[time-nuts] GPS discipline oscillator vs phase lock

Charles Steinmetz csteinmetz at yandex.com
Fri Jun 16 23:57:02 EDT 2017

life-speed at yahoo.com wrote:

> Perhaps I could implement an ISM band radio link for the purpose of locking the two oscillators.  Of course that wouldn't reach a couple miles either.

There appears to be some amount of talking past each other going on here.

First, I think you may have a fundamental misconception of phase locking 
as it applies in your proposed case.  If there are two GPSDOs, the 
oscillators are *already* phase locked -- each one to the GPS network as 
received at its actual location.  If you were to try to do some other 
phase-locking, at least one of them wouldn't be a GPSDO any more.  (That 
may not be a bad thing, if common-view GPSDOs can't achieve the required 

The two GPSDOs would, ideally, produce clock "ticks" identical to each 
other within picoseconds, which would be plenty sufficient for the vast 
majority of applications.   Of course, there are inevitably various 
errors, so in reality we do not achieve the full theoretical precision 
of the system.

The largest contributors to the differential errors (i.e., the phase 
difference between the two oscillators) are (or should be)  (i) mismatch 
in the cable delays due to differences between the lengths of the coax 
connecting each receiver to its antenna and/or the propagation 
velocities of the antenna cables (including the temperature coefficients 
of the cables), and (ii) differences in the GPS "solutions" in use at 
the two locations, which includes differences between the satellite 
constellations being used moment-to-moment by the two GPS receivers and 
the local reception conditions (quality of sky view, multipath, etc.).

Then there is (iii) the jitter of each GPSDO, which is not synchronous 
one to the other.  This includes the ionospheric path distortion [maybe 
this should be its own item], the GPS receiver electronics, and the 
locked oscillators themselves (including noise on the EFC line and the 
different instabilities of the two OCXOs).

Item (i) cable delay differences due to the cable lengths and/or 
isothermal propagation velocities result in a static offset.  Most 
timing-grade GPS receivers have a "cable length" setting that allows one 
to compensate for the cable delay (although that will not correct for 
the cable temperature coefficients).  This is all avoided if you use 
integrated GPSDOs (GPSDO built into an antenna housing).

It may be possible to reduce item (iI) by operating the GPS receivers in 
"single satellite" mode, both looking at the same satellite.  If long 
observations are required (such that the satellite in use must be 
changed during the measurements), this would become messy.  The tradeoff 
in "single satellite" mode is that while you eliminate errors due to the 
different satellite constellations being used moment-to-moment by the 
two receivers, but the the RF path errors and noise may increase, giving 
back at least some of the gain.

To reduce item (iii) errors, use identical GPSDOs with the very best 
OCXOs available.  You may need to select samples of the GPSDOs to 
minimize these errors.

All that will hopefully get you down to a differential phase of a few 
nS, at least for substantial stretches of time (using single-frequency 
receivers).  Times of day with low ionospheric distortion will produce 
lower differential phase than times with higher ionospheric distortion.

If you need better differential phase than this, you may want to 
consider solutions like White Rabbit (note that there may be issues with 
portable-type [movable] applications).

Best regards,


More information about the time-nuts mailing list