[ntp:questions] PRN23 failure (was Re: wireless routers beating on NTP servers)
jcs at no_s_p_a_m.zoo.bt.co.uk
Sun Jan 25 15:11:31 UTC 2004
In article <x7vfn61qgu.fsf at capsicum.wsrcc.com>,
"Wolfgang S. Rupprecht" <wolfgang+gnus20040120T090554 at dailyplanet.dontspam.wsrcc.com> writes:
> jcs at no_s_p_a_m.zoo.bt.co.uk (John Sager) writes:
>> In article <x7fzed3oe3.fsf at capsicum.wsrcc.com>,
>> "Wolfgang S. Rupprecht" <wolfgang+gnus20040117T190822 at dailyplanet.dontspam.wsrcc.com> writes:
>>> Lets home the critical infrastructure can be convinced to buy their
>>> own GPS based stratum-1's to supplement external time sources.
>> Excuse me if I laugh, Wolfgang. The recent PRN23 fault (you wouldn't
>> have seen it in SF) showed up some shortcomings in GPS time sources.
>> I suspect it didn't have too big an impact on NTP clocks, but
>> a lot of our GPS-referenced backup network frequency sources dropped
>> off line for an hour or two.
> Finally something exciting happens with the GPS constellation and I
> entirely miss it? Sheesh! (I see I'm going to have to google to see
> what I missed.) I do have full second-by-second lat/lon/elev/time
> logs covering the last 2 months so if it effected me or anyone my ntpd
> talked to I should be able to spot it. The map at Shriever seems to
> indicates the SV had the good graces to fail when it was over Asia and
> not visible from the Lower-48 US states.
> I did think a bit about which mode to run my gps in and I use normal
> "3-D" (really 4-D) mode. I would hope that that mode is more tested
> than the timing mode. Perhaps the code to remove the outliers would
> have worked??? Was there a any resolution in terms of which
> units/firmware did the right thing?
I'm not sure very many receivers did well. Our time source receivers
drifted up to 700ns over a period of time & then rapidly jerked back
when the sat disappeared from the solution. Someone posted on
sci.geo.satellite-nav that some Scandinavian CORS-type stations
went down for a while. I checked out some of the RINEX data from the
Ordnance Survey reference stations, and it showed problems. The exact
behaviour depended on receiver manufacturer but they showed anomalies
in pseudorange & phase observables from all sats, not just PRN23.
My view is that with 7 or 8 sats visible, it shouldn't be that hard to
detect outliers & remove them from the solution, especially for
receivers that know, a priori, where they are. I hope we'll see
some improved software releases coming out of this.
More information about the questions