[time-nuts] How close can you trim a Cs?

[Chuck Harris]

John Ackermann N8UR wrote:

> 
> Again, sorry for being unclear in my other message -- I'm measuring the 
> time difference between two 1pps signals, and looking for a change in 
> that difference over time to determine the offset, so there's no gate 
> time involved.  I'm averaging 100 of those readings to generate each 
> line of logged data, then plotting that data to show the change over time.
> 
> John

Hi John,

That is exactly what I thought you were doing, and I don't think it will
work correctly.

Your counter, if I understand you correctly, is taking a measurement from the point
where it is tripped by the C-beam, to the point where it is stopped by the
GPS, making note of the time duration (to a 2ns resolution), storing the
result, and then computing the average after it has made 100 of these readings.

The problem with this method is the counter cannot be relied apon not to
add(or subtract) a certain amount of time (less than its resolution) to
each measurement.  If this was simply jitter, it should average out to zero,
but if it is a delay time, it will remain as a bias on your result.  This
bias will survive any amount of averaging.

The only way I know of reducing the affect of this sort of error is to start
your timer at the beginning of the C-beam's first 1 second pulse, and stop the
timer at the beginning of the GPS's pulse that comes nearest to the beginning
of the 100th 1 second pulse that comes from the C-beam.

This will give you a true average of the differences between the two 1pps rates,
+/- the 2ns resolution of your counter.  The best you can hope for with this
counter based technique is 2 parts in 10E11.  If you want 10E13, you will have
to up your time window to 10,000 seconds, or find a counter with finer resolution.

At least it seems that way to me.

-Chuck