[time-nuts] Testing frequency using NTP Bruce GPS ps
XDE-L2G3 at myamail.com
Mon Oct 13 07:58:43 UTC 2008
"Tom Van Baak" <tvb at LeapSecond.com> wrote:
>> Now I am confused. I thought it used the internal crystal clock
>> for the timing, and selected which pulse to output for the 1PPS,
>> and this is what causes the sawtooth. If so, the timing is
>> determined by the crystal.
> Correct. For example, if the xtal were 20.00 MHz then the hardware
> timing pulse could be off by up to +/- 25.00 ns from the nominal
> value. But the firmware knows approximately how far off, and so it
> outputs this value as a correction message. The correction is not
> perfect, but if used, is always better than hardware alone.
So where did the 1ns granularity come in?
>> I don't use software to correct the sawtooth. It is a whole new
>> way of locking to the 1PPS.
> You might want to check out James Miller's excellent GPSDO.
> His simple design also does not use sawtooth correction and the
> results are very good. See:
Thanks, already studied and analyzed. He uses an XOR phase detector,
but the funny hump at 1,000 seconds bothers me. I can't see a good
reason for that. The drop starting at 1e4 is interesting, and it
would have been nice to see what happens further out.
His results also show that some other methods using phase/frequency
detectors are badly implemented. That is to be expected. PLL's are
tough for people not used to working with noise and crosstalk.
More information about the time-nuts