[ntp:hackers] A stop-gap authenticated time service
magnus at rubidium.dyndns.org
Sun Nov 15 12:31:53 UTC 2015
On 11/15/2015 12:12 PM, Poul-Henning Kamp wrote:
> In message <5648650F.1060502 at rubidium.dyndns.org>, Magnus Danielson writes:
>> If I get the time, I might derive the various ADEVs for noise through
>> PLL. Might be a useful article.
> You're provably better at math than me, so I'd probably not be a very
> useful reviewer, but to the extent I can assist in any way, just yell
> my way.
Will keep it in mind. I think Francois can verify that work when I get
around to it.
> My personal "hands-on" experience is that the P term is invaluable
> for initial capture, but once you're captured, it usually just adds
> high frequency noise and for tracking you are better off with a
> pure I regulator.
This comes as no surprise at all. As with any PI loop, the P term is the
AC path and well, noise will take that path. If you do not have an
additional low-pass filter to combat your white noise, lowering P (and
hence damping) is your only option. You do want to keep the damping up a
bit because otherwise you will get a high Q with it's side-effect.
> I'm not sure if this is why, but I do note that all of the HP506x
> atomic frequency standards only use an I term.
But they are by their nature FLLs, so it is to be expected.
> When I realized that the NTP kernel-pll's P-term noise profile was
> a source of instability, I started experimenting with different
> "decay" strategies for the P-term. I need to finish that and
> publish the results and code.
> Time, never enough of it...
Indeed, as always. If we could just have people produce some time for us...
More information about the hackers