New Cesium reference

[Mike Monett]

My Efratom uses a 20 MHz VCXO. A divider chain creates 5 MHz and
312.5 KHz. They're mixed in an xor gate to make 5.3125 MHz. The 20
MHz is also multiplied to 60 MHz. The 60 is combined linearly with
the 5.3125 and drives an SRD in a resonant cavity. Somehow some
line in this mess hits the 6.834687 GHz rubidium resonance, 114th
harmonic or something. The power must be minute.

I don't quite get this, but that's what the manual says. The
circuitry is fairly simple, just HC logic and a few transistors.
The whole thing is about as big as a coffee mug.

John

Thanks, John. That's for rubidium - cesium is a bit higher at
9,192,631,770 Hz. This means an even higher multiplier and more
phase noise or jitter, if you prefer. The snr degredation at a
multiple of 114 is already 41 dB, but that's not counting the 1/F
noise. It would seem they need the performance of a YIG to hit the
cesium line and stay on it. So the basic oscillator must be very
good!

Now how do they package a low 1/F VCXO, synthesizer logic, and the
needed multipliers and mixers into such a small low-power package?

The problem is even worse with the NIST-F1 cesium fountain. There,
the synth must generate frequency steps of about 9uHz at 9 GHz and
hold them for at least one second between measurements:

http://www.boulder.nist.gov/timefreq/cesium/fountain.htm

That must be some synthesizer - maybe it needs a NIST-7 to drive
it. Definitely not a weekend project in the basement

Regards,

Mike

 

[Mike Monett]

OK, I found the SRS PRS10 page - you're right. Rubidium is a very simple
design, no synthesizer needed. Just a 22 bit DAC driving a VCXO. At only
US$1495, everyone should have one

http://www.thinksrs.com/products/PRS10.htm

The NIST-F1 cesium fountain still presents a problem. If it's like the
previous cesium references, it uses a stable VCXO, perhaps driven by a
DAC the same as the SRS unit. Now where do they find a crystal oscillator
stable enough to hold 9uHz at 9GHz?

Certainly a quartz crystal would not be good enough for the mercury ion
clock that runs 1e5 times faster and is 1e3 times more accurate:

http://www.boulder.nist.gov/timefreq/

So even after all the discussions on 32,768 KHz crystal oscillators, we
seem to be many orders of magnitude behind the current state of the art

Best Wishes,

Mike Monett

 

[doug]

In message

Certainly a quartz crystal would not be good enough for the mercury ion
clock that runs 1e5 times faster and is 1e3 times more accurate:

http://www.boulder.nist.gov/timefreq/

So even after all the discussions on 32,768 KHz crystal oscillators, we
seem to be many orders of magnitude behind the current state of the art

Best Wishes,

Mike Monett

I had my attention drawn to the tiny caesium some months ago and
researched the sources and provided some observations for a daily
newspaper.
The system elements are being developed independently and combined in a
later separate contract.
The key advancement seems to me to be the application of a narrow line
width laser giving greater energy density so that a smaller volume of
gas can be used.
MEMs is an added goodie where we all migrate to if we can (I'm now in
MEMS gyros) .
The final product oscillator is not on the market .
I note with interest that the current quoted stability is only a good
TCXO which I take to mean 1ppe7 /1ppe8 pa
Most systems have been engineered to live with current stabilities
available at the cah dictated by the system
If 1ppe11 was on the market for 1$ there would not be a rush, .
Question :
What new applications woulf be first to take up the product?

 

[John Larkin]

Thanks, John. That's for rubidium - cesium is a bit higher at
9,192,631,770 Hz. This means an even higher multiplier and more
phase noise or jitter, if you prefer. The snr degredation at a
multiple of 114 is already 41 dB, but that's not counting the 1/F
noise. It would seem they need the performance of a YIG to hit the
cesium line and stay on it. So the basic oscillator must be very
good!

Now how do they package a low 1/F VCXO, synthesizer logic, and the
needed multipliers and mixers into such a small low-power package?

The problem is even worse with the NIST-F1 cesium fountain. There,
the synth must generate frequency steps of about 9uHz at 9 GHz and
hold them for at least one second between measurements:

http://www.boulder.nist.gov/timefreq/cesium/fountain.htm

That must be some synthesizer - maybe it needs a NIST-7 to drive
it. Definitely not a weekend project in the basement

Regards,

Mike

OK, here's the Efratom math:

20 MHz * 3 = 60 MHz (tuned transistor tripler)
20 MHz / 4 = 5 MHz (HCMOS)
5 MHz / 16 = 0.3125 MHz (HCMOS)
5 MHz * 0.3125 MHz = 5.3125 MHz (xor gate)

60 MHz * 114 - 5.3125 = 6834.6875 MHz (via SRD)

This is about 5 KHz, or about 1 PPM, off the hyperfine rubidium
resonance of 6834.682612, which is apparently close enough to get
exactly 20 MHz at the VCXO if the loop is offset a bit.

I'm impressed by the x114 multiplier/mixer (the cavity power must be
nanowatts) and by the stability they get working a full PPM off the
exact resonance. The optical modulation amplitude that they detect is
typically only 0.1%.

John

 

[Ken Smith]

Ken, you seem to have a great deal of inside information on these
clocks. Do you work with any of the groups in development?

I have known, slightly, one of the people in the project for many years.
I met with them just about a month ago to discuss their work. I design
systems that involve the same physics and some of the same parts as atomic
clocks (ie: magnetometers) so I end up talking to the clock folks from
time to time.

Also, how do they synthesize the required frequencies with sufficient
resolution and low enough jitter to find the cesium resonance?

They usually use a crystal to make the 10MHz output. They then usually
PLL multiply up to something near 100MHz and finally use a step recovery
diode to make the 9GHz harmonic.

The dithering usually is done in the 100MHz PLL section.

The high frequency jitter components (ei: far from the carrier) don't
really matter that much. The physics is a lot like super narrow tuned
circuit. It simply doesn't see the side bands.

The folks at Varian got this sort of stuff to work back in the 1960's
using bear skins and flint knives.

 

[Ken Smith]

I'm impressed by the x114 multiplier/mixer (the cavity power must be
nanowatts)

There abouts.

The light is in the tens of uW. The light that actually interacts is
about 10% of that. The RF has to be small compared to the light or you
"burn a hole" in the line. This lowers the stability my making the line
smaller and wider.