[time-nuts] Clock accuracy

[Chris Caudle]

On Fri, July 19, 2019 5:20 pm, donald collie wrote:
> Assuming the "accuracy" of the GPSDO is 1 part in 10^12
> then the inaccuracy after 100 years will be up to

That would assume that your oscillator maintained accuracy of 1 part in
1^12 without correction, and the GPS system stopped providing corrections
for 100 years.  Does not seem likely unless your GPSDO has a cesium beam
tube.

More likely is that without correction your oscillator is more like a few
parts in 10^9, in which case the drift would be magnitudes larger, or
there is some equivalent to GPS for the entire time, so your clock is only
off in the medium term by parts in 10^12 but gets constantly corrected by
GPS, so there is essentially 0 long term error, and maybe parts in 10^9
short term error, depending on the time constants in the correction loop
which compares the local clock output to GPS time.

> : 60x60x24x365.25x100x1x10^-12= 3ms [approximately]
> - which is probably good enough for an old fella. I have
> I wanted to know what sort of long term
> accuracy I could expect from the GPS constellation - looks as if 1 part in
> 10 to the 12th is about right.

That is not cumulative though.   Think of it more like the average error
from "true" GPS time when averaged over an hour, and GPS time should be
within some small number of nanoseconds of "true" UTC essentially always.

-- 
Chris Caudle