No spikes, no errors. Garmin over wetsuit, on wrist, no duct tape, band tight.

Select to expand quote

vosadrian said..
If some software processing GPSTC data are getting different results for 5X10 speeds with this issue, it suggests that positional data is being used.

The difference could be related to what I call "run detection", determining when one run starts / ends and another begins.

Whilst the Doppler-derived speeds are used for results, positional changes and derived heading may be used for run detection.

Just a quick update.

Scott added an additional timed event to APPro that repeats every 1000 ms. The time of the event is logged to the FIT file and allows us to see whether it runs without experiencing any drift. I've centred the data in all of these charts around 500 for the sake of clarity, and the timed events on both the FR 255 and F7 Pro ran without drifting. It is my expectation that the FIT logging and perhaps some other events in the watches are running every 1000 ms in much the same manner, without drifting. There might however be the odd hiccup when the watch syncs its clock with the time from GNSS solutions.

The location events continue to show a drift on the FR 255 and in this instance, drifted by approximately 250 ms during this test. The F7 Pro did not exhibit any drift for location events, and neither did the FR 265 in the one test session that I have seen. I'm pretty sure it's this drift that causes the FR 255 to start logging a lot of repeated GNSS values, where slight variations in timing will be enough to cause repeated values. Just as a reminder, if fields aren't calculated before the next write to the FIT file then the previous values are written.

I've also spent some time looking at the relationship of GPS metadata and the main records in FIT files, but haven't had time to document how those can be useful when the timing goes awry. We can definitely use them though, and with some additional data logged to the main FIT records it should be possible to interpolate in the correct places, without any complex logic. I just don't have the bandwidth to document it right now. I'm still keen to identify which watches are affected so would like to hear from anyone with a FR 955, and also a NON-PRO version of the Fenix 7 or EPIX (gen 2).

I combined the two FIT files from that day when I moved the watch and the signal degraded. That's what convinced me that the wet neoprene next to the watch was the problem. I saved the file, moved the Garmin on the water in about a minute and restarted logging. The error rate went from one per two per run to dozens. But this also fits with the thinking that there could be some sort of timing bug (that I don't understand) being introduced when logging starts.

Is it possible that the watch is trying to connect to your phone to save the previous file. It may even connect if your phone is not too far away. Then perhaps it gets caught up with some partially complete phone syncing operation.

^ now that is an interesting thought. When I saved the session and restarted a new one, I would have been about 70 metres away from the car. On two occasions the watch randomly started "find my phone" while I was sailing.

Select to expand quote

tbwonder said..
Is it possible that the watch is trying to connect to your phone to save the previous file. It may even connect if your phone is not too far away. Then perhaps it gets caught up with some partially complete phone syncing operation.

Good thought Andrew, I think we'll investigate that possibility further.

That got me wondering if it had something to do with starting logging while some distance from the phone. But this session starts next to the car and is consistently terrible for the entire two hours. www.ka72.com/Track/t/537819

Select to expand quote

remery said..
On two occasions the watch randomly started "find my phone" while I was sailing.

I've disabled that feature on my Garmin watches. It can be pretty annoying when close to the shore.

Regarding the timing, I believe it can sometimes be amiss as soon as a session starts (e.g. badly affecting the second session for remery) or starting in the middle of a session, due to the drift. I have sessions which started fine, then contained lots of repeated values for half an hour or so (much like the remery example) and then returned to normal.

The change mid-session is explainable by the the drift that has been diagnosed in the FR 255. The session starts fine, drifts into problem territory and then settles down again, some time later. The lag for subsequent logging also changes by 1 second (either forwards, or backwards), and I've seen this several times on my FR 255. I believe Peter has observed it as well.

I don't recall seeing the quality of a session change mid-session. I reckon mine have been excellent, bad and terrible all the way through.

Select to expand quote

remery said..
I don't recall seeing the quality of a session change mid-session. I reckon mine have been excellent, bad and terrible all the way through.

Perhaps just a matter of time. I've racked up more than 100 genuine sessions on my FR 255, which is excluding test drives, etc.

A couple of additional insights.

Firstly some FR 965 data from vosadrian, showing very similar timing behavior to my F7 Pro and the FR 265. Generally stable, but a couple of glitches exactly an hour apart. When the timing changes significantly, we see repeated values, but I won't post a graph of speeds.


Secondly, data from an Edge 530 which I immediately recognised as the Sony chipset. Motion in blue, Edge in red. The Sony is really bad, but that is not news to anyone who has studied its data.
Speed data from the Sony is bad, but positional data is even worse. This is evident in the position-derived speeds (dotted blue).


What is most relevant is that a period of frequently repeating values appears during the session and then disappears for the end of the session.


The data from the Edge 530 therefore confirms that frequently repeating values also occur on some of the Sony based Garmin devices. It would appear that some devices have implemented their timings / events in a reliable manner, and others have not.

Thanks for that Mike.Looks like the timing on the 965 is on the better side for this timing issues with minimal repeated positions. We will wait for more info on whether the speed is comparable to the 255/Motion. I hope to get a sail in this weekend and will pass the files when I have them.

Regarding sailing, for the purpose of this test is there a recommended way to wear the devices for best comparison? I normally just wear my watch as normal (on top side of arm as a normal watch) and I presume the Motion should be worn strapped to the same arm?

Select to expand quote

vosadrian said..>>I normally just wear my watch as normal (on top side of arm as a normal watch) and I presume the Motion should be worn strapped to the same arm?

Yes please, the closer together the better.

I have DIY-GPS data from one side of the boom for all my sessions comparing the 255 and motion of that's a help.

Consistent numbers and no big spikes today.

I had a sail at Kyeemah (Botany Bay) with the borrowed Motion and my FR965 today. My first GPSTC session in years since I had an approved device. I will send the files through for checking in a moment.

KA72 results for Motion:

2 Second Peak (kts):33.271
5x10 Average (kts):31.027
Top 5 5x10 speeds: (1)32.297
Top 5 5x10 speeds: (2)31.497
Top 5 5x10 speeds: (3)30.789
Top 5 5x10 speeds: (4)30.444
Top 5 5x10 speeds: (5)30.108
1 Hr (kts):12.663
Alpha 500 (kts):19.575
Nautical Mile (kts):23.057
100m peak (kts):32.802
Total Distance (km):41.668

KA72 results for FR965:

2 Second Peak (kts):33.192
5x10 Average (kts):30.945
Top 5 5x10 speeds: (1)32.214
Top 5 5x10 speeds: (2)31.401
Top 5 5x10 speeds: (3)30.801
Top 5 5x10 speeds: (4)30.324
Top 5 5x10 speeds: (5)29.985
1 Hr (kts):12.676
Alpha 500 (kts):19.903
Nautical Mile (kts):23.017
100m peak (kts):32.578
Total Distance (km):41.578

Looks pretty good to me!

Here's a comparison of Adrian's top 5 results using GPS Speedreader.

The most notable differences are a couple of 2s results - one is 0.48 kts too low and one is 0.42 kts too high. That is normal / common for the modern Garmin watches that I have studied.

Differences in 2s results often aren't apparent on KA-72 because that's often a comparison of two different runs from the two devices. Comparisons using GPS Speedreader also requires you to take care and match up the times of the runs.

I've seen differences of up to 1.5 knots in 2s results (e.g. one Garmin 0.7 too low, and one 0.8 too high) which wouldn't have shown up when you comparing two runs. Something to look out for when reviewing results.

Anyways, 965 seems to have performed well during this session.

Thanks Mike, took the words right pout of my mouth. Needs to be a time sort, not a result sort.

Makes some sense .. but for the purpose of GPSTC postings, the times actually do not matter. One also needs to keep in mind that differences are due to errors from both devices. Even if one device tends to have higher accuracy in general, there will always be stretches where the less accurate device will be closer to the (unknown!) truth than the more accurate device.

Select to expand quote

boardsurfr said..
Makes some sense .. but for the purpose of GPSTC postings, the times actually do not matter. One also needs to keep in mind that differences are due to errors from both devices. Even if one device tends to have higher accuracy in general, there will always be stretches where the less accurate device will be closer to the (unknown!) truth than the more accurate device.

Agreed, both devices have unknown errors but the Motion has greater precision. As way of an illustation, I've previously compared two Motions and two Garmins to see how the reported speeds differ throughout a session. The images below are from a wing session last year. Of course, the Motions had the benefit of being on my biceps, rather than wrists. The differences can be attributed to measurement errors (several possible causes) and arm movements.

Whilst the distributions are not perfectly Gaussian they aren't far off. I can't find the spreadsheet but I guess the most notable thing is how the variance is so much lower on the Motions. It's pretty rare for a rolling 2s to differ by more than 0.2 knots. The variance is much greater on the Garmins, partly due to measurement and partly the wrist mounting.

Neither the Motion or Garmin are error free, but it doesn't make much sense to compare a Motion run at 2pm against a Garmin run at 2.30pm. That is not uncommon when simply looking at the best run, hence my warning of simple one-run comparisons.

When I run the same files I get this:


Something very different with the Alphas?? I dont know why? But that does not look very good to me.

When I run the files sorted by time stamps, which I think is a more valid comparison, it does not improve much at all in the 2 sec runs, but is much better in the Alphas.

I would not call this comparable with what I have seen with the vast majority if the G255M comparisons, where the 2 sec results are usually under 0.1 sec difference, and those rarer ones that go over that figure, are not much over.

I'm not at my computer, but ISTR the best alpha was not present in the Motion data. Both devices recorded similar data but the Motion one was not in the top 5.

Select to expand quote

K888 said..
I'm not at my computer, but ISTR the best alpha was not present in the Motion data. Both devices recorded similar data but the Motion one was not in the top 5.

Ah, yes, you are correct. Speed-reader finds a better Alpha for the Fit file, so it is probably a proximity circle thing.

So here is what I get if I sort the Alphas properly it looks a lot better. this matches closely what you posted, but it's a slight worry that we find a better Alpha result from the fit file that is not found on the OAO file. And the 2 sec is still a bit of an anomaly.

Thanks for looking into this guys!

I will try to get more data, but I don't get to sail as often as some as I am not that close to the water and work a normal working week. I'll do my best.

I have read previously that hte motion provides error data which can be used to determine when the GPS signal may have been poor and cause erronous results. Is it possible to check this error data for where the differences are? Unfortunately the same is not available for the Garmin I believe, but would this be a way to determine which of the devices was likely in error when the differences in speed are larger?

Also, the image below was posted by @decrepit last week from my bike ride last week (with Motion and FR965). The speed is quite high with a relatively constant acceleration and would have been rolling down a hill. This is on a road bike on smooth roads with the Motion strapped to my handlebars (and Garmin on my wrist with hand on the handle bars within about 15cm of the Motion). The point to note is that the Motion doppler (blue solid) has a lot of jitter. I presume it is showing a higher sample rate compared to the Garmin. The Garmin (red solid) is smooth. There is no way that the bike did what the blue line says. The blue line shows decelerations which it simply could not have done (an 80kg bike+rider decelerating for a fraction of a second while having an average acceleration rolling down a hill??). If I saw that while windsurfing, I could believe it due to chop possibly causing the speed to behave like that.... but I don't ride at that speed down a hill on a bumpy road on my road bike. The actual speed of the bike is likely to be very close to the smoother red line.

The reason I post this is that I believe the Motion is posting at 10Hz, and the Garmin at 1Hz. So for a 2S run, the calculations on the Motion would be based on the speed of 20 time periods (averaged) and for the Garmin it would use 2 time periods. If those 20 time periods have a lot of jitter as seen below, it is expected that the results will not be the same. The 2S calculation on the motion could start half way between the 1 second samples of the Garmin, so could in fact be over a different period. The 2S period could include more time when the jitter was high than low... or vice versa. Both devices could be just as accurate, but get a significantly different result if one has a higher sample rate and lower filtering/smoothing.

Select to expand quote

vosadrian said..
Thanks for looking into this guys!

I will try to get more data, but I don't get to sail as often as some as I am not that close to the water and work a normal working week. I'll do my best.

I have read previously that hte motion provides error data which can be used to determine when the GPS signal may have been poor and cause erronous results. Is it possible to check this error data for where the differences are? Unfortunately the same is not available for the Garmin I believe, but would this be a way to determine which of the devices was likely in error when the differences in speed are larger?

No worries. Thanks for making the effort to get the comparison data.

The error data in the Motion file for the 2 second is very normal looking at around 0.12 knots, so it looks like the Motion was getting good reception. The difference in the results is most likely at the extreme end of what we will see, but it would be prudent to see more sessions to confirm this. I guess this highlights that individual samples cant give us the full picture.

For the Alpha, the Motion error data is again very normal, but GPS-Speedreader finds an Alpha from the Garmin file that it doesn't find in the Motion file, where the track misses the proximity circle by around 1 meter or less! This is probably to be expected occasionally with really close calls, just because we must use the less accurate positional data for the proximity, and position can vary a bit more. When we compare the Alphas that match up on the time stamp, they are quite OK. It is inevitable that we will get the occasional discrepancy with Alphas because of this. When this is the case, you will see correlation in the 60m Alpha, which we do in this case, with only a 0.08 difference in result.

Select to expand quote

vosadrian said..
Also, the image below was posted by @decrepit last week from my bike ride last week (with Motion and FR965). The speed is quite high with a relatively constant acceleration and would have been rolling down a hill. This is on a road bike on smooth roads with the Motion strapped to my handlebars (and Garmin on my wrist with hand on the handle bars within about 15cm of the Motion). The point to note is that the Motion doppler (blue solid) has a lot of jitter. I presume it is showing a higher sample rate compared to the Garmin. The Garmin (red solid) is smooth. There is no way that the bike did what the blue line says. The blue line shows decelerations which it simply could not have done (an 80kg bike+rider decelerating for a fraction of a second while having an average acceleration rolling down a hill??). If I saw that while windsurfing, I could believe it due to chop possibly causing the speed to behave like that.... but I don't ride at that speed down a hill on a bumpy road on my road bike. The actual speed of the bike is likely to be very close to the smoother red line.

The reason I post this is that I believe the Motion is posting at 10Hz, and the Garmin at 1Hz. So for a 2S run, the calculations on the Motion would be based on the speed of 20 time periods (averaged) and for the Garmin it would use 2 time periods. If those 20 time periods have a lot of jitter as seen below, it is expected that the results will not be the same. The 2S calculation on the motion could start half way between the 1 second samples of the Garmin, so could in fact be over a different period. The 2S period could include more time when the jitter was high than low... or vice versa. Both devices could be just as accurate, but get a significantly different result if one has a higher sample rate and lower filtering/smoothing.

This is perfectly normal, and what you are seeing in the 'jitter' or 'sawtooth' is predominantly real micro variation in the speed calculations due to micro movements (vibrations) coming through you handlebars. Remember that those speeds variations are very small. In the order of tenths or hundredths of knot. That does not take much movement!! Of course, there is also a smaller amount of random error, but it is predominantly real measurement.

Below is a speed graph of a test I did on my motorbike with two side by side Motions on my helmet. Note that you can clearly see the tiny pauses in speed/accelleration during the split second gear changes. And both GPS's correlate very well. Also note that when I reach higher speed, even on a flat road, there is sawtooth in the speeds from buffeting and vibrations, and those actually correlate very well also.

Yes, the results will not be exactly the same between 10Hz and 1Hz devices, for a lot a reasons. But we have seen very close correlation in the results most of the time, in the range of 0.5 to 0.15 most of the time for 2 secs, and well under 0.1 for 10 sec and 5 x 10 secs.

But both devices are not 'just as accurate'. See K888's post above. The inherent precision of the Motions is considerably higher, but we have deemed the results from the G255's 'close enough' for our purposes. Hopefully we can say the same for the G965 with more study.

Can we respectfully request that we refrain from doing this sort of individual file analysis online in the forum please? It is better done slowly and carefully, with time for team consultation, and when not feeling under pressure to reply immediately.

Please just send the files to info@gpsteamchallenge.com.au and we will do our best to analyse and compile the results among the team thoroughly, and then reply/report on it as appropriate.