[time-nuts] How do I know my GPS stabilizedoscillator isworking?

[Magnus Danielson]

From: "Tom Van Baak" <tvb at leapsecond.com>
Subject: Re: [time-nuts] How do I know my GPS stabilizedoscillator isworking?
Date: Mon, 31 Jul 2006 22:08:02 -0700
Message-ID: <003d01c6b528$78275240$0b0ff204 at computer>

Tom,

> > Well, in your above measuring case with 34 missing pulses, you *DO* have a
> > problem if you are going to do Allan Deviation (ADEV), Modified Allan
> Deviation
> > (MDEV) or Time Deviation (TDEV). Replacing the start and stop as you did
> above
> > will destory that information. The gap will cause a measurement error.
> 
> Right, unless otherwise corrected, the missing data
> may mess up statistics. But you can't correct for the
> missing data unless you know the data is missing.
> Hence the emphasis on a data collection method
> that can detect gaps, should they occur.
> 
> Note in that example yesterday, the 34 second data
> gap exists regardless of which start/stop method you
> choose. The only difference is that one way you get
> visible notice that a gap exists and the other way
> you'll never know.

Indeed.

In ADEV for instance, may no part of the sum use sample points from the
different measurement blocks. The double-derivate may only calculate on
consecutive sample points having the intended tau difference. Also, the
drift-rate compensation can only be performed per block, as the gap will
destroy the continous property of drift-rate compensation. You may calculate
the drift average from each block, but care must be taken to weight each such
contribution properly.

> Timestamps -- I should mention the other way to
> detect gaps in GPS 1PPS data is to timestamp each
> measurement. I use an RS-232 logging tool that
> prepends a MJD timestamp to each serial data line
> as it read. The log for the GPS18 looks like this:
> 
>     53893.093681 8.9755  us
>     53893.093693 8.9368  us
>     53893.093704 8.8986  us
>     53893.093716 8.9214  us
>     53893.093728 8.8829  us
>     53893.094131 34.0000088063 s
>     53893.094143 8.8310 us
>     53893.094155 8.7332 us
> 
> Using this method one can detect gaps in the data
> simply by looking at the deltas in successive time
> stamps. And then it doesn't matter which channel
> is start/stop.

Either that, or any other method providing a local clock which ticks on.

> > For ADEV/MDEV/TDEV measurements you also want to compensate out driftrate
> > before doing the actual calculation. The frequency error is not a problem
> for
> > the calculation, but the frequency error can be a problem for some, but
> not
> > all, counters.
> 
> I was under the impression that these statistics were
> invariant with respect to phase and frequency offset
> (but not frequency drift).

ADEV is certainly not invariant to frequency drift, but it is to phase and
frequency offset. The second derivate of the continous time will null out
phase error (first derivate will do that) and frequency error (second derivate
will do that) while the drift rate will become a offset in the second derivate.
So, by averaging the second derivate, subtracting that from the derivates and
then square the residues and average that you end up with the propper Allan
Variation, rootsquade and you have your ADEV.

MDEV/TDEV has the same sensitivity to frequency drift rate.

Maybe I should write a TN on that, there is infact a trick to calculate the
ADEV directly while compensating it without having to store all the data.
I haven't seen it in any literature or articles, infact drift rate compensation
is rarely mentioned either.

> Can you elaborate on the counter problem?

I will have to come back on that to make a fair presentation.

Cheers,
Magnus