3 ms·

~1ms drift for a GPS clock sounds like a lot to me. Have I missed something?

by majke 13y ago

ajross 13y ago
It's graphing "offset", not "drift". The actual fluctuations are more like 0.1ms it seems. The DC bias is presumably inherent in the measurement mechanism (asymmetric IP delays, etc...). Best guess, anyway. Honestly the data looks surprisingly uninteresting. There isn't much there in the way of notable observation or correlation with other effects...
- majke 13y ago
  Even 0.1ms drift sounds like an awful lot of time for a serious clock. [edit] if 1ns is 11.8 inches of wire (at the speed of light) then 0.1 ms is 30 km of wire! [1] http://highscalability.com/blog/2012/3/1/grace-hopper-to-programmers-mind-your-nanoseconds.html http://highscalability.com/blog/2012/3/1/grace-hopper-to-pro...
  - jrockway 13y ago
    GPS is pretty accurate, but remember that the time calculations depend on ionospheric conditions, which are random. (The two biggest sources of GPS error are propagation changes and delays introduced by the signal processing code. Both are very small, but not 0)
  - ajross 13y ago
    Yes, but again this is measuring via NTP. A 0.1ms offset in a GPS signal would be a failure state, but the same injected by a gigabit switch (for example) would be virtually undetectable and persist in the market. Having great underlying hardware doesn't make the problem NTP is trying to solve any easier.
  - VLM 13y ago
    300 KM is not too unbelievable WRT multipath distortion thru the ionosphere. You can game it a bit by averaging or building a clock that "knows" it cannot move or that its oscillator is more stable than ionospheric scintillation. Fundamentally the whole system works by having a constellation of satellites with precision clocks squirting out the "exact" time and where they think they are WRT each other. So you're 0.01 light seconds from satellite 3 and 0.005 light seconds from satellite 4 blah blah repeat at least 4 times and draw some circles of a certain light second diameter and there you are. But its going to be inherently sensitive to anomalous propagation, the kind of thing ham radio operators actually like but drives GPS nuts. It would work a lot better if radio waves traveled at a constant speed, but if your system design demands nine decimal places of constant speed and the world won't deliver, well..
    - mikeash 13y ago
      Typical real-world GPS accuracy is on the order of 10m, so that means that the error in the time differences cannot be more than about 30ns typically. To see anything close to a 0.1ms/300km offset, you'd need every signal to be shifted by almost the same amount. To maintain 10m position accuracy while suffering from 300km time error, all of the signals would have to be delayed to within 0.003% of each other. That's an extremely tall order for a natural phenomenon, considering that the satellites are positioned such that signals can come from any direction in the sky. It's vastly more plausible that this is simply a result of troublesome hardware after the core GPS clock hardware gets done doing its thing.

~1ms drift for a GPS clock sounds like a lot to me. Have I missed something?

by majke 13y ago

~1ms drift for a GPS clock sounds like a lot to me. Have I missed something?

It's graphing "offset", not "drift". The actual fluctuations are more like 0.1ms it seems. The DC bias is presumably inherent in the measurement mechanism (asymmetric IP delays, etc...). Best guess, anyway. Honestly the data looks surprisingly uninteresting. There isn't much there in the way of notable observation or correlation with other effects...

majke 13y ago
Even 0.1ms drift sounds like an awful lot of time for a serious clock. [edit] if 1ns is 11.8 inches of wire (at the speed of light) then 0.1 ms is 30 km of wire! [1] http://highscalability.com/blog/2012/3/1/grace-hopper-to-programmers-mind-your-nanoseconds.html http://highscalability.com/blog/2012/3/1/grace-hopper-to-pro...
- jrockway 13y ago
  GPS is pretty accurate, but remember that the time calculations depend on ionospheric conditions, which are random. (The two biggest sources of GPS error are propagation changes and delays introduced by the signal processing code. Both are very small, but not 0)
- ajross 13y ago
  Yes, but again this is measuring via NTP. A 0.1ms offset in a GPS signal would be a failure state, but the same injected by a gigabit switch (for example) would be virtually undetectable and persist in the market. Having great underlying hardware doesn't make the problem NTP is trying to solve any easier.
- VLM 13y ago
  300 KM is not too unbelievable WRT multipath distortion thru the ionosphere. You can game it a bit by averaging or building a clock that "knows" it cannot move or that its oscillator is more stable than ionospheric scintillation. Fundamentally the whole system works by having a constellation of satellites with precision clocks squirting out the "exact" time and where they think they are WRT each other. So you're 0.01 light seconds from satellite 3 and 0.005 light seconds from satellite 4 blah blah repeat at least 4 times and draw some circles of a certain light second diameter and there you are. But its going to be inherently sensitive to anomalous propagation, the kind of thing ham radio operators actually like but drives GPS nuts. It would work a lot better if radio waves traveled at a constant speed, but if your system design demands nine decimal places of constant speed and the world won't deliver, well..
  - mikeash 13y ago
    Typical real-world GPS accuracy is on the order of 10m, so that means that the error in the time differences cannot be more than about 30ns typically. To see anything close to a 0.1ms/300km offset, you'd need every signal to be shifted by almost the same amount. To maintain 10m position accuracy while suffering from 300km time error, all of the signals would have to be delayed to within 0.003% of each other. That's an extremely tall order for a natural phenomenon, considering that the satellites are positioned such that signals can come from any direction in the sky. It's vastly more plausible that this is simply a result of troublesome hardware after the core GPS clock hardware gets done doing its thing.