[time-nuts] Phase locked local oscillator results

[Peter Schmelcher]

Hi all,

I have been playing with a phase and frequency locked local oscillator in 
my 3816A GPS receiver. The simple setup uses the OCXO output as an external 
reference input to a HP 3325A and with a trivial mod to the VP GPS receiver 
the 3325A becomes its local oscillator. The setup improves the short term 
stability of the OCXO. Which I guess in hindsight is not very surprising. 
Logging data every 5 seconds using GPSCon the AD drops to under 0.5. I 
tested two different approaches. One setup used a fixed LO frequency of 
19095750.191 Hz. To my surprise the prime number fraction chosen to give 
the time interval interpolator a workout results in an auto correlation 
peak at 1000 seconds when the EFC is viewed with the plotter program. The 
second approach used dithering. The LO center frequency was set to 
19095750.2 Hz and continuously ramped up and down every 10 seconds +-0.035 
Hz. This causes the pps pulse to dither slightly more than one sawtooth 
period over a 5 second interval (the GPS pps pulse edge is just a LO clock 
edge). Dithering however is a pain to adjust because timing the start of 
the ramp with respect to the pps output is trial and error and not at all 
repeatable with my setup. Both approaches result in a pps standard 
deviation of about 6.4ns. Further small improvements might be possible but 
would require adjusting the phase of the 3325A over the gpib interface to 
lock the sawtooth residual to some constant value.

For the truly adventurous the next level requires some artificial 
intelligence like 
<http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/19/30540/01408297.pdf?arnumber=1408297>http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/19/30540/01408297.pdf?arnumber=1408297 
unfortunately only the abstract is on the web. As I recall the authors 
report 2.5ns of uncertainty over a day when compared to a cesium reference 
and over 10 times improvement in the MDEV plot of the OCXO.

A second rather simple modification to reduce temperature induced tuning I 
tried is to power the oven with a separate isolated 12V supply. The only 
common connections are the EFC pin the reference output voltage pin and 
whatever point you decide to attach in the oven as a signal ground (if you 
are worried about power supply sequencing you can always add a diode). I 
connected the signal ground to the inner oven ground pin which does not 
have any voltage drops caused by the outer oven heater current. The ideal 
signal ground point would be somewhere inside the inner oven where no 
current induced voltage drops would modulate the EFC tuning voltage. To get 
an idea of the magnitude of the improvement you can measure the voltage 
drop from the oven ground pin to the ground pin of the DAC voltage 
reference. I haven't combined the isolated oven supply and the phase locked 
LO in a single test yet.

For the curious I have a 6 day GPSCon log file of the two LO setups 
compressed down to a half Meg let me know if interested and I will email it 
to you.

Enjoy,
Peter
nebula at telus.net