[time-nuts] Re: Timestamping counter techniques : phase computation question
Bob kb8tq
kb8tq at n1k.org
Mon Jan 31 16:04:36 UTC 2022
HI
> On Jan 31, 2022, at 10:23 AM, Erik Kaashoek <erik at kaashoek.com> wrote:
>
> There was a small error in my previous post
>
> Here are the complete and corrected numbers
> Below table shows:
>
> Event counter(Y) : The exact count of events since the start of the counter captured at the time of the event.
> Clock Counter(X) : The clock cycles since the start of the counter captured at the time of the event
> Calculated clock count: Calculated Y at time of event using the outcome of a linear regression using 20 captured (X,Y) pairs starting from the time of the event
> Wrapped event phase: as the Clock runs at a frequency different from the events the calculated clock count is converted to the event frequency and wrapped between 0 and 1 to see the variation in the phase
>
> The event and clock periods where constant. As the event period is a factor 20 longer than the clock period the phase resolution without regression would have been 0.05. With the interpolation over 20 captures it is at least a factor 2 better, consistent with the calculated accuracy improvement using a regression.
> In the actual implementation the regression can run on up to 100000 captures per second measurement time
>
> Any feedback?
Mostly a lot of questions:
What is the overall objective? What does the system do, what does it need to do?
Are the two clocks independent from each other or do they share a common source?
Iâm guessing they are independent.
What are the stability / noise characteristics of the clocks (and the events)?
Any math you do needs to consider the characteristics of the clocks. If you donât
do that, you can easily head down this or that rabbit hole.
One example:
If you simply dump a few hundred samples and only look at wide spaced points,
you do improve resolution. It may well be that the stability of the clocks at some
wide spacing actually does not support anything more exotic than this.
Bob
>
>
> Event Counter(Y) Clock Counter(X) Calculated clock count Wrapped event phase
> 280687207 1693025665 1693025665.303 0.061080
> 280688207 1693046999 1693046998.727 0.065341
> 280689207 1693068332 1693068331.970 0.061080
> 280690207 1693089665 1693089665.303 0.061080
> 280691207 1693110999 1693110998.727 0.065341
> 280692207 1693132332 1693132331.970 0.061080
> 280693207 1693153665 1693153665.303 0.061080
> 280694207 1693174999 1693174998.700 0.064056
> 280695207 1693196332 1693196331.947 0.059997
> 280696207 1693217665 1693217665.339 0.062770
> 280697207 1693238999 1693238998.732 0.065544
> 280698207 1693260332 1693260332.029 0.063852
> 280699207 1693281665 1693281665.326 0.062161
> 280700207 1693302999 1693302998.719 0.064935
> 280701207 1693324332 1693324332.016 0.063244
> 280702207 1693345665 1693345665.313 0.061553
> 280703207 1693366999 1693366998.706 0.064326
> 280704207 1693388332 1693388332.003 0.062635
> 280705207 1693409665 1693409665.382 0.064800
> 280706207 1693430999 1693430998.762 0.066964
> 280707207 1693452332 1693452332.046 0.064664
> 280708207 1693473665 1693473665.330 0.062364
> 280709207 1693494999 1693494998.710 0.064529
> 280710207 1693516332 1693516331.994 0.062229
> 280711207 1693537665 1693537665.278 0.059929
> 280712207 1693558999 1693558998.658 0.062094
> 280713207 1693580332 1693580331.942 0.059794
> 280714206 1693601644 1693601643.893 0.057494
> 280715207 1693622999 1693622998.543 0.056683
> _______________________________________________
> 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.
More information about the Time-nuts_lists.febo.com
mailing list