[time-nuts] Averaging effects

[Tom Van Baak]

Hi Magnus,

Another test that's really fun to do is to measure what effect
DUT frequency differences have on stability measurements.

Let me explain. Say you measure the ADEV of DUT1 using
DUT2 as the reference. Theoretically this measurement is
independent of the relative frequency offset between the two
independent frequency standards. This because, ADEV is
immune to any constant phase or frequency offset.

But in the real world, as you have found, and as I have seen,
you can sometimes see unexpected biases or perturbations
depending on the value of the frequency difference.

This affects the quality of the measurement -- and unless you
know ahead of time how susceptible or immune your system
is to this, the fidelity of your results are in question. And by
system, I mean the sum total of both DUT and the instrument
used to measure.

Call it injection locking, or frequency pulling, or some sort of
measurement system artifact involving interpolators or DDS
or ADC quantization, or analog or digital filtering, or output
resolution, or whatever. At a high level it doesn't matter what
the actual cause is -- you just want to know if it exists and if
so, what its magnitude is.

So here's how to see this with your own eyes. Start with two
free-running, slightly drifting BVA. Find the one that is, say,
drifting up in frequency and calculate the approximate drift
rate per hour. Then set its frequency low by about that same
amount.

Now run your measurement system continuously for 2 hours.
The relative frequency difference between the two will very
gradually go from some negative value, through zero, up to
some positive value. All the while you're collecting data. The
more data the better; the more gradual the better.

Using DAVAR, or by binning the samples and using ADEV
(or even just STDEV) you can then clearly see if there are
any gradual changes, or asymmetry, or quantized spikes in
the measured stability as a function of frequency difference.

/tvb