More time server goofing
After dicking around with the time server last week, I noticed that AliExpress were having a “sale” and they had “marine style” GPS antennae for about $12 shipped, which sounded good enough to gamble on. The package arrived today, so after work I set about figuring out how to get it on the roof.
I’d originally intended to put it on the post that old FoxTel dish used to live on (the dish removed because it cast a shadow on the solar panels at certain times of the day), however doing so would likely necessitate some sort of lightning protection. Instead, by tucking it under the TV antenna next to the evaporative tower for our air conditioning, I can negate the risk of that significantly.
That’s farther than 5 meters though, and I don’t know how good the “active” part of this $12 GPS antenna is to push the signal adequately down an extension cable, so I resolved to move the GPS into the house rack instead of the garage, run the cable through the penetration into the ceiling void and up to the roof. A cursory test shows that signal is significantly improved with it just laying on the roof. Why, I don’t think I once saw less than 7 satellites that met the filters, a far cry over the “three or four but sometimes two” (and definitely no hope of SouthPAN reception whatsoever) of the puck on a west-facing window.
The antenna mounts by what’s allegedly a 1"-14 TPI thread, so I started trying to find something to mount it with. Absolutely nothing in Australia that I can find, and the only thing I could find mail ordering was small plate-mount things that I’d have to screw into my roof. No thanks. Jaycar had bullbar mounts for a UHF antenna, but even that wasn’t really right as I’d still need some sort of bolt in the right thread, and I’d need a way to enlarge the hole in it.
Off to Hammer Barn (twice, as the first time I neglected to bring the antenna with me to test fit anything, and guessed wrong on the threads and thus had to return one), and it turns out 20mm x 3/4" valve take-off adapter (Holman PVVA2020) fits the thread on the bottom of the antenna just well enough. You’d not want to swing off it or anything, and it’s certainly not water-tight, but it pulls up tight enough without jumping a thread to keep the antenna in place. A couple of 45 degree elbows and a short length of the appropriate sized pipe, and then some glue to stick it all together and I have a reasonably nice looking little mount. A couple of antenna clamps to squeeze it to the TV mast, snake the cable back in to the rack and we’re away. Ironically, because of the clamps, the mounting solution cost more than the antenna did.
Interestingly though, once I let it sit with the SouthPAN test-mode signal enabled for a while, precision was really good for location, but my time signal with Chrony was absolutely dogshit. Bad enough it threw out my PPS signal as a falseticker and switched to an NTP source instead. Inspecting it, and I learned why: for some reason when the test mode signal is on (even if it’s not got a clear enough signal to get a DGNSS fix), it periodically clobbers the GPS->UTC offset, which is currently 18 seconds, to zero. Whichever NMEA message contains the offset (almanac?) would come in and set it back to 18 seconds, and back and forth they’d go, at roughly 7 minute intervals.
So that means that periodically, the NMEA clock the PPS uses for a reference is out by 18 seconds. No wonder Chrony was throwing it out.
I spent a little while faffing about with u-center trying to figure out what I had configured wrong, before I gave up and just figured the receiver is too old to do it properly. But what I can do is crank the position-hold settings, let it sit in DGNSS mode until the survey-in period is done (I picked an hour at 1.5M, it came up with a standard deviation on the position it picked of 0.4M, which I think is good?). I then turned off test mode in the SBAS configuration, did a warm start, and let it go.
I’ll see how accurate it is over the next week, but hopefully it’s an improvement… at least I’m not out too much money if not.
Update: 2026-01-18: It’s sat pretty rock-solid with a nice flat graph of skew since the installation. CloudFlare’s dances around either side of it but it’s fairly consistently below a microsecond out too. Of course with something like this it’s hard to tell which is correct. I have found the performance of CloudFlare’s time server to be pretty stellar compared to most of the others in the pool, but the point is they’re in agreement a lot more of the time than the previous solution, so I’m happy.
fwagglePublished:
Modified:
Filed under:
Location:
Horsham, VIC, AustraliaNavigation: Older Entry Newer Entry