[time-nuts] The amazing $5 timestamper, part 3 (and: how do you calibrate an LPRO-101?)

[Jeremy Elson]

Fellow nuts,

This week, I've been working more on my "$5 timestamper" based on the
STM32G4 chip. I've finally been able to use it to get some nice results
comparing the frequency of a couple of LPRO-101 rubidium frequency
standards I have to SI seconds via a GPS receiver.

My original email to this list on my new timestamper, in February, had a
version of this experiment. Unfortunately it was flawed because there were
still some bugs in the analog front-end of my timestamper that I had not
yet discovered. As a result, there were discontinuities in the timestamps
when the clock line going into the timestamping chip had noise (generating
extra pulses) or wouldn't quite be high enough voltage to go over the
chip's threshold (causing missed pulses). These have been fixed, as I
reported in my second email on the timestamper (in April).

I moved recently, and now that I have GPS set up in my new lab I was
finally able to redo my February experiment to measure the frequency of two
LPRO-101 rubidium standards I bought on eBay for about $200 each. The
seller ("test_tool") claimed to have calibrated both before sale. However,
I discovered the performance of one of them was almost two orders of
magnitude better than the other. The test setup was:

1) The device-under-test (LPRO-101) was used as the 10mhz reference clock
for my timestamping board.

2) An early eval board of a ublox M10 GNSS (EVK-M101) with a decent sky
view, was configured to listen to 3 constellations (GPS, Galileo, GLONASS).
I did not use location surveying so the accuracy is probably less than it
could have been but the reported 3D position was quite stable.

3) The PPS output of the uBlox M10 was attached to one of the timestamper's
input channels.

I did this with two LPRO-101 units. The resolution of the timestamper is
currently ~6ns, i.e. the inverse of the 170mhz clock speed of the chip. (On
my todo list is to create another revision of my board with the higher-end
STM32G4 chip that will get the timestamper resolution down to 184ps.) I
plotted the error in the timestamps of the PPS signal with time, i.e. the
x-axis is the time the experiment has been running in seconds; the y-axis
is the difference between the actual timestamp and what the timestamp
"should have been" if the timestamps were actually received exactly 1
second apart. Ideally it would be a flat line indicating no frequency
difference.

The better of the two units showed a frequency error (t=10000s) of about
4e-11, which (as I understand it) is typical performance for an RbXO:

https://www.circlemud.org/jelson/time-graphs/2021-06-16-gpspps-rubidium-unit2-test2-after-warmup.txt.time.plot.png

The other unit was about 20x worse, about 1e-9:

https://www.circlemud.org/jelson/time-graphs/2021-06-16-gpspps-rubidium-unit1-test2.txt.time.plot.png

I'm very pleased with the performance of my timestamper, which seems to be
working perfectly, and I look forward to getting it "cooked" enough to be
able to share with all of you.

I'm less pleased with the performance of one of the frequency standards,
which leads me to my question: has anyone calibrated one of these things
before? The manual I found online says I should be able to turn a
calibration screw on the cover, but the units I have don't seem to have any
exposed screws!

Regards,
-Jeremy

PS: If needed, the raw data behind the two graphs below is here:

https://www.circlemud.org/jelson/time-graphs/2021-06-16-gpspps-rubidium-unit1-test2.txt

https://www.circlemud.org/jelson/time-graphs/2021-06-16-gpspps-rubidium-unit2-test2-after-warmup.txt