Best practices question

Paul Theodoropoulos paul at
Wed Jun 26 22:11:47 UTC 2019

On 6/26/19 13:35, Gary E. Miller wrote:
> Yo Paul!
> On Wed, 26 Jun 2019 13:14:36 -0700
> Paul Theodoropoulos<paul at>  wrote:
>> I think maybe I'm  hitting a bug or I'm just not understanding
>> (usually the latter), as it doesn't seem possible:
> No one totally understands this.
There is some comfort in that.
>> jitter fluctuating between 400 parts per trillion and 800 ppt?? From
>> a Raspberry Pi running at 1.4Ghz?
> Here is mine:
> My Standard Deviation is 48.504 ppb (10e-9)
> Your Standard Deviation is 82.8 10e-12)
> So about 500x better then mine.
> But my numbers are over 24 hours, not one.  So the diurnal ambient changes
> are in mine, not yours.
> I get a lot better when I put the RasPi in a temp chamber and control
> the CPU temp.  But not 500x better.
Yeah. 500x seems impossible to me - I mean, the processor clock on the 
RPI3 is 1.4 billion cycles per second - so I don't see how you could 
quantify an interval a small fraction of that cycle. But I've never 
claimed that math was my strong suit!
> That said, I switched to min/maxpoll 1 for PPS, and saw a
> corresponding change - so Time jitter appears to be inversely
> correlated to Frequency jitter - less frequency jitter, more time
> jitter, and vice-versa. This is all fascinating.
> This sort of relates to a current discussion on time-nuts.  They
> are discussing the relative merits of a PLL vs an FLL.  Similar effect
> here.
> The shorter poll keeps the time closer to the PPS, but magnifies the
> jitter.
> "And if you go chasing rabbits, And you know you're going to fall..."

Yup - here's a shot showing the effects more clearly - at min/maxpoll 0, 
2, then 1.  1 sure seems like the best compromise, as the time jitter 
returned to roughly similar values as the original 0 -  though a longer 
window into this will obviously be more informative.

Paul Theodoropoulos

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the users mailing list