GP-E1 Time Accuracy

[koolkurkle]

     I just measured the time accuracy of my newly purchased GP-E1.  The advertised accuracy as shown on the Canon USA website is +/- .01 sec.  I installed an NTP time app on my iPhone (Emerald Time), which claims a corrected time to the nearest tenth of a second.  (I verified Emerald Time's accuracy by calling the NIST time recording at 303-499-7111.)  Below is a photo of Emerald Time and the associated exif time data. The GP-E1 was more than a second off.  I force-updated the camera's clock off the GPS multiple times and the results were consistent and repeatable.

[jrista]

Where does the GP-E1 get its time from?

[viggen61]

Where does the GP-E1 get its time from?

Like all GPS devices, it gets it from the atomic clocks that are on the GPS satellites.

One second off? Doesn't sound like a big deal to me.

[jrista]

Where does the GP-E1 get its time from?

Like all GPS devices, it gets it from the atomic clocks that are on the GPS satellites.

One second off? Doesn't sound like a big deal to me.

I just wanted to verify that was the case. There is a degree of latency with any satellite communication. Could that latency be the issue? Even at the speed of light, there is at least a fraction of a second difference in time between when the satellite sends its response and the GPS receiver gets it.

[PeterJ]

I just wanted to verify that was the case. There is a degree of latency with any satellite communication. Could that latency be the issue? Even at the speed of light, there is at least a fraction of a second difference in time between when the satellite sends its response and the GPS receiver gets it.

In short, no. The GPS system maintains ephemeris data for each satellite and the delays through space and the atmosphere are accounted for as part of the model. There are errors for sure, such as delays caused by different ionspheric conditions, but internally any receiver will typically be accurate to well under a microsecond. Light travels about 300m in that distance, so you can imagine the accuracy if it was worse.

More likely is something funky in Canon's implementation, maybe it does something like get the time when constructing the EXIF data after the image is taken rather than when the shutter opens. Or maybe it gets the time over the serial link to the GPS and doesn't take the latency into account. Or it might be a straight bug, or some combination of all the above. To get accurate time from a GPS you normally need to use a seperate hardware PPS (pulse per second) line rather than read the serial data alone, maybe they don't do that.

Actually it's useful to know for future reference, a few times in the past I've taken a photo of a time reference and moving objects to verify the accuracy of GPS systems. While not an issue for most users it means it wouldn't be useful for that, at least not without determining the time offset in advance if it is 100% repetable.

[jrista]

I just wanted to verify that was the case. There is a degree of latency with any satellite communication. Could that latency be the issue? Even at the speed of light, there is at least a fraction of a second difference in time between when the satellite sends its response and the GPS receiver gets it.

In short, no. The GPS system maintains ephemeris data for each satellite and the delays through space and the atmosphere are accounted for as part of the model. There are errors for sure, such as delays caused by different ionspheric conditions, but internally any receiver will typically be accurate to well under a microsecond. Light travels about 300m in that distance, so you can imagine the accuracy if it was worse.

More likely is something funky in Canon's implementation, maybe it does something like get the time when constructing the EXIF data after the image is taken rather than when the shutter opens. Or maybe it gets the time over the serial link to the GPS and doesn't take the latency into account. Or it might be a straight bug, or some combination of all the above. To get accurate time from a GPS you normally need to use a seperate hardware PPS (pulse per second) line rather than read the serial data alone, maybe they don't do that.

Actually it's useful to know for future reference, a few times in the past I've taken a photo of a time reference and moving objects to verify the accuracy of GPS systems. While not an issue for most users it means it wouldn't be useful for that, at least not without determining the time offset in advance if it is 100% repetable.

Ah, well if GPS maintains satellite ephemeris data and accounts for delays, it should be pretty darn accurate.

[koolkurkle]

I just wanted to verify that was the case. There is a degree of latency with any satellite communication. Could that latency be the issue? Even at the speed of light, there is at least a fraction of a second difference in time between when the satellite sends its response and the GPS receiver gets it.

In short, no. The GPS system maintains ephemeris data for each satellite and the delays through space and the atmosphere are accounted for as part of the model. There are errors for sure, such as delays caused by different ionspheric conditions, but internally any receiver will typically be accurate to well under a microsecond. Light travels about 300m in that distance, so you can imagine the accuracy if it was worse.

More likely is something funky in Canon's implementation, maybe it does something like get the time when constructing the EXIF data after the image is taken rather than when the shutter opens. Or maybe it gets the time over the serial link to the GPS and doesn't take the latency into account. Or it might be a straight bug, or some combination of all the above. To get accurate time from a GPS you normally need to use a seperate hardware PPS (pulse per second) line rather than read the serial data alone, maybe they don't do that.

Actually it's useful to know for future reference, a few times in the past I've taken a photo of a time reference and moving objects to verify the accuracy of GPS systems. While not an issue for most users it means it wouldn't be useful for that, at least not without determining the time offset in advance if it is 100% repetable.

I think PeterJ nailed it.  I sent it back for "repair", and I guess we'll see what they say.  It probably is fixable in firmware, but the number of units sold may not justify the investment in correcting an issue most people probably wouldn't even notice.

[koolkurkle]

Almost a month to the day of receiving my GP-E1 for repair, a Canon supervisor called me today.  It took a while, but the personal service from Canon was really outstanding.  There is a known (but apparently not yet publicized) bug with the GP-E1/E2 and the leap second (http://maia.usno.navy.mil/ser7/series14.txt) that was implemented by IERS June 2012.  This is the cause of the 1 second discrepancy I was experiencing.  The current temporary fix involves 1) Disabling auto power down, 2) Selecting a 1 second update interval and 3) Waiting 15 MINUTES before force updating the camera time to GPS time.  This doesn't sound nearly as bad as it could have been and a firmware update seems likely. 

[Drizzt321]

Almost a month to the day of receiving my GP-E1 for repair, a Canon supervisor called me today.  It took a while, but the personal service from Canon was really outstanding.  There is a known (but apparently not yet publicized) bug with the GP-E1/E2 and the leap second (http://maia.usno.navy.mil/ser7/series14.txt) that was implemented by IERS June 2012.  This is the cause of the 1 second discrepancy I was experiencing.  The current temporary fix involves 1) Disabling auto power down, 2) Selecting a 1 second update interval and 3) Waiting 15 MINUTES before force updating the camera time to GPS time.  This doesn't sound nearly as bad as it could have been and a firmware update seems likely.

Wow, crazy. Maybe they should have publicized this, rather than waiting until people dig into it like you have been.

[PeterJ]

Good to know. The 15 minutes would be to download a fresh almanac which takes 12.5 minutes. If they didn't mention it make sure you leave it in a good position for those 15 minutes, if it misses a small part of the transmission you have to wait again and sounds like it's not all that apparent if it's downloaded it OK or not.

[mackguyver]

It looks like I found a new winner for most minor issue inconvenience ever sent back for repair.  I understand that some very specific applications might need absolutely perfect atomic time, but to send this back for repair of a clock that is one second off seems to defy all logic.  I guess it's good he found a fix at least.

I was just searching for some info on the GP-E1 which I'm thinking about getting now that I seem to have lost my Garmin GPS for good...

[Don Haines]

Here is how it is supposed to work:

The satellite is transmitting the time signal. If the satellite is in geosynchronous orbit the signal will take about 0.240 seconds to reach you. If your satellite is in a lower orbit the signal will take less time to reach you, approximately 0.100 seconds... If you can only pick up one satellite and you know both where you and the satellite are, you know the distance and delay and can work backwards and get accurate time. If you don't know both locations, you don't know the signal delay and therefore, can not calculate the real time, but you can guess within a quarter second.

Once you can see three satellites, you can calculate the delay from each one and triangulate your position. This gives you an accurate time fix.

Your electronic device has it's own clock. This clock will keep good time, but it will not be exact. Over a period of time this clock will drift away from real time. Lets say that at noon your device's clock was synchronized to GPS time and that 5 hours later, when it was synchronized again, the time had drifted by 10 "ticks" away from the real time. A well implemented timing service would then realize that every half hour that it needs to adjust it's time by 1 "tick" to keep up with the GPS clock. This is known as a "GPS Trained " clock and they can be extremely accurate. I have one in my lab that is accurate to 10^-14 seconds. (0.00000000000001)

Most commercial devices use standard modules that communicate with the device through a series of standard commands. For most GPS modules, these commands are standardized as NMEA, and the time command reports time with a precision of 0.01 seconds. That is as accurate as you should be able to get with your camera. The least accurate devices out there have internal clocks that are accurate to 10^-7, and that's a drift of about .08 seconds per day. Being off by a second is a big deal and indicated that something is wrong.....