a 15 MHz lpf.
Now I am working on a 10.808 MHz LO (stabilized by Isotemp 131 OCXO) and found that it travels about 10-15 Hz, quite rapidly. As I also have a good MV89A OCXO I used this as clock for the counter and now the signal only travels a couple of Hz, and slowly. But I need the GPSDO for accuracy. So I did some measurements on the GPSDO.
I had available a diode ringmixer in a tinned box with connectors. Th IF has a lpf at about 100 kHz. The 10 MHz signals go to the RF/LO port of the mixer. The IF signal goes to a Ultra low noise LF amplifier (60 dB), similar to what Wenzel published, and then to a Focusrite audio card. Signal analysis is with Audiometer, by DG8SAQ.
As sanity check I first connected the Isotemp and MV89A. This the red curve. The vertical scale is not really calibrated, but the curve looks good. The oscillators are not locked as you would do to measure phase noise correctly. This gives a kind of cross correlation spectrum.
Then I changed the Isotemp for the normal output of the GPSDO. This is the blue curve. This shows a lot of trouble that is not quite understood. I hope someone with experience with this particular GPSDO can shed some light on what is going on, what is wrong.
Finally, with the datasheet of the Star 4+ I managed to find an internal connector of the Star 4+ that was not used in this case, but should give a 10 MHz sinusoidal signal. I connected this signal to the mixer: the green curve. That is much better, but here the catch is that the signal goes slowly up and down about 10-15 dB. I measured the dc output from the mixer and find a dc voltage slowly traveling from -300 to + 300 mV. Clearly an indication of the actual phase difference. Guess what: the two phenomena are completely in sync. What is going on here? The input cap of the ULNA is about 1-2.2 uF. Is the phase voltage somehow modulating its gain?
Well I am looking forward to your comments.

[MIS42N]

For good performance, a GPSDO needs a good GPS signal. Do you have a well positioned antenna?

[GigaJoe]

if i understand it right : you try to measure 10800 Mhz - 10 gigahertz , with reference signal of 10M ?

[Yrrah]

For good performance, a GPSDO needs a good GPS signal. Do you have a well positioned antenna?

For sure. My antenna is on the roof and has 360 free sight. The Star4 reports lock. But..... it could be defunct of course. It does not by any means explain the odd spectrum I measure....

[Yrrah]

if i understand it right : you try to measure 10800 Mhz - 10 gigahertz , with reference signal of 10M ?

That's correct. Nothing wrong with that. Many microwave counters rely on 10 MHz clocks. It's the quality of the clock that counts....

[GigaJoe]

gps constantly adusting frequency of ocxo by changing control voltage

all i can suggest to create very very stable control voltage for standalone mv98 ocxo , even use a styrofoam chamber to minimize affect of environment.
adjust against gpsdo and use as the reference, to minimize jitter ,   cause by control voltage and thermal regulator circuits ...
 

[MIS42N]

Now I understand. Operating at 10GHz you get deviation of 10-15 Hz.

Most GPSDO are phase locked to the GPS and in the long term will be accurate. However in the short term most GPSDO rely on the 1PPS (one pulse per second) for timing, this is dependant on the GPS module. The cheaper modules such as u blox neo 6/7/8 follow the received signal faithfully, this is affected by various things so can change +- many nanoseconds. The short term output of the GPSDO is dependant on the algorithm that averages the 1PPS. This may be the problem. Better results are obtained using a timing GPS module such as u blox lea m8t.

I designed a GPSDO, it averages the 1PPS usually over 512 or 1024 seconds. With an m8t gps module and 1024 second averaging, it is stable to 1 part in 10^10, or 1Hz @ 10GHz.