We are trying to check if a HP GPSDO is still functioning correctly. To do this we used a tinyPFA to measure the GPSDO against a known good Rb during about 7 days. The ADEV looks good,  4e-11 at tau=0.1 s going down to 4e-12 at tau=10 s, staying flat at 4e-12 till tau = 400 s and after a small bump going down from 6e-12 @ tau=3000 s to 8e-13 at tau=20000 s However, above 20000 s the ADEV is going up again. Looking at the linear residue of the phase it showed indeed the phase was not (yet) in linear drifting mode. Maybe warming up time was insufficient, not a big problem. Our biggest worry is visible in the frequency plot (see attachment) which shows periodic frequency changes sometimes appearing for some days with a period of about 2 hours or about 30 minutes These changes are in the order of 4e-11. Our question is: In what direction should be look for the cause of these periodic disturbances? - Constellation changes? No effort has been done yet to select only one constellation - Sub optimal antenna placement? The antenna has no full 360 degrees clear sky view as there are some trees in the way and there has been a lot of mist/rain - Fundamental limitation of the measurement and we should not bother? - Is the Rb approaching end of life and starting to have locking problems? Any hint is most welcome Erik.

-------- Erik Kaashoek via time-nuts writes: > Our biggest worry is visible in the frequency plot (see attachment) > which shows periodic frequency changes sometimes appearing for some days > with a period of about 2 hours or about 30 minutes > These changes are in the order of 4e-11. Oscillations at those rates are almost always thermal. If they are sort of sinosodal, it's probably the ventilation or air-con, cycling on and off. The rule of thumb on cooling compressors is that they should cycle three times an hour in a well tuned system. If the ecursions are irregular and spiky, look for doors opening and closing, or equipment in the room turning on or off, or doing more or less work. Computers, both server and desktop have very spiky power consumption and thus heat generation. Start by covering the entire setup with cardboard box or a sheet of cloth, but make sure not to insulate the setup so much that it cannot get rid of the heat. The idea is to prevent air-currents in the room from getting directly to the setup you are measuring. And if that changes the picture, you are on the right track... -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 phk@FreeBSD.ORG | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence.

Hi, On 2025-01-17 23:06, Poul-Henning Kamp via time-nuts wrote: > -------- > Erik Kaashoek via time-nuts writes: > >> Our biggest worry is visible in the frequency plot (see attachment) >> which shows periodic frequency changes sometimes appearing for some days >> with a period of about 2 hours or about 30 minutes >> These changes are in the order of 4e-11. > Oscillations at those rates are almost always thermal. Indeed. This is completely normal and expected. > If they are sort of sinosodal, it's probably the ventilation or air-con, > cycling on and off. The rule of thumb on cooling compressors is > that they should cycle three times an hour in a well tuned system. The bang-bang regulation is typical and works well enough for normal operation. It's just not very nice to T&F stuff except avoiding overheating. > If the ecursions are irregular and spiky, look for doors opening > and closing, or equipment in the room turning on or off, or doing > more or less work. Computers, both server and desktop have very > spiky power consumption and thus heat generation. > > Start by covering the entire setup with cardboard box or a sheet > of cloth, but make sure not to insulate the setup so much that it > cannot get rid of the heat. The idea is to prevent air-currents > in the room from getting directly to the setup you are measuring. It may seem silly, but just a windshield stabilize the thermal system a lot, while not making much of an insulation long-term. However, the short-term variations can hit you both as single event and as repeating events. In an extreme form, a quick change makes regulation goes out of linear region, thus making it uncontrolled. Also, for crystals, temperature gradients is worse than temperature errors. Just shielding from wind makes temperature gradients even out through the thermal conduction, and have time to do that even as temperature changes. Also, control loops gets a fair chance to compensate changes. That said, yes we do not want to insulate too much. > And if that changes the picture, you are on the right track... > Indeed. It takes very litte cost and effort to test the theory. I remember we had a couple of students doing extra summer work at the company. They where doing measurements and found that the overnight measure had an unexpected deviation. What they showed me was a slow variation that suddenly stopped. I asked them when they started the measurement, could it be around 15:00? Sure, that sounded about right, but how would I know that? Well, three hours in, their oscillation stops, and that's when the AC in the building shuts off. They where flabbergasted about my conclusion, but I then explained to them that we have so power hungry boards that shift air from side-to-side in the rack unit, then I pulled out the switch board and showed them how the oscillator effectively sits in that airflow. I then explained how just a little windshield would help it, and then brought out the foam tape and instructed them to just make a small shield. Next measurement was much flatter. We have even seen how turbulence cause the oscillators to be unstable, when the PLL bandwidth did not allow for it. Again foam tape to the rescure. Putting small plastic caps over oscillators can really help creating a calmer local climate. There is no point in testing oscillators unless dunked into a small box with some bubble-wrap. However, there is one reason to have oscillators exposed to temperature changes, the larger they are, while still reasonable, the easier it is to measure their temperature dependence with good quality. As you lover the variations, the sensitivity of temperature changes makes it harder to characterize. However, this may or may not be a concern you need to care about on daily basis. Cheers, Magnus