|Seems the nondoppler/doppler errors spoil both alpha and 1 hour runs. Because I use auto session posting it isn't possible to post one session with both doppler data for avg 2sec 500 aso, and non doppler for alpha, 1hour and session distance. |
Please give a clear sollution by changing the auto sent function or software so allways nondoppler is used for alpha and 1hour, stating wich can be used in the rules.
|I've done some high-level comparisons of my recent alpha results:|
- GT-11 and GT-31 <- I have been wearing both for the past month
- Doppler vs Positional
- GPSResults 6.02 vs GPSAR Pro 4.6
To summarise the differences in Alpha 500 results:
1) Doppler speeds are practically identical for GT-11 and GT-31 (<0.02 knots difference)
2) Positional speeds are approximately 0.5 knots higher than Doppler speeds when using GT-11
3) Positional speeds are approximately 0.2 knots higher than Doppler speeds when using GT-31
4) Positional speeds are about 0.3 knots higher on GT-11 than GT-31 (n.b. same as 0.5 - 0.2)
5) GPSResults reports approximately 0.2 knots more than GPSAR Pro
These high level figures lead me to suspect that the positional speeds are perhaps not to be completely trusted. The Doppler results are almost identical on the two devices and make comparisons fairer between GT-11 and GT-31 users.
Note that I haven't looked at the track data in great detail. All that I have done is match up runs and compare the results.
I have uploaded the spreadsheet to my web server:
|A major part of the reason for the positional error, is due to "dead reckoning" (http://en.wikipedia.org/wiki/Dead_reckoning).|
The easiest way to demonstrate this is to walk/drive with the GPS, then cover the GPS with a tin-can (or enter a tunnel) - now change direction 90deg, move some distance, then uncover the GPS. The resultant track will have extra/bogus data where the GPS didn't have a signal lock.
Dead-reckoning helps with bush-walkers, boating, etc. which might have occasional signal loss, as most moving objects dont change direction quickly. However, it would be nice to be able to disable it for speed-sailing (but doing so would show other errors too).
Note that a similar algorithm is applied to the Doppler data, but the effect is not as apparent.
|It's great to see someone else doing some testing Michael. Well done! :-)|
In my testing I found results were affected by the position of the GPS on the body. I had more errors and spikes (unrealistic shaped speed graphs) when the ZGPS is mounted on the arm than when it was mounted on the head/helmet. Where did you mount your GPS's?
When doing my comparisons I looked at the overlaid speed graphs of each run. This showed up some strange errors where either the Doppler or the Positional (and sometimes both) would have unrealistic steps or spikes in the speeds (usually mid gybe). In these cases there could be very large differences between the D and P. Did you find the same errors? If so, did you filter your results to remove the obvious errors?
My initial testing with GT-11 showed some seemingly quite unrealistic slow spikes in the doppler results compared with what I saw as more realistic speed graph curves for the Positional. There were also more of the obviously erroneous speed graph shapes in the Doppler. This was why I initially recommended sticking to the use of Positional for Alphas.
When the GT-31 come along, the positional results were horribly full of significant errors (when viewed on the speed graph) and the Doppler results sometimes showed the same errors. In this case I thought it better to use the doppler for Alphas if you had no other choice (GT-11).
An update of the GT-31 firmware made a huge difference to the GT-31 results (I think it was v1.2 1205 but not sure). After that the GT-31 gave very consistent results between the Doppler and Positional with very few errors (but still some so you have to check every result).
My comparisons at this stage showed very good correlation between the GT-11 and GT-31 Doppler, although I am still not totally convinced that the results are totally realistic/accurate.
Another factor come into it when you compare Doppler Alphas with GPS-Results and Realspeed.
Realspeed uses the positional data to find the start and end of a run and the proximity. GPS-Results uses the Doppler data for both the speed and to calculate position for the proximity. I have found some significant errors in using Doppler for the proximity positioning. A small Dopper error in the gybe (the most likely place for this) throws the rest of the positional calculation off. This can lead to a run not being counted when it should be, or not being counted when it should not be. The RealSpeed method is much better because there can not be any positional accumulated errors. Each positional calculation is independent. I think GPSAR uses the same method as RealSpeed for Alphas but I am not sure as I have done the comparisons.
Can you email me Michael so we can compare and collaborate. firstname.lastname@example.org
|Damn this non ability to edit posts!|
the sentence in the above post should read:
"This can lead to a run not being counted when it should be, or being counted when it should not be."
|To briefly answer your questions:|
- The GPS units are mounted on my upper right arm, top to tail in the same aquapack. The GT-31 is always at the top because it is the better one to look at. It's quite conceivable that this could lead to the units picking up my rig flip / rotation, etc.
- I am yet to do any detailed analysis on the runs themselves. Now that I have some high level figures, interesting ones can be picked out from the list and examined in more detail.
- The results themselves haven't been filtered in any way, other than the standard num sats, max hdop and max accel. However, they provide a good idea on the differences / errors when no manual examination is done (probably the norm for most users) and the GPS is worn on the forearm (again, probably the norm).
I'll drop you a quick e-mail before I go to work.