[time-nuts] Venus838LPx-T PPS stability measurements
nsayer at kfu.com
Fri Oct 7 10:46:12 EDT 2016
This is a bit overdue, but I finally got around to making at least an attempt to measure the stability of the Venus838LPx-T timing module’s PPS stability.
The results are a bit of a mixed bag.
The module under test is one built into one of my OH300 GPSDOs. It’s inside of a closed chassis, mounted about a half inch away from an OCXO, sitting on my workbench inside an unconditioned garage. My guess is that over the course of the test the ambient temperature varied by perhaps 15°F. The PPS output of the module - as well as being fed into the controller and PLL - is fed into a buffer before being presented on the diagnostic port. From there, it went straight into input 1 of my 53220A. Input 2 came from the PPS output of a Thunderbolt. The 53220A’s 10 MHz reference comes also from that same Thunderbolt. No effort has been made to apply the sawtooth corrections indicated by the PSTI,00 sentences. Both receivers are fed from the same antenna and splitter. Reception is nearly ideal, with the Thunderbolt having 7 or 8 satellites at all times, most of them with S:N > 45.
The first surprise is that although both PPS signals are ostensibly synced with GPS, there’s a 135 ns residual between the two. The residual (after ~15 hours or so) has a ~2E-13 slope. This graph is the phase difference with that residual removed.
There are three variances visible. Firstly, there’s about a ±6ns “fuzz” around the center on almost all samples. That can be explained by the quantization error indicated in the NMEA output. Plotting those errors shows a similar fuzz with a range of ±6 ns. Secondly, there’s a much slower wander that’s mostly confined to a ±10 ns corridor. I attribute this to GPS itself. How much of the wandering is due to which receiver is something I haven’t attempted to figure out. Running this test with an undisciplined rubidium oscillator might help, but the short term stability of the 5680A isn’t very good, so I didn’t want to make it the first test standard.
The third variance is more serious. Periodically the variance is much larger - sometimes ±25ns or even more. These variances are not accounted for in the sawtooth correction values. The only good thing that can be said about them is that they’re fairly well balanced and can be easily averaged out.
If you take a closer look at one, they sometimes appear adjacent to hanging bridges:
This isn’t always the case, but it’s often enough to potentially be more than coincidence.
All that said, I believe the resulting ADEV is in line with expectations for a GPS receiver:
I’ve made an inquiry with SkyTraq to ask about the excessive excursions. I’ll report back what their answer is. Given that the excursions are easily averaged away, and given how inexpensive these modules are, I’m not bent out of shape about it. And if it’s something SkyTraq can figure out how to update in the firmware to report in their QE messages (so that it can be corrected for externally), I won’t mind at all.
More information about the time-nuts