[time-nuts] METAS tour report (was Tour of METAS (Swiss Federal Institute of Metrology) time lab: any questions or requests?)

[Pete Stephenson]

On 5/7/2015 11:38 AM, Pete Stephenson wrote:
> I recently inquired about taking a tour of the METAS time lab[1] and
> they said they'd be willing to show me around.
> 
> Is there anything in particular that fellow time-nuts would be
> interested in me asking them about or (if possible) photographing?

Hi all,

On Friday I toured the METAS Time & Frequency lab with two others:
Patrick, a colleague from work and Atilla, a fellow time-nut.

I'd like to report a bit about what we learned today that may be of
interest to other time-nuts.

UTC(CH) is composed of an ensemble of four commercial HP 5071A cesium
clocks and an active hydrogen maser.

METAS changed the definition of UTC(CH) several years ago. Details
regarding the old and new systems were presented at a conference in
2006. The only archive I can find of the conference paper is, for some
reason, available at a US military website[1]. I've mirrored the paper
at my site[2].

I'll quickly summarize the two systems below:

===== Old Definition of UTC(CH) =====

UTC(CH) had previously been defined as a computed "paper clock" named
UTC(CH.P) which was the weighted average of the four HP 5071As and the
maser. This paper clock was steered monthly to track UTC.

In the old system, the maser (the reference clock) drives a micro phase
stepper and the output of the stepper is steered to the paper clock to
form UTC(CH.R), the hardware real-time realization of UTC(CH). UTC(CH.R)
is connected to the distribution hardware.

Using a paper clock had two main advantages:
- It's more stable than any of the individual clocks in the ensemble.
- It is tolerant of hardware failure of the non-reference clocks. For
example, if the ensemble normally has N clocks and one fails, it
continues to work with N-1 clocks.

However, there are several disadvantages: UTC(CH.P) was computed for
only a single epoch each day so time measurements made at other epochs
can only be related to UTC(CH.P) via interpolation. Also, if the
reference clock fails the whole system is disrupted.

===== New Definition of UTC(CH) =====

In the new system, UTC(CH) is defined in real-time without interpolation
instead of being computed for a single instant each day.

UTC(CH) is now defined as a hardware master clock named UTC(CH.RT).

To quote the paper, "The UTC(CH.RT) hardware definition of UTC(CH) is
chosen from one of two independent master clocks: UTC(CH.A) and
UTC(CH.B). Each is the output from a DDS synthesizer, used as a
MicroPhase Stepper (MPS). Each is driven by one of the free-running
atomic clocks, and steered to track the paper time scale UTC(CH.P).

Auto-sense Fault Switches (AFS) are used to choose UTC(CH.RT) between
the A/B master clocks. The hardware redundancy between UTC(CH.A) and
UTC(CH.B) has two advantages. One is reliability: if one master clock
fails, switching to the backup master clock is instantaneous. The second
is continuity of service: if some maintenance of one master clock
becomes necessary, for example for the purpose of calibration, it is
possible to switch to the other master clock at one’s convenience
without interruption of service."

Typically clock A is the hydrogen maser and clock B is an HP 5071A. In
addition to both being reference clocks for UTC(CH.RT), both contribute
to the paper time scale.

===== Clock Vault =====

In addition to the maser and four HP 5071A clocks, there are several
other clocks that are used for various purposes but which do not
contribute to UTC(CH): a passive hydrogen maser, several rack-mountable
quartz oscillators, and at least one rubidium oscillator. All the clocks
are kept in a single thermally-regulated clock vault in the basement at
METAS.

While it won't be used to contribute to UTC(CH), a continuous-beam
cesium fountain is being constructed in the room next to that containing
the UTC(CH) clocks. Details of this clock are available at [3] with a
mirror at [4].

Although the Swiss are well-known for their fine watches and other
timekeeping devices, METAS is a rather small national time lab (compared
to, say, PTB, NIST, or USNO) with comparatively limited resources.

Their primary function is to provide a service to customers or users
rather than advance the state of the art of timekeeping: for example,
they provide NTP service to the public and UTC-traceable calibration
service to paying customers.

In the past they were the time source for the long-wave time signal
radio station HBG, but that service was discontinued in 2011. They are
doing some interesting research in regards to time and frequency, but
that is not their main focus.

===== Questions & Answers =====

Several fellow time-nuts had sent me questions that they wanted me to
ask METAS. Here's the questions and answers, paraphrased from my
shorthand notes:

1. Q: How does METAS generate UTC(CH)?

A: With four HP5071As and a hydrogen maser that contribute to a "paper
clock". See above, [1], or [2].

2. Q: How is the Swiss time scale linked to the rest of the world? GPS?
Two-way satellite transfer?

A: The primary link is a two-way satellite link (TWSTFT). They also have
dual-frequency geodetic GPS receivers for all-in-view multi-channel
dual-frequency P code time transfer.

According to BIPM Circular T, they use a combined smoothing of the
TWSTFT and GPS methods.

3. Q: Why was HBG discontinued?

A: It was expensive to operate and signal reception in mountainous
terrain was often limited. The antennas needed expensive renovation and
it was decided that it was easier and cheaper for users to switch to
DCF77 rather than repair HBG.

4. Q: What's the explanation for the bug in the transmission of the leap
second by HBG in 2005-2006? (See [5] for details.)

A: No idea. The scientist giving us the tour worked at METAS in the time
and frequency group at the time, but he didn't recall what might have
caused the bug. In fact, this was the first he had heard of it.

He then expressed dismay at leap seconds in general and wishes TAI was
more widely used as a time standard (e.g. computer clocks should use
TAI), with the conversion to UTC only occurring for display purposes.

===== Photos =====

Photos are available on Flickr at [6].

Unfortunately, conditions were not ideal for photography and photos of
the clock vault were taken from outside the thick, thermally-insulated
windows. Nevertheless, if anyone wants the original high-resolution
photos, please contact me off-list and I'd be happy to provide them.

===== Interesting Notes =====

1. The IT department at METAS is concerned about the security of
publicly-accessible systems such as NTP servers. As such, none of the
master clock systems or the computers that manage them are connected to
the internet.

Rather than use a typical computer network such as ethernet, the time
distribution systems attached to the master clock send IRIG B timecodes
over coaxial lines to the public METAS NTP servers[7]. This setup is
accurate to within a few microseconds, which is sufficient for NTP, and
helps isolate the master clock systems from the internet.

2. The servo loops for the HP 5071As and active hydrogen maser run with
time constants of one hour and one day, respectively. (Is this typical?)

3. METAS previously provided a time sync service over analog modems (the
actual modems and time code generators can be seen in one of the
photos). When the decision was made to discontinue the service, the
METAS staff wanted to notify users of the service so they could prepare
for the eventual shutdown without any surprises.

Unfortunately, Swisscom (the incumbent telephone company) refused to
provide METAS information on who was calling the modem lines because
their privacy policy (and Swiss law) forbids it. As such, METAS was
forced to discontinue the service with no warning. They never received
any complaints.

4. Our guide, Dr. Laurent-Guy Bernier, related an amusing story: the
Swiss watch industry used to conduct state-of-the-art research on
improving the accuracy of wristwatches. Later, they realized that if
people wanted accurate watches they were buying cheap quartz ones --
people weren't buying expensive Swiss watches because they used them to
tell time, they were buying them as status symbols.

Unfortunately, a similar situation came about with METAS: they provide
the legal Swiss time scale and offer traceable calibration services to
customers, but there's not really a lot of political incentive to spend
money on having METAS to offer more services (DCF77 and GPS are run by
the Germans and Americans respectively, meet people's needs, and cost
the Swiss nothing, so why pay to run HBG?) or do more than a little bit
of advanced research.

Practical? Perhaps, but I think it's too bad.

5. The exploits of several time-nuts are known to the cloistered
researchers at METAS: they find Tom's Project GREAT to be particularly
entertaining and interesting.

===== Thanks =====

Dr. Laurent-Guy Bernier at METAS was kind enough to take over an hour
out of his busy day to show the three of us around and I would like to
thank him for his time and expertise.

===== Links =====

[1]
<http://oai.dtic.mil/oai/oai?verb=getRecord&metadataPrefix=html&identifier=ADA474255>
[2] <https://www.heypete.com/pete/time-and-frequency/archive/ADA474255.pdf>
[3] <http://hal-obspm.ccsd.cnrs.fr/hal-00165376/document>
[4]
<https://www.heypete.com/pete/time-and-frequency/archive/EFTF07_FOCS_2.pdf>
[5] <http://phk.freebsd.dk/Leap/20051231_HBG/>
[6] <https://flic.kr/s/aHskcPUTko>
[7]
<http://www.metas.ch/metas/en/home/fabe/zeit-und-frequenz/time-dissemination.html>

If anyone has any questions or comments, let me know and I'd be happy to
do what I can to answer.

Cheers!
-Pete