[time-nuts] Cs stability
Richard (Rick) Karlquist
richard at karlquist.com
Tue Jul 17 12:52:22 EDT 2007
In case anyone still doesn't know who I am, I need to
mention that I designed the RF electronics in the
5071A in an earlier life, circa 1990.
Now that we have that out of the way, I will agree
with Magnus that the cited paper is severely flawed.
We don't know what models of cesium were used, and
the paper seems to assume that any deviations with
respect to GPS are due solely to the cesium standards.
We also don't know anything about the GPS equipment.
Citation  is a 5071A product note. I guess this
is thrown in to imply that one of the cesiums is a
5071A; but we can't be sure. A statement is taken
out of context from this product note to the effect
that "some cesiums don't have an independent means
of frequency setting", which might lead the casual
observer to think that this refers to the 5071A.
Actually, the 5071A does have an independent means
of frequency setting, as opposed to older cesium
clocks where you had to vary the C-field to change
the frequency. For this and other reasons, I don't
put much stock in the paper.
Having said that, let me speak about how we addressed
some of these issues in the 5071A. First, the matter
of temperature dependent phase shift. If you have a
really good frequency standard, you have to start to
worry about a temperature ramp resulting in a phase
ramp. Of course, a phase ramp is a frequency step,
by definition. The previous modele, the 5061B, like most
cesium clocks, had an architecture where the crystal oscillator
"flywheel" was connected to a splitter that drove
a microwave frequency multiplier from 10 MHz to 9.2
GHz and also an isolation amplifier. Both of these
were absolutely full of narrow band tuned circuits
with substantial temperature coefficients. The combination
of these "rubber bands" caused the 10 MHz at the output
connector to be fairly loosely correlated (in terms of 5071A
specifications) with the microwave signals sloshing
around the Cs cavity. In the 5071A, I used a combination
of a new architecture and better circuit design to
drastically reduce the temperature coefficient of phase.
You can read about this in my 1992 FCS paper (available at
ieee.org or www.karlquist.com/FCS92.pdf). We measured the RF
hardware by itself to determine its phase contribution to
the overall 5071A. Our resident perfectionist, Len Cutler,
insisted that all electronic error contributions be less than
a part in 10^14, including this one. I can assure you that Len was
a very happy camper with respect to the RF chain, and, if
you knew Len, you know that is saying something.
Shortly after the introduction of the 5071A, we had some
colleagues at JPL perform environmental tests on an early
5071A. For the better part of a year, they ramped up
and down the temperature, humidity, and pressure to see
if they could measure these tempcos. By using correlative
techniques, such as ramping the temperature on a daily
basis and looking for a fourier peak at 1 day, they were
able to establish a measurement floor of something like
a part in 10^15. They were never able to see any correlation
whatsoever of frequency to the environment for any of the variables.
Regarding magnetic field sensitivity. The mu metal shields
of the 5071A are so effective that it is not necessary to
demagnetize them. Even Len Cutler didn't think so. A standing
joke was the 5071A demagnetizer accessory that we were going
to build "one of these days". I think we finally did build
one for some customer who was even more of a perfectionist than
Len. He didn't see any improvement from using it. Anyway,
I take any statements about the earth's magnetic field affecting
5071A with a large grain of salt.
AC magnetic fields like 50 Hz, or 60 Hz, are also not a concern.
The frequencies used internally in the 5071A are carefully chosen
to avoid any correlation with those power line frequencies.
Magnetic fields even in the milliTesla range have absolutely no
effect, let alone nanoTesla.
Regarding vibration, acceleration, etc: earlier cesium beam tubes
did have a problem with the beam wandering around when the
clock was installed on a ship. In the 5071A, Len carefully
devised a technique to prevent beam wander from having any
effect on accuracy. Now if you tipped a 5071A upside down,
the 10811 would experience a "2g turnover" frequency step.
The control loop would respond to this within its time constant.
So, admittedly, you might have a temporary frequency shift.
However, there is no known application where a clock needs
to work upside down, so even Len couldn't justify worrying about this.
I hope this clears up any confusion about the 5071A, now
made by my good friends at Symmetricom.
Rick Karlquist N6RK
(now with Agilent)
These are my own opinions and don't represent HP, Agilent,
> > > In the article "OBSERVATIONS ON STABILITY MEASUREMENTS
> > > OF COMMERCIAL ATOMIC CLOCKS", Pekka Eskelinen claims to
> I think the paper lacks a number of necessary details in order to fully
> understand the conditions under which this exercise took place.
> There is no
> clear list of equipment used. It can be implied that it is 5071A
> being used.
> What Time Interval counters where used is for instance not known.
> He have compared PPS outputs between GPS receivers (which model
> are those?)
> and Cesium.
> I'd like to see some investigations into how stable the GPSes and
> equipment is to temperature.
> In the end, there are a few pages of material missing in there
> for it to give a
> fairly comprehensive picture. Many of these issues could probably
> be cleared up
> after some discussions with the author, but we should not have to
> rely on side-
> channels like that.
> It's an interesting article, but it leaves me with a few
> questionsmarks. It is
> a bit unsatisfying.
> > Does anyone have contact with the authors (Finland)?
> Should not be too hard as his email adderess is at the top of the
> first page.
More information about the time-nuts