Some may have seen the discussion on GPSTC recently where Boro got substantially higher speeds on the GW-60 than on his GT-31s. The conclusion was that this was due to poor GPS reception when he had the watch on the front hand and used an under grip during speed runs. I have been able to reproduce this during a simple test drive with 6 GPS units (GW-60, GW-52, and GT-31).
It has been suggested here many times not to use the watch together with an under grip on speed runs, or to wear the watch to the inside, or (even better) to wear it on the arm or on the helmet. What surprised me somewhat, though, was to see that the error was not random - at times, the speed would deviate in the same direction for 10 seconds or longer. I posted details about the issue at boardsurfr.blogspot.com/2017/05/the-fastest-way-get-faster.html

Interesting post.

I would be interested to see the error-bars for each graph

Select to expand quote

mathew said..
I would be interested to see the error-bars for each graph

Matt I've posted them here.
www.gpsteamchallenge.com.au/sailor_session/show?date=2017-05-11&team=124

Im guessing this is the data for the gps chip used in these www.locosystech.com/en/product/gnss-module-s5-1010-2r.html
Like boardsurfer thought there is a dead reconing mode and although i didnt read the whole thing in detail it apears that it can record what mode it was in at the time ( top of pag 11). Maybe just ask locosys to turn off what they call "Estimated (dead reckoning) Mode" or make it so it just reads the slow end of guessing in that mode so our nms dont get messed up.I looked at the sbp file and have no idea how to enterpret that but im sure some one else knows enough to work that out.

Select to expand quote

Simon100 said..
Im guessing this is the data for the gps chip used in these www.locosystech.com/en/product/gnss-module-s5-1010-2r.html
Like boardsurfer thought there is a dead reconing mode and although i didnt read the whole thing in detail it apears that it can record what mode it was in at the time ( top of pag 11). Maybe just ask locosys to turn off what they call "Estimated (dead reckoning) Mode" or make it so it just reads the slow end of guessing in that mode so our nms dont get messed up.I looked at the sbp file and have no idea how to enterpret that but im sure some one else knows enough to work that out.

Theoretically, it would be possible to add information about dead reckoning to .sbp files, since the last byte in the sbp record is not used. But I doubt that dead reckoning is used in an "all or nothing" mode; it's more likely that it (or Kalman filters or similar) are used more and more as data quality gets worse.

But even a small change in the data structure is problematic. It should not be needed since we already have the SDOP values. The issue is that while errors are random enough at low SDOPs, they are non-random (biased, coupled) at high SDOP values, which can lead to large speed errors (similar in size to the SDOP values of single points, even when averaging many points). For GPSTC, this means that 2 second speeds may be 2 knots to high, and 10 second speeds may be 1 knot too high, if the watch is facing towards the water. This happens much more often than expected: the second time Boro used his GW-60, and during the first 15-minute test drive I did to investigate this issue.

The problem data can be excluded when analyzing with GPSResults by reducing the SDOP threshold to 1.5 (that is the default value for 1 Hz data, but it was increased to 3.0 for 5 Hz data). Unfortunately, that has some drawbacks. On my Mac, it's wicked slow, since GPSResults re-calculates everything for every 0.1 knot change. More importantly, GPS Action Replay does not support SDOP filters. I don't know if ka72 does, and what values it uses if it does.

Select to expand quote

mathew said..
I would be interested to see the error-bars for each graph

Below is a screen shot from the 6 GPS test drive. I think I turned my arm a bit sideways during the last few minutes.

I am feeling very ignorant, but can someone please tell me what SDoP means?

Select to expand quote

Windxtasy said..
I am feeling very ignorant, but can someone please tell me what SDoP means?

It refers to the estimated error range value for the Doppler speed. It is calculated by the GPS using the data on signal strength, number of satellites used, position of satellites in the constellation and possibly some other factors.

It is reported as a +/- figure for each Doppler speed data point. Dr Chalko coined the term SDOP to describe it (Speed Dilution Of Precision).

Select to expand quote

boardsurfr said..

Simon100 said..
Im guessing this is the data for the gps chip used in these www.locosystech.com/en/product/gnss-module-s5-1010-2r.html
Like boardsurfer thought there is a dead reconing mode and although i didnt read the whole thing in detail it apears that it can record what mode it was in at the time ( top of pag 11). Maybe just ask locosys to turn off what they call "Estimated (dead reckoning) Mode" or make it so it just reads the slow end of guessing in that mode so our nms dont get messed up.I looked at the sbp file and have no idea how to enterpret that but im sure some one else knows enough to work that out.

Theoretically, it would be possible to add information about dead reckoning to .sbp files, since the last byte in the sbp record is not used. But I doubt that dead reckoning is used in an "all or nothing" mode; it's more likely that it (or Kalman filters or similar) are used more and more as data quality gets worse.

But even a small change in the data structure is problematic. It should not be needed since we already have the SDOP values. The issue is that while errors are random enough at low SDOPs, they are non-random (biased, coupled) at high SDOP values, which can lead to large speed errors (similar in size to the SDOP values of single points, even when averaging many points). For GPSTC, this means that 2 second speeds may be 2 knots to high, and 10 second speeds may be 1 knot too high, if the watch is facing towards the water. This happens much more often than expected: the second time Boro used his GW-60, and during the first 15-minute test drive I did to investigate this issue.

The problem data can be excluded when analyzing with GPSResults by reducing the SDOP threshold to 1.5 (that is the default value for 1 Hz data, but it was increased to 3.0 for 5 Hz data). Unfortunately, that has some drawbacks. On my Mac, it's wicked slow, since GPSResults re-calculates everything for every 0.1 knot change. More importantly, GPS Action Replay does not support SDOP filters. I don't know if ka72 does, and what values it uses if it does.

GPSAR does have an SDOP and VSDOP function. Version 5.23

Select to expand quote

Te Hau said..

GPSAR does have an SDOP and VSDOP function. Version 5.23

All functions in GPSAR Pro are for display only; SDoP is not used in speed analysis by GPSAR Pro. You can show SDOP graphs (which I did above - the screen shots are from GPSAR). But only GPSResults removes points with high SDOP values. GPSAR Pro has only HDoP and minimum satellite filters in the speed panel:


GPSAR Pro has a separate "Filter" function, which allows filtering by max speed and max acceleration, but again not by SDOP.

BTW, the graphs have been in GPSAR for many versions. What's new in recent versions is that the SDoP (+-) numbers are shown in the track points tables. I had a bunch of email exchanges with Yann about this feature, which I find very useful for analyses like this one.

Is there an easy solution for this possible 2 knot error most people would be completely unaware of where of when they post keeping in mind that most people have no interest in checking the data in different programs and are only sailing for fun ? Cant the watch just be updated to always minus the margin of error if its over the acceptable limits.

So If I want to get faster speeds I should get rid of my GW52 and get a watch ? I always believe I was faster than what my gps read ??

I wear both my GW52 and GW60 for each sailing session. I use GPS Results to process logged data from each devise. There is rarely any difference (greater than 0.1kt) between the processed results for each device for each session. If there is a difference, then I use the result that has the least SDOP error. I then post from this chosen device via KA72 to GPSTC. This is usually the GW52. Interestingly, the GW52 is usually also faster.

Thanks John that gives me confidence in the old girl, I won't rush out to buy a watch ??

Select to expand quote

Tinlyds said..
Thanks John that gives me confidence in the old girl, I won't rush out to buy a watch ??

The watch is worthwhile getting because it is great for feedback on the water.

Like John I use both, but the GW52 on my head is more accurate than the watch on my wrist and usually gives slightly faster speeds.
Also like John, I'm going to invest in an extension strap so I can wear the watch on my upper arm. That will improve both it's accuracy and visibility on the water.
As John says the big advantage of the watch is the alpha firmware's increased options.
It also has a bigger memory, I found the advantage of this today. Because there's no need to conserve the watches memory, I have min speed set to 0kts, whereas the GW52 is set to 2kts. May not sound like much but it meant the watch gave me a 4kt better NM than the GW52. The best NM had a short period where speed dropped below 2kts, so GPSResults ruled this run invalid for the GW52.

Select to expand quote

decrepit said..
edit...
....It (GW60) also has a bigger memory, I found the advantage of this today. Because there's no need to conserve the watches memory, I have min speed set to 0kts, whereas the GW52 is set to 2kts. May not sound like much but it meant the watch gave me a 4kt better NM than the GW52. The best NM had a short period where speed dropped below 2kts, so GPSResults ruled this run invalid for the GW52.

For GPS-Results users: My normal practice to to reset the min speed value to 0 in the GPS-Results software. It is very annoying that this has to be done every time the software is opened, because the default setting is 5 knots. With this setting it will calculate the same distance as the other software packages (KA72, GPSAR-Pro,Realspeed), subject to very small, occasional differences.

For GW52 users: Unless you have a session that lasts for more than 5 hrs, there is ample memory even with the min speed set to 0. If you may be at the venue for longer than that, and your session could include one or more long breaks between sailing, there is no problem with just turning the GPS off during breaks, and on when you resume.

Select to expand quote

John340 said..
I wear both my GW52 and GW60 for each sailing session. I use GPS Results to process logged data from each device. There is rarely any difference (greater than 0.1kt) between the processed results for each device for each session. If there is a difference, then I use the result that has the least SDOP error. I then post from this chosen device via KA72 to GPSTC. This is usually the GW52. Interestingly, the GW52 is usually also faster.

My experience is almost exactly the same..

Since the GW60 was released I have been using it concurrently with at least 1 x GW52 (and often 2 x GW52) for almost every one of my many scores of sessions. (and often also with a ublox 10Hz logger). So far I have not seen any of the large (or small) errors such as the ones Boardsurfr has described, despite the fact I wear the GW-60 on my front (right) wrist, and therefore under grip, on my predominantly Starboard speed runs. Yes, the SDoP is almost always higher on the GW-60, but almost never over the standard filter values I use in GPS-Results (the old default, lower value of 3). Because of the higher SDoP values from the watch, my normal practice is to post the data from the GW-52, but it is very rarely different enough to have any significance.

I must add that all the places I sail have very good and wide sky view and usually excellent reception.

That said, there are clearly situations where errors may occur, and the software SDoP filter looks like the best way to cope with that.

I am keen to try an extension strap to wear the GW-60 on my upper arm for comparison.

There seem to be two things that can cause GW-60 data to be incorrect by a knot or more:
1. Poor GPS reception, combined with an underhand grip.
2. Under-water situations, for example after crashes and when cleaning weed of fins.

1. Underhand grip artifacts: It seems that typically, the reception is good enough even with an underhand grip. The exceptions I have seen are from a mountain lake in Nevada. Even though the mountains are a few miles away, they apparently can block enough satellites to create a marginal signal. It's still good enough for an overhand grip, but an underhand grip can cause the signal to become so bad that the GPS chips starts "guessing" the speed. The typical SDoP errors in such regions are around 1.2-2, below the filter thresholds in GPSResults. This should not affect many speed surfers in Australia or the Netherlands, but definitely could be an issue at spots like Lake Garda in Italy.

2. "Under water" artifacts: This kind of artifact can happen anywhere, I have some examples from Australia. The doppler speed can shoot up to 60 knots or more, even if the surfer is clearly swimming or trying to waterstart. Often, these artifacts are short, and have SDoP (+/-) numbers above 4. In that case, artifacts are easily identified by GPSResults. Future versions of ka72.com and GPS Action Replay will also have SDoP filters, and remove such artifacts.

However, the speed in an "under water" artifact sometimes stays high for 3-4 seconds, and the SDoP values can drop below the current filter thresholds (see the image below for an example). In such cases, all programs (including GPSResults) will give incorrect results. The SDoP filters as currently implemented cannot catch these artifacts. This would require filters that use the average SDoP values for the entire 2-second region (or something similar, like the standard deviation). I have not heard from any of the authors that they are planning to implement such a filter, so we cannot assume that software will identify such artifacts. However, they (a) only seem to affect 2-second values, and (b) can easily be identified by checking the numbers and looking at the traces. If the 2-second number is more than 2 knots higher than the best 10-second number, that's suspicious; if it seems to be at the very start or end of a run, that makes it very likely to be false (unless you had a huge crash just after reaching top speed). This can even be seen on ka72.com. In GPSResults or GPS Action Replay, the doppler speed graphs make it easy to spot such problems: the top speed will be in the middle of a very short "run", as shown in this example:

If you're curious about GPS reception, there's a cool tool at www.gnssplanning.com which can show the number of satellites visible, where they are, etc, for any spot. For example, changing the "cutoff" setting from the default of 10 degrees to 20 degrees to simulate a lake mountain situation shows that the minimum number of visible satellites can drop from 11 to 8 (using GNS and Glonass satellites). With just 8 visible satellites, any problems with the signal from just one or two satellites can cause the signal to become too weak for accurate data.

Select to expand quote

sailquik said..

John340 said..
I wear both my GW52 and GW60 for each sailing session. I use GPS Results to process logged data from each device. There is rarely any difference (greater than 0.1kt) between the processed results for each device for each session. If there is a difference, then I use the result that has the least SDOP error. I then post from this chosen device via KA72 to GPSTC. This is usually the GW52. Interestingly, the GW52 is usually also faster.

My experience is almost exactly the same..

Since the GW60 was released I have been using it concurrently with at least 1 x GW52 (and often 2 x GW52) for almost every one of my many scores of sessions. (and often also with a ublox 10Hz logger). So far I have not seen any of the large (or small) errors such as the ones Boardsurfr has described, despite the fact I wear the GW-60 on my front (right) wrist, and therefore under grip, on my predominantly Starboard speed runs. Yes, the SDoP is almost always higher on the GW-60, but almost never over the standard filter values I use in GPS-Results (the old default, lower value of 3). Because of the higher SDoP values from the watch, my normal practice is to post the data from the GW-52, but it is very rarely different enough to have any significance.

I must add that all the places I sail have very good and wide sky view and usually excellent reception.

That said, there are clearly situations where errors may occur, and the software SDoP filter looks like the best way to cope with that.

I am keen to try an extension strap to wear the GW-60 on my upper arm for comparison.

Andrew,
I've used a strap extension. The watch works really well on the upper forearm. However the extension strap I bought on EBay failed and I nearly lost the watch. I'm waiting until Locosys brings out their extension strap before I try it again.

Select to expand quote

boardsurfr said..
There seem to be two things that can cause GW-60 data to be incorrect by a knot or more:
1. Poor GPS reception, combined with an underhand grip.
2. Under-water situations, for example after crashes and when cleaning weed of fins.

1. Underhand grip artifacts: It seems that typically, the reception is good enough even with an underhand grip. The exceptions I have seen are from a mountain lake in Nevada. Even though the mountains are a few miles away, they apparently can block enough satellites to create a marginal signal. It's still good enough for an overhand grip, but an underhand grip can cause the signal to become so bad that the GPS chips starts "guessing" the speed. The typical SDoP errors in such regions are around 1.2-2, below the filter thresholds in GPSResults. This should not affect many speed surfers in Australia or the Netherlands, but definitely could be an issue at spots like Lake Garda in Italy.

2. "Under water" artifacts: This kind of artifact can happen anywhere, I have some examples from Australia. The doppler speed can shoot up to 60 knots or more, even if the surfer is clearly swimming or trying to waterstart. Often, these artifacts are short, and have SDoP (+/-) numbers above 4. In that case, artifacts are easily identified by GPSResults. Future versions of ka72.com and GPS Action Replay will also have SDoP filters, and remove such artifacts.

However, the speed in an "under water" artifact sometimes stays high for 3-4 seconds, and the SDoP values can drop below the current filter thresholds (see the image below for an example). In such cases, all programs (including GPSResults) will give incorrect results. The SDoP filters as currently implemented cannot catch these artifacts. This would require filters that use the average SDoP values for the entire 2-second region (or something similar, like the standard deviation). I have not heard from any of the authors that they are planning to implement such a filter, so we cannot assume that software will identify such artifacts. However, they (a) only seem to affect 2-second values, and (b) can easily be identified by checking the numbers and looking at the traces. If the 2-second number is more than 2 knots higher than the best 10-second number, that's suspicious; if it seems to be at the very start or end of a run, that makes it very likely to be false (unless you had a huge crash just after reaching top speed). This can even be seen on ka72.com. In GPSResults or GPS Action Replay, the doppler speed graphs make it easy to spot such problems: the top speed will be in the middle of a very short "run", as shown in this example:

I've noticed the underwater error as well. For me, it has only occurred when I've come to an abrupt halt after either hitting something or a gear breakage. Hence it's easy to identify and discount the result.

Select to expand quote

John340 said..

Andrew,
I've used a strap extension. The watch works really well on the upper forearm. However the extension strap I bought on EBay failed and I nearly lost the watch. I'm waiting until Locosys brings out their extension strap before I try it again.

If only I'd read this last night, I just ordered one!

Maybe I'll experiment with a lanyard as a watch saver.

Select to expand quote

decrepit said..

John340 said..

Andrew,
I've used a strap extension. The watch works really well on the upper forearm. However the extension strap I bought on EBay failed and I nearly lost the watch. I'm waiting until Locosys brings out their extension strap before I try it again.

If only I'd read this last night, I just ordered one!

Maybe I'll experiment with a lanyard as a watch saver.

Mike, I like the band to be stretched a little, so it holds the watch face in a fixed position. What I found is that this puts pressure on the pin on the strap extension. The pin wore through its polyurethane sleeve. Hopefully the Locosys extension will be better made.

Select to expand quote

John340 said..
Mike, I like the band to be stretched a little, so it holds the watch face in a fixed position. What I found is that this puts pressure on the pin on the strap extension. The pin wore through its polyurethane sleeve. Hopefully the Locosys extension will be better made.

Thanks John, I'll keep an eye on the pin, and see if I can rig a safety lanyard as well. I'm keen to document the accuracy difference between wearing on the upper arm and wearing on the wrist. And I also want to see if watching the watch while sailing, is any better than listening to speed talk on the phone.

Had my first sail today with the extension strap and lanyard. It didn't do much to improve my ability, to see what speed I'm doing while sailing. But it was a bit bouncy today, may be better if I get some smoooth stuff.

Had the "screen flip" option active, so the antenna points at the sky while the watch faces you.

It certainly improved the accuracy and removed the underhand/overhand effect.
Here's a comparison with the gw52


GW52 results GW60 results
500m:
time time
13:04:55 28.839+/-0.086 13:04:55 28.823+/-0.051
13:41:31 28.483+/-0.042 13:41:32 28.471+/-0.044
13:25:02 27.628+/-0.030 13:25:03 27.621+/-0.043
14:06:10 25.906+/-0.044 14:06:10 25.913+/-0.064
13:19:27 24.756+/-0.064 13:19:27 24.756+/-0.052

1852m:
time
13:41:48 26.928+/-0.024 13:41:48 26.930+/-0.025
13:06:36 26.873+/-0.045 13:06:36 26.854+/-0.026
14:06:19 23.708+/-0.022 14:06:19 23.703+/-0.024
13:26:11 23.411+/-0.016 13:26:11 23.409+/-0.024
12:59:06 20.077+/-0.025 Watch wasn't on for this one

2s:
time
13:06:59 32.178+/-0.333 13:06:59 32.097+/-0.247 overhand
13:38:50 30.557+/-0.223 13:38:49 30.505+/-0.112 overhand
12:58:07 29.957+/-0.399 Watch wasn't on for this one
13:18:59 29.951+/-0.367 13:18:59 29.964+/-0.127 Underhand
13:24:58 29.836+/-0.101 13:24:58 29.702+/-0.183 overhand

10s:
time
13:07:04 30.915+/-0.129 13:07:04 30.939+/-0.118
13:38:53 29.226+/-0.080 13:38:53 29.176+/-0.100
13:25:04 29.007+/-0.062 13:25:04 28.973+/-0.095
12:58:12 28.905+/-0.172 watch wasn't on for this one
13:19:04 28.674+/-0.086 13:19:04 28.698+/-0.163

alpha
time
13:48:37 15.367+/-0.052 13:48:37 15.391+/-0.047
13:38:13 15.353+/-0.037 13:38:13 15.350+/-0.045

Interestingly there's no swap over of runs, the time sort and speed sort produce the same order.
The differences between the GW52 and watch are well withing the +/- values, And most of the differences are under 0.1knots
I suppose I should work out which runs are overhand and which underhand, because I was going both ways. Just checked the two secs and there's only one underhand in the top 5, but it's got the second best SDoP

If anybody wants the files, just PM me with your email address.

Great work Mike.

Those values are certainly a LOT closer than when the watch is worn on the wrist, especially when worn under grip.

I am going to try forearm mounting as I used to do with the old Foretrex GPS. In that position I could actually see the display while sailing with the GPS on my front arm and under or over grip hardly changes the GPS position on the upper forearm. Not that I really need that now with the GPS-Logit speed talk.

I've just had a reply from Willy at Locosys.




He says these strap extenders are now available, and some retailers are stocking them.
Looking at where the buckle pin joins the strap, I think I see a bulge, or thickening of the strap where the pin passes through it.
The ebay strap doesn't have the extra thickness for the pin, I guess that's why John's broke on him.

And here's a couple of screen shots from today. GW52 on my head and watch just above the right elbow.
The first one is the speed graph and tracks from the watch of my alpha today



The SDoP envelope is very tight and the doppler and positional tracks are very close even through the gybe.

And here's the GW52 for comparison.




So extremely close, +/- value the same

Select to expand quote

decrepit said..
I've just had a reply from Willy at Locosys.




He says these strap extenders are now available, and some retailers are stocking them.
Looking at where the buckle pin joins the strap, I think I see a bulge, or thickening of the strap where the pin passes through it.
The ebay strap doesn't have the extra thickness for the pin, I guess that's why John's broke on him.

I've just ordered one from surfsail australia. I'll report back when it arrives.