How accurate are GPS 1PPS signals

[VK1LW]

A question for people who may have had equipment within a couple of hops of a calibration lab. 

How accurate (true) are the time and 1PPS signals from GPS/GNSS receivers?  Once antenna delay is compensated for, and clock jitter removed, how far are they generally biased from correct system time?  (Repeatability I can measure myself.)

If it answer is, 'it depends', then insight into:

 - Regular GNSS Receivers
 - RTK-capable Receivers
 - Proper Geodetic Receivers

 ... would be helpful.

Yes, I probably need to pay a calibration lab for a fully correct answer.  I'd like to understand the ballpark I am in first.

Thanks

[Berni]

The GPS receivers typically have a spec in the datasheet for the accuracy and jitter of the pps signal. Usually it is in nanoseconds.

Tho there are two kinds of accuracy to talk about here. One is how close to 1Hz that signal is and another is how accurately is the phase of it aligned to the global time. The 1Hz should be incredibly accurate in the long run as it is basically synced to the GPS atomic clocks, but short term it might have a fair bit of jitter cycle to cycle (depending on the internal architecture of the GPS receiver).

Easiest way to check is to take two GPS receivers that are built on different chipsets and run both of their PPS outputs into a oscilloscope.

[iMo]

For example short term around 30ns rms for NEO-6 to -9 under good receiving conditions.
You have to discipline a good OCXO with it to get much better results..

[Hydron]

Some receivers send a correction to account for quantisation of the PPS by the internal clock. But short term GNSS will be a bit jittery even with corrections, long term it will be incredibly accurate. This is where a GPSDO comes in, you can discipline a (short term stable) oscillator to the long term stable GNSS signal, the trick is getting the filter right.

[VK1LW]

Thanks.  I'm trying to convince myself of the 'trueness' as opposed to repeatability.  I can deduct the quantisation error to produce a smoother set of numbers, but how do I tell if these closely align to a timescale?

[dietert1]

Recently i tested a dual band GNSS receiver (NEO F10N) with a helical antenna outside yet close to the window - roughly half sky view. Height determinations once a second exhibited a rms of about 1 m, which corresponds to 3 nsec. Of course, in order to get there, one will need a fairly complex correction scheme.
I just executed a similar exercise synchronizing a local clock (STM32 RTC) using LwIP with SNTP, with a time comparison every 120 seconds. Averaging 30 of those time comparisons i got 0.14 msec RMS during a week. This is with the RTC clock inside a thermal chamber. The systematic error yet unknown might amount to some msec.
In the case of GPS there will be systematic errors, too: antenna cable and other communication delays, maybe 50 nsec.
I hope to do the triple comparison soon: RTC, SNTP and GPS. And the GPS antenna should move above the roof to get the full sky view.

Precision time isn't something easy. For example there is a difference of 18 seconds between our local time as reported by the SNTP and GPS as GPS ignores leap seconds. Don't know yet what happens when using GNSS. Looks like Galileo implements UTC and takes leap seconds into account.

If the OP question isn't about time but about frequency: Assuming a jitter of 3 nsec, in order to reach a nominal 10E-13 accuracy one needs about one day of second pulses. After an hour one reaches about 10E-12, good enough to calibrate a Rubidium oscillator. The NEO F10N receiver has a configurable reference frequency output and a TXCO. Doing the comparison at 10 MHz instead of 1 Hz should be somewhat faster.

Regards, Dieter

[Berni]

Thanks.  I'm trying to convince myself of the 'trueness' as opposed to repeatability.  I can deduct the quantisation error to produce a smoother set of numbers, but how do I tell if these closely align to a timescale?

If the aim is to generate a pulse that is really really really exactly 1 second long then you wouldn't directly use the PPS output anyway.

The GPS receiver is constantly hunting for a lock with the satellite signals so on a short time scale it will jitter around (made worse by the fact that the GPS receiver does the computations on its own internal clock). The receiver can't directly watch the 1.5GHz RF carrier coming from the satellite to simply lock onto that (the signal is way too weak), it can just find the correlation peak for the PRBS code clocked by a divided version of that carrier. But that's more than good enough for positioning use.

The thing that a PPS signal is useful for is to watch it long term and see if your local oscillator (be it TCXO, OCXO, rubidium etc..) phase drifts in relation to it. You can use a digital scope to record the timing difference and plot it. If the plot doesn't exhibit any long term upwards or downwards trend then your local oscillator is pretty much perfect. Then to get your perfect 1Hz signal divide down your local oscillator to that.

If you are calibrating something like a hydrogen maser then you will have to be pretty patient in waiting long enough for a phase error to show up. But for any sort of oscillator a mere mortal might have at home waiting around a few minutes for a clearly visible phase change is already really really accurate.

[dietert1]

The NEO F10N receiver i mentioned provides carrier phase tracking and i seem to remember that its reference frequency output is derived from there. Seems they provide the phase measurement for each of the satellites once there is a position lock.

Regards, Dieter