[time-nuts] Measuring gpsdo vs itself

[Magnus Danielson]

On 11/05/2012 06:30 AM, Tom Van Baak wrote:
> Ah, I see what you mean now. Yes, that setup can give you a rough estimate of the counter's noise floor.
>
> I can't give you specific numbers but one danger with this sort of test is that the input and the timebase are artificially locked together (i.e. fixed phase relationship) through the common reference. Your measurements may thus show artificially less noise than a real-life case of independent input(s) and reference.
>
> This can happen if your sub-ns counter is based on interpolators. Because the input and the timebase are locked in phase, the counter lands near the same point of the interpolator scale on every single measurement, rather than experiencing the noise (and non-linearity) of the entire scale.

It's a little more complex than interpolator non-linearities alone. You 
also need to include cross-talk between the signals. This cross-talk is 
usually higher between A and B inputs than from reference, but never the 
less.

You would need to sweep the trigger input delays to illustrate these 
non-linearities. From a single measurement you can get both a better or 
worse number compared to the average which is what you would expect to 
see for free-running signals.

So, you can get a rough idea about the baseline, but it is not a 
sufficient method.

See the SR620 manual for a plot of non-linearities.

Cheers,
Magnus