[time-nuts] +/- TI button on 5370B

[Magnus Danielson]

On 07/06/2013 11:02 PM, Tom Van Baak wrote:
>> Or have I misunderstood what you were saying?
>
> But what happens if, over a day, your DUT 1PPS wanders ahead and/or beyond the REF 1PPS? This is common with GPS 1PPS boards or with too-accurate house 1PPS references or when comparing poor quartz with a GPSDO.
>
> One symptom is that all your TI numbers look positive and right, but the *measurement interval* subtly changes from one clean measurement every 1 second to one clean measurement every 2 seconds. This kind of raw data is hard to process with standard time/frequency stability tools, because they expect tau to be constant.
>
> The solution is either a time-stamping counter, or to deliberately offset the DUT or REF by enough microseconds to avoid any sign changes in TIC measurements ever. I'm curious if you've discovered a reliable third alternative.

As I point out, sub-rate triggering can resolve the issue completely, at 
the expense of rate in samples. That is, you get stable results at a 
lower rate.

Another approach that has been used is to use a pair of counters, where 
the measurement-period of one covers the dead-time of the other. It's an 
alternative approach being relative to the sub-rate triggering.

> Without reproducing your ARM/START/STOP scenario here myself, it sort of sounds like you're moving the start ahead in time. True, this will give nice valid positive TI measurements but your consistent tau is now silently corrupted.

Which is what I wanted to avoid, unstable tau.

I did hint about a possible "cover-up" action by the software. There 
should be a separate trace in the measurement to indicate when covered 
up samples exists and a Performance Monitoring counter. This is an 
indicator of problems and quality. Potentially should the "degrees of 
freedom" be reduced accordingly.

Not indicating the missing samples on trigger-missing, is bad. Really bad.

Cheers,
Magnus