[time-nuts] GPS 1PPS, phase lock vs frequency lock, design

[Charles Steinmetz]

I think you may be missing the most likely primary contributor.

Each GPS receiver (and, thus, each GPSDO) tracks a constantly-changing 
"constellation" of satellites.  Each rx switches constellations as it 
sees fit, depending on reception conditions as it sees them, and no two 
receivers will track the same constellations, switching at the same 
time, even if they are fed from the same antenna.  Most GPS receivers 
switch constellations quite frequently (at least several times per 
minute, sometimes much more frequently) even with strong signals.  At 
each switch, "GPS time" as computed by each rx changes by a few nS 
(maybe more, depending on the quality of the unit's "time solution" 
algorithms and the signal environment).  You can see this dynamically if 
you run each receiver into a separate instance of LH (there seems to be 
some finite latency in LH's constellation reports, but I'm not sure how 
much -- perhaps Mark will comment).

So, for short tau (averaging times), there is quite a bit of jitter on 
each receiver's time solution, which is *not* correlated between 
receivers even if they are fed from the same antenna.  (I.e., the jitter 
is almost all differential, very little common-mode.)

Of course, we already knew that raw GPS data at low tau has bad jitter 
(compared to the jitter after averaging for 1000+ seconds, which is what 
we think of naively as the precision of GPS), so all this should come as 
no surprise).

Some GPS receivers let you switch into "reduced switching" or even 
"single satellite" modes, but this turns out to be much less helpful 
than you might think with real-world signals.

Hanging bridges can cause significant phase jumps, but they should be 
much less frequent than most of the changes you are reporting.

Best regards,

Charles


ed wrote:

> I think I have a setup that exemplifies this situation, and some
> anecdotes. A while back, I acquired two "identical" GPSDO boards, and
> boxed them up together, with common environment, power supply, and GPS
> signal via a splitter. I've mentioned this thing a couple of times here,
> and had planned to do some experimenting to see how they track each
> other, if crosstalk at the front-ends may have effects, etc. I haven't
> done any of this yet beyond looking at the relative phase of the 10 MHz
> outputs on a scope, over various periods from minutes to days.
>
> I had expected them to agree quite closely after enough running time,
> and be quite stable, but was disappointed. The phase drifts up and down,
> sometimes very, slowly, over an hour or so, and sometimes quickly,
> noticeable over a few minutes observation time. After some pondering on
> why identical units with the same GPS signal should drift like this, I
> realized that besides possible front-end interactions, and noise, that
> this was likely mostly from the sawtooth effect - the discreteness of
> phase comparison of the 1 PPS vs 10 MHz counting, and discreteness of
> OCXO tuning voltage via the DACs. They each responded differently, for a
> number of reasons. More time and voltage resolution would help, of
> course, but they will never perfectly agree, even in this idealized
> setup with identical units. Virtually identical, that is - there's no
> such thing as truly identical units, and operating in identical conditions.