[time-nuts] Interesting paper: Don't GPSD' your Rb...

[ws at Yahoo]

Thanks, Poul
If that was their point then, I missed it completely.
The way I read it, I think you are giving them WAY too much credit about 
them understanding subtle things like that.

Yes we do agree that using low accuracy 1PPS signals can cause some 
unforeseen problems.
They are using 10 MHz for the Rb loop, so the 1PPS problems do not apply 
(much).
But note there is still the same approx 1PPS concerns even if the hanging 
bride things are filtered with a OCXO, 1PPS GPSDO.

A 1PPS signal with hanging bridges has a higher ADEV noise level so if used 
for disiplinng ANYTHING, the loop needs to be set a lot slower.
First order, it is about as basic as that.

If disciplining a low end Rb where temperature change causes the majority of 
the errors, then need the Rb loop time constant to be in the one or two Hour 
range.
If using a supper RB then the loop can be up to one or two days.
Like has been stated many times before, the discipline loop constant is 
(mostly) determined by the ADEV cross-over point between the Rb used and the 
GPS engine used.

One of the things I noticed is that their check data was plotted at ONLY one 
point per day.
That fact alone makes the details of the actual Rb, the GPS engine, and the 
disciplined loop of VERY minor importance.
Most any reasonable loop or Rb would give about the same results if that is 
the only data points considered.

ws

*******************

Poul-Henning Kamp said:

In message "WarrenS" writes:
>Magnus wrote:

>This is not a paper about "Don't  GPSD your RB", as the nut subject line
>suggest.

Actually, that is their point, or maybe more specifically:  Don't GPSDO
it the way we do with OCXOs.

At least as I read the paper.

It chimes pretty well with my own experiments and measurements.

The short term noise and offsets of GPS-PPS signals in most cases
is quite detrimal to disciplining a Rb.

Case in point:  Oncore + PRS10.

If you do not apply the negative sawtooth, there is no guarantee
that the PPS signals average will be coincident with the epoch the
receiver tries to mark.

Tom has some plots of the "hanging bridges" you will experience
and they can have durations so long that it starts to leak through
the PLL steering the Rb, thereby ruining the result.

This is the reason why you need to configure a PRS10 with a
timeconstant a fair bit shorter than theory predicts if you
just hook it up to a GPS-PPS signal.

Interestingly enough, applying the PRS10's 1/256th filter makes
worst case behaviour worse, because the filter makes the PRS10
"latch onto" even shorter hanging bridges.

If you cannot apply the negative sawtooth, you will get better
results by disciplining almost any random quartz xtal, ovenized
or not to the GPS, divide it down to PPS and then discipline
the PRS10 to that.

I belive that is the same thing the paper advocates, although
they communicate it very badly.


-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.