[aklug] Re: ntp and freebsd?

On 3/16/12 4:47 PM, Shane Spencer wrote:
> On Fri, Mar 16, 2012 at 3:45 PM, Leif Sawyer <lsawyer@gci.com> wrote:
>> GCI has stratum one time here in Anchorage and down in Seattle.
>>
>> ntp.gci.net =A0slaves off of it, so it's stratum-2 =A0 which is fine
>> for nearly everybuggy.
>>
>
> Sweet.. what makes up the Stratum 1 devices?

The most common source for a Stratum 1 NTP system is GPS.

At the base level, it could be a high end consumer type GPS units with
timing outputs (has a X PPS (pulse per second), usually 1PPS) to provide
interval, and serial output of time that can be identified accurately
against the pulse it applies to (there are a number of low end units
with 1PPS out but faulty Q/C in the design phase. They output the data,
but you cannot accurately match serial fed time data with which pulse it
applied to). Generally needs a good fixed external antenna with a clear
view to work well.

The NTP server then uses the X PPS output to discipline its internal
clock rate (slowly tuning over time to match), and gets the time from
the serial data.

This is more than enough to get +/- 1ms for log matching, etc but not
nearly good enough for things like one way latency/jitter measurements
with better than 1ms resolution or clocking uses (more telco stuff).

Next step up is a timing specific GPS. These are different in that they
generally do not solve _location_, only time and usually have 1PPS and
10Mhz outputs. A fixed mount antenna is not an option, it is required.

A timing GPS will start up, acquire sats and look for stored location
data. If stored location data is not present, or does not match within
limit it will go into "self-survey". It will collect hundred to
thousands of fixes, or run for a predetermined time (6hrs, 24hrs, 48hrs)
and average all of the good 3D fixes it gets over time. This averages
out the impact of atmospheric, weather, sat acquire/loss, etc.

The end result is a very, very accurate location in 3D.

The timing GPS then stops using the sats for both location and timing,
and uses all of them for timing info based on its known location.

The think about time/freq is that when you get to high precision, it is
all about average over time. GPS is very accurate over the longer term
(1000 seconds and up), but over the short term it is a bit jumpy.

The timing GPS will have a very accurate local oscillator, most commonly
a OCXO (Ovenized).

The OCXO receives very stable voltage, and the oscillator lives in side
the "oven" kept at a stable temp well above the environment. A OK OCXO
will be more stable than GPS over say 10 - 100 seconds. A really good
OCXO will be better than GPS for up to 1000 or 2000 seconds.

All the extra software, sensors, etc in the timing GPS will
compare/average over time the GPS data v.s. the OCXO. It will slowly
adjust the OXCO based on the long term average GPS info. The OCXO
provides high stability time ref up to a few hundred or thousand
seconds, the GPS signals "steer" things over time (compensating for
temp, OCXO aging, etc). A good stable GPSDO in a controlled
environment, wandering is generally measured in parts per trillion,
because parts per billion is too coarse. Should the GPS lose lock for a
time, the OCXO stability, plus some of the software magic that has been
collecting ternd data for it will generally keep the time/freq within
parts per billion even without GPS lock for at least a few hours.

There are other setups, for example GPS steered Rubidium but they are
less common in the field as they have fixed life-times (and Rb standards
are very stable over long periods but have a lot of phase noise in the
short term).

Good time/freq GPS gear is actually pretty common. Enhanced 911 for
cellular, as well as the timing needs of CDMA, GSM, EVDO, LTE, etc
require very accurate freq and time refs. Pretty much any modern cell
site will have a GPSDO timesource.

I don't qualify as a "time nut" I am fine with my home systems being +/-
1ms for NTP. But I am a HAM, and needed a good frequency ref to frive
the test gear. I have 3 trimble Thunderbolts upstairs. One is online,
working on a temp controlled case that will house 2 TBolts running full
time. One will provide the 10Mhz freq referency to the measurment gear,
the other will be a live cross-check of the first and the third will be
a shelf spare.
---------
To unsubscribe, send email to <aklug-request@aklug.org>
with 'unsubscribe' in the message body.
Received on Sat Mar 24 13:52:05 2012