[time-nuts] Re: Timestamping counter techniques : phase computation question

[Bob kb8tq]

Hi

One thing that is still unclear: Why do you have two clocks tagging 
the events?

You have a simple crystal oscillator and are looking at 0.1 seconds. 
That oscillator likely “bounces around” ( = has a noise floor) at about 
1x10^-8 or so. Indeed, your “perfect” clock likely has similar issues.

Net result is that any process that gets you past 8 digits likely is fibbing 
about what’s going on. Yes this depends a lot on exactly how you define 
what you consider to be “going on”. 

If you have a 0.01Hz bandwidth filter, noise through it will not be much 
10 Hz off center. That averaging is part of the “fibbing” part of this. If 
you are trying to see how much the oscillator moves around, you just
lost that information.

Put another way: You are looking at a signal ( something completely 
constant) plus noise (the degree the result moves around). How you filter
the result impacts that noise. As you impact the noise part of what you
are looking at (or looking for) goes away. 

Bob

> On Jan 31, 2022, at 11:42 AM, Erik Kaashoek <erik at kaashoek.com> wrote:
> 
> Excellent questions.
> 
> The goal is to measure with a certain interval the phase of repetitive
> events versus a clock.
> The clock is assumed to be perfect. Also the counters are assumed to be
> perfect.
> The events can change their frequency and phase but any change within one
> measurement interval is allowed to be averaged.
> The measurement interval can be much larger than the period of the events,
> this allows the measurement to make use of the many events for for one
> measurement.
> 
> Practical example:
> The clock is a 200MHz perfect clock.
> The events come from an unknown quality XCO at 10 MHz
> The measurement interval is 0.1 seconds
> The goal is to derive the phase (and also the frequency) of the XCO versus
> the clock from the counter data in the highest possible accuracy. Once this
> is done the rest of the post processing of multiple measurements can be
> done in tools like Timelab.
> It would be extremely interesting if it is also possible to calculate a
> "quality"  figure that quantifies the amount of deviation of the events
> versus the presented measurement.
> This would help someone looking at result of the measurement to understand
> the real accuracy
> 
> And I am aware that as I don't know anything about this subject it seems
> simple, and probably is not.
> Erik.
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com -- To unsubscribe send an email to time-nuts-leave at lists.febo.com
> To unsubscribe, go to and follow the instructions there.