Raspberry Pi clock trivia

On 04/02/2016 01:48, rickman wrote: []

I've been using typically uBlox units, and they are well within a microsecond, typically within 100 ns, two to three orders of magnitude better than you suggest. You can see the effect of poor GPS antenna location on some of my flock here:

whereas the devices with well situated antennas don't have the loss of sync. By the way, these RPi cards all have indoor antennas.

As you say, both GPS and WWVB have reception issues. The RPi makes a nice NTP server for quite a large network, whatever the source.

On 04/02/2016 00:59, Dennis Lee Bieber wrote: []

.. and if the receiver location is fixed, and has been determined by the GPS, even fewer satellites are needed just for timekeeping. Some GPS receivers have this operational mode - initial location determination, followed by optimal timekeeping reception.

These are the old units. In the old days, a GPS receiver could really only receive a single satellite at a time, even when it had 6, 8 or 12 channels. Those were multiplexed, in reality.

So to get the best time fix using those old GPSDO boxes, they first do a "survey", i.e. they determine their position by averaging 200-2000 fixes. Then they store that (some in RAM, some in EEPROM) and switch to "single satellite mode" where they lock only their clock to a single satellite, using the known distance from that satellite to the fixed location.

The advantage is that they can concentrate on receiving a single satellite, with better performance in those old receivers. The disadvantage is when one satellite provides wrong time (as happened last week), they go completely haywire. A receiver that always receives 8-12 satellites in parallel can easily determine that one satellite is faulty (because its reported time does not agree with the time computed from the other satellites and the location), but these old boxes cannot, and they will change their local clock to agree with the faulty reference for the time period that satellite is selected.

I have both a GPS and a DCF77 (similar to WWVB) receiver connected to the same system and in the past I have made plots of their relative time, where the day/night propagation differences could easily be seen. The jitter on the DCF77 signal during night was about 200us. (the receiver in use only demodulates the ASK, not PSK)

Not only "old units" - even some of the latest u-blox receivers offer survey/fixed operation, for example:

See para: 1.10.6. Incidentally, its specification for accuracy is:

Time pulse accuracy

Clear sky ? 20 ns

Indoor ? 500 ns

Actually survey/fixed is not the same as single-satellite operation. I see this unit allows fixed mode with multiple satellites, or with a single satellite. In the old unites, that mode was preferable due to receiver limitations, I expect that in these new receivers it is only present for backward compatability, e.g. to operate in an environment where single-satellite (1SV) operation has been defined.

Last week's incident (and it has happened before) showed that it is a bit dangerous to do so.

Don't believe the specs on units will be met all the time or even any of the time. I was working for a military radio company and we evaluated uBlox and Fastrax units for a new product. I believe it was the Fastrax unit that would not meet the spec for time to first fix. This was not a constellation problem as it happened repeatedly over a couple of days and the rest of the setup was the same as was used with the uBlox device. I see now they are the same company. lol

My point is that the specs are under rather ideal conditions and are not always met. There is also the issue of just what the spec is saying. Consider how they generate the PPS and explain to me where the above 20 ns number comes from. I'd be willing to be it is specifying the jitter in the timing pulse due to being clocked by a 50 MHz oscillator. If that is true, every other source of error would have to be zero.

But then it appears my recollection is a bit out of date. Our efforts were done nearly 10 years ago and I see GPS receivers have improved a lot. The location accuracy is speced at 2 m CEP which is many times better than the units we evaluated. Location and timing accuracy go hand in hand.

If you'd have been the OP there would have been mention of your home brewed 6809 system and how ICL mainframes did their timekeeping.

On Thu, 4 Feb 2016 11:23:48 -0500, rickman declaimed the following:

Besides the discontinuation of Selective Availability, North America (or, at least, the US) now has the Wide-Area Augmentation System (Europe equivalent is EGNOS, as I recall)... WAAS is a pair of non-Navstar satellites that transmit a set of "corrections" based on the Navstar position errors computed at a number of fixed ground stations... Sort of a differential GPS for the continent. (Differential GPS makes use of "local" ground transmitters which broadcast on a different frequency -- requiring an adjunct receiver -- giving the local correction between their surveyed position and a GPS fix. As I recall, most DGPS transmitters are located near coastal ports, allowing more precise shipping).

The newest receivers not only use Navstar, but GLONASS birds.

Some of the top end receivers may also make use of both GPS frequencies even if they can't decrypt the precise position signal -- since matching the second frequency chips to the first allows for some estimation of ionospheric effects. (I haven't kept up on this, but recall some hearsay that the Navstar will be adding two more frequencies giving military GPS four frequencies for ionospheric corrections and reducing the odds of being jammed, and was going to open the current second frequency for civil signals).

I had your fine work partly in mind. :-)

It seems a shame to have such high quality time information in the air around us and not use it.

Particularly when it no longer costs an arm and a leg! Educational, too.