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

[Bob kb8tq]

Hi

What you run into on a lot of setups are dead bands. You get a series
of (effectively) same / same /same /same sample groups. Applying math
to them gives you the (obvious) result … everything is perfect. 

If you have (say) a 200 MHz clock, your data will be “chunked” into 5 ns
spaced buckets of some sort. At 1 second you will have 1/2x10^8 sort of 
resolution. ( or 1/2x10^7 at 0.1 sec).  Playing with math to try to do better 
than this underlying structure is likely to get you in trouble. ( = the math
is fibbing, the results are limited by the physical setup). 

Indeed with some setups, your 200 MHz clock does not do equal sized
buckets (for various reasons) and the resolution will be a bit less. You very
much need to dig into the structure of the device you are using to see how
it does this or that. You might find that the 5 ns internal clock effectively 
is limited down to 20 ns for external events ….

So: Is this sort of resolution good enough for your application? If not, there
are well known techniques to improve on it. They are not crazy hard to do.
Yes, you are heading off into the world of analog this or that to implement 
them. 

Bob

> On Jan 31, 2022, at 3:54 PM, Erik Kaashoek <erik at kaashoek.com> wrote:
> 
> In the nist documents a separation is made between a measurement device
> that provides single observations spaced in time and the statistical
> processing that uses these observations over a long time period. My focus
> is the on methods to come to the best possible single observation as input
> to the statistical process, possibly extended with some confidence
> information. So a kind of decimation that is done using many underlying
> observations (the captured counters) almost equally spaced in time.
> From what I understood of the feedback is that the linear regression is the
> best method to do this decimation both for phase and frequency measurement
> and that modeling of the reference clock or other factors influencing the
> counters is possible but will become very complex very soon. Making a good
> event counter is still difficult
> The reason to go for the two captured counters is because I assume it is
> the most simple way to relate phase and frequency of a repeating event to a
> reference clock, in particular given the abilities of modern
> microcontrollers. If this assumption is wrong let me know.
> I will study the calculation method by Magnus as it may increase the
> obtainable speed of the underlying observation.
> _______________________________________________
> 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.