Monday, September 2, 2013

FlySight - a better GPS?

The short version for my non-geeky readers: 
There is a better GPS unit out there. If you want to by a GPS for speedsurfing, do not buy the GT-31! If you are neither geeky nor in the market for a GPS, I suggest you stop reading now.

---

The long version (consider yourself warned!):
For GPS speedsurfing, the GPS that is strongly suggested or even required at sites like gps-speedsurfing.com and the GPS Team Challenge is the Locosys GT-31. The GT-31 is more accurate than most other GPS units because it records doppler-based speeds, which are less prone to random errors that the position-based speeds used by most other units.

However, the GT-31 has a few shortcomings - I described its limitation in jibe analyses in a previous post. It is also getting pretty old - in fact, Locosys has issued an official phase out notice for the GT-31. So I started looking for a commercially available alternative a few months ago.

The first unit I tested looked promising at first, but turned out to be a disappointment. It was based on a MediaTek GPS chip, which apparently cannot record doppler data. The position-based data showed huge spikes when tested for windsurfing - my top speed reached several hundred knots (albeit just for 1/10th of a second).

So I was a bit skeptical when I found a GPS that had been developed for sky diving and wingsuit jumps, the FlySight. The promising things about this unit was it can record 5 data points per second, and that it uses a chip (the u-blox 6 chipset) that definitely can record doppler speed. But before spending $250 to test it, I first contacted the developer/seller to see if I would be able to return the unit if it proved not to be useful. Then, the fun began - I got an enthusiastic answer within a few hours. About a week later, the unit arrived by mail, and the testing began. Here's a picture of the FlySight and the GT-31:
The FlySight is smaller and lighter, but does not have a display like the GT-31. It is not water proof, but the GT-31 also needs to be worn in a waterproof armband since it is not sufficiently waterproof for windsurf crashes. When testing the FlySight, I put it into a small ziplock bag inside a waterproof armband.

So far, I have tested the FlySight in three separate windsurf session. I wore both the FlySight and the GT-31 on separate armbands on the same arm. The results were very similar in each test - let's look at a doppler speed graph:

GT-31 data are blue, FlySight data red. The curves are very similar - most of the time, so similar that one curve is drawn on top if the other. How about speed?

Speeds are very similar, usually within 0.1-0.2 knots. The one exception are the Alpha-500 values (sailing 500 meters, with a jibe in the middle and the end points close to each other). For alphas, the GT-31 gives about 0.7 knots higher speeds than the FlySight, so this needs closer examination. Here are tracks for a couple of jibes:
The top graph shows the positions. The GT-31 (blue) determines position once per second, the FlySight (red) 5 times per second. The bottom shows the doppler speeds on the left, and the positional speed on the right. Here's another jibe:

We see a similar picture: the tracks reported by the FlySight are much smoother, and closer to what a jibe looks like - it is a gradual turn, not a very sudden direction change. Note that for the FlySight, the regular (position-based) speed graphs and the doppler speed graphs look quite similar. But for the GT-31, the regular speed graphs show multiple spikes that are absent from the doppler speed. However, the doppler speed in the GT-31 graphs drops about 2-3 seconds after the speed FlySight doppler speeds. Why?

I can only speculate, but keep in mind that all GPS data are filtered. It seems the doppler data from the GT-31 has a filter against sudden drops and direction changes; this makes sense when driving around normally in a car. The positional GPS data are often filtered using the doppler data, which are generally more accurate. So it the doppler data say that we still have a lot of forward speed, the position data are adjusted accordingly. This is why the GT-31 overshoots in jibes. After 2-3 seconds, the filter suddenly sees that the drop appears to be real, and doppler speed finally drops. Then, when we start accelerating in the new direction, the GT-31 needs to adjust the position in leaps and bounds to get to the correct position - hence the spikes. But since the GT-31 overstates the doppler speeds for 2-3 seconds, the reported alpha values for jibes like the ones above are artificially high.

-- please read my note at the bottom of this post that was added on 7/25/2014 about why the GT-31 tracks look so bad! --

The GT-31 accuracy problems in turns become obvious when comparing jibe analyses with and without doppler. First, here let's look at the jibe analysis from GT-31 data using positional speeds:
According to this, the best jibe had a minimum speed of 13.2 knots, or 61% of the entry speed. But the doppler data paint a completely different picture:
Here, the minimum speed in jibe #35 was only 7.1 knots instead of 13.2 knots! Now, lets look at the FlySight data - positional speeds first:
And doppler-based results:
The FlySight results are much closer to each other, with the same two jibes being on top of the list at minimum speeds that are within 10% of each other. This also means that the tracks (which are position based) are much more accurate for the FlySight than for the GT-31 - something that already was obvious by just looking at the tracks.

Most jibes that I looked at in the GT-31 data from different sessions had artifact similar to the ones shown above, but every now and then, I saw a jibe that looked decent on the GT-31 tracks, too - here is an example:
But even in this jibe, the delayed drop in doppler speed as well as the spikes in positional speed are visible in the GT-31 data. Clearly, the FlySight data give a much more accurate picture of what happens in jibes.

Most of the data shown above were from my first test session, which was on flat water (Duxbury bay). The second test session was at a different location, Kalmus, where we had quite a bit of chop, and small swell that invited a bit of wave play (not real wave sailing - the waves did not break, and I was on a 110 l freestyle board with a 30 cm weed fin). In the tracks from this session, I found a few spots where the doppler speeds between the two units differed:
Let's zoom in:
In the FlySight tracks (red), we can see a bit of wave play - three or four little turns down a wave, followed by turning back upwind a bit. The doppler speeds nicely show the speed gain when going down the wave, and the slowdown when turning back upwind. In stark contrast, the GT-31 tracks do not show most of the turns, and the speed changes are less pronounced and offset by a few seconds. Again, the FlySight data paint a much clearer picture of what was going on.

After seeing these results, I decided to do one more test in my car: driving around in circles on an empty parking lot. On a windsurf board, the speed in a jibe can suddenly change when you hit a piece of chop, sink the tail, or many other reasons. In a car, the speed will be much more constant if you don't stomp on the gas or brake. First, let's look at the doppler speeds of the ~10 minute test drive:
Good agreement overall, but a few spikes in the GT-31 data (I removed a few irrelevant drops where the GT-31 lost the GPS signal). The region in the middle deserves more attention:
The low-speed section is when I drove around in circles. Here's a Google Earth overlay of the FlySight tracks:

(BTW, the parking lot was empty - I did not drive over cars!) Here are the GT-31 tracks:
I know that we cannot sail around like this on a windsurf board, even if our 360s are great - but this clearly illustrates (a) the accuracy problems the GT-31 has in turns, and (b) that the FlySight provides more accurate data.

Just for fun, look at the overlay of the tracks from the start of this test drive. FlySight data:

GT-31 data:
I think the pictures say it all, but let me assure you - I was driving on the road, not next to it!

In summary, the FlySight GPS clearly provides data that are not only higher resolution, but also much more accurate. The FlySight is based on a newer GPS chip that can track Russian and European GPS satellites in addition to the US satellites that the GT-31 is limited to. In my windsurf tests, the GT-31 would typically track 6-8 satellites, while the FlySight tracked 10-14. More satellites mean more accuracy. But the number of satellites tracked is just part of the story - it appears that data processing algorithms in u-blox 6 GPS chip that the FlySight uses are superior to the algorithms in the older Sirf-3 chip that the GT-31 uses.

For my really geeky readers who want to examine the data from my tests, a zip file with all data is available here. The later tests were done with a version of the firmware that I modified to write binary UBX files (with much-appreciated help from Michael Cooper, the FlySight developer). I am currently working on additional modifications of the firmware related to the spoken speed announcements. I really like that the FlySight can be used with waterproof headphones to announce speed while you are sailing. Since the speeds are doppler speeds, this is much more accurate than using GPS Speed Talker with an Android phone, as I had described earlier.  But the lack of a display on the FlySight means that you cannot easily see your top speed without plugging the unit into a computer (or perhaps smartphone). I already made a version that remembers the top speed, and announces it occasionally when the GPS is stationary, so that you could just plug earphones in at the end of a session to see how fast you were. I plan to modify the firmware so that it can also remember 10-second averages, and announce them at the end of a session. Not sure when I'll get to it, though - the next weekends will be busy with ABK camp, the East Coast Windsurfing Festival, and a windsurf instructor class in Hatteras.
--
Added 7/25/2014:
I am still convinced that the FlySight gives more accurate data then the GT-31. However, I also discovered why my GT-31 in the images shown above looked to bad: I had my GT-31 set to "low power" mode. The low power mode restricts the number of satellites the GT-31 tracks, and leads to substantially lower accuracy. In "normal" power mode, tracks are a lot more accurate, and position-based speeds are much closer to doppler speeds. For more information, read the post "Low Power Is Bad" from July 22, 2014..