[time-nuts] Low-long-term-drift clock for board level integration?

Chris Albertson albertson.chris at gmail.com
Mon Feb 20 19:47:36 UTC 2012

On Mon, Feb 20, 2012 at 12:15 AM, mike cook <michael.cook at sfr.fr> wrote:
> Le 20/02/2012 07:18, Bill Woodcock a écrit :
>> Murphy says we won't.  Bell curve, again.  A very few will have good
>> symmetric paths to Stratum-1 servers, most will have mediocre asymmetric
>> paths, and some will have nothing usable at all.
>>> Are you targeting homes, offices, or machine rooms?
>> The vast majority will be office buildings.  A few datacenters.  Probably
>> only a handful of homes.
> With the exception of the mobile units, the units' position can be
> determined with google maps or local survey data prior to installation so if
> you are not relying on the GPS receiver for time sync, why not forget it,
> set the data prior to shipment  and invest the saved cost on a better
> oscillator.

I've been thinking that all along.  The GPS receiver is not going to
be useful at all for timing.  Even for the first setup, even if GPS
was able to get you "perfect" time as soon as you connect top NTP that
perfect time will be wiped out with NTP's idea of the time.  So it's

Now that I've read you are building hundreds of these then I'd say
there is ABSOLUTLY ZERO chance of ntp working at the uSec level.  Yes
it might work now and then but NOT all 100+ of your systems.    For
that to happen would would require a statistical miracle right up
there with buying a dozen winning lottery tickets in one day.  It
might happen.    But getting 100+ NTP installations to work that well
is near impossible.   You WILL need GPS and even then getting to 100%
is hard.  Some sites simply will not have a good view of the sky or if
in a high rise building running antenna lead down 25 floors through
conduit would cost to much.       The ONLY way to get 100% success
rate with hundreds of installations is to have multiple options.
"coockie cutter" or turn-key just can't work.  MOST places will not
have a good enough network connection for uSec level times.  Some data
centers will.  Small offices and homes will not.    Most places will
have a way to set up GPS.   Then in the remaining places you can use
cell phone based radio clocks.  They cost more then GPS and have 100
times worse performance but they might be the only option in most

You said the person installing this will be rather clueless.   The
"fix" for that is good customer tech support.  You can walk them
through the decision process about what kind of clock is best for
their site then help them set it up and test it.  Testing is key.
Have a few fall backs if the tests fail.

Bottom line is that open loop timing at the level of 10 uSec per year
costs well into 7 digit figures (Cesium clocks) so you need some
connection to the outside world.   So must choose a connection type.
The available options are

1) NTP.  Attractive because it is free but in most cases it will not
meet your accuracy requirements, only in places with "good" networks
like co-location facilities and some large data centers or if you are
lucky and the customer already base GPS connected NTP servers on his
local network (many places do have this.  It is common with larger

2) GPS.  Works perfectly, not expensive but it does require an antenna
that can see much of the sky.   You may need a long antenna lead cable
and running long wires in a commercial building can be expensive.

3) CDMA reference clocks.   These can work at the uSecond level any
place there is decent CDMA cell phone coverage.  The clock and its
antenna can be located indoors.
Here is an example of one:

4) There may be other ways to connect to the outside, phone modems,
WWB broadcasts.  Loran is still running in some parts of the world.
Your customer may already have a precision timing system of some kind
and you can use that. (We use IRIG in our lab)   But you do this case
by case.

If this were me.  I'd have GPS as the default option but then tel the
user he does not need it if a GPS connected NTP server is already
installed at his facility, then he could just use NTP.  Then use CDMA
as a fall back if those fail.

Chris Albertson
Redondo Beach, California

More information about the time-nuts mailing list