Breaking the Wall

A.M. Ran with the kids, and Rob. Rob did 4, Jacob 1, Joseph 2, Jenny 2, Julia 3, William 0.5, Benjamin 8. I did a post-VPB pickup for 0.8 miles at 5:21 pace down the canyon. Felt strong. Would have been faster if the road was completely dry, it was OK, but I had to hold back in some spots for safety reasons.

A.M. Finished the Fast Running Friend data collection feature that allows me to gather and analyze the GPS data from it. Tried it out. During the 1 mile with Jacob and Julia it crashed, of course. I neglected to release the JNI references after each flushing operation and it eventually overflowed the 512-entry reference table, plus I also had a bug that was flushing the data wrong. I ran 2 more with Julia, Jenny, and Joseph, and 1 more with Rob. Rob ended up with 4, Joseph 2, Jacob 1, Jenny 2, Julia 3.

Then I fixed the bugs, and tried again. During the 8 mile run with Benjamin and later on everything went well as far as timing is concerned, and the distance also was reported correctly. I did a tempo run for 2 miles down the Provo Canyon. Benjamin joined me for 1 mile. He would have joined me for 2, but his foot has been giving him trouble so we decided to be cautious. We went through the first mile in 5:27, then I did 5:33. Felt like I did not have the top gear. Not surprising - I have not been sleeping well in the last two weeks - Matthew is keeping us up. And with all the snow I have not done anything fast. So all in all I think 11:00 for 2 miles is a good starting point.

After that I ran 0.5 with William and 2.5 myself. Collected more GPS data, analyzed in the afternoon. Discovered a small bug that does not affect anything for the user, only my analysis. Will continue working on finding a good GPS algorithm.

Total of 15.25 miles. 

Day of rest. Went to church. It was a Fast Sunday. I recall the kids asking when they were younger why it was called Fast Sunday since we did not run on Sunday at all. We had a good lesson on the translation of the Book of Mormon in Sunday school.  I find my life to be quite more simple if I accept the origin of the Book of Mormon to be what Joseph Smith said it was. It solves so many problems that you have to deal with otherwise trying to explain how it got to us. Of course, that has certain issues as well - you have to believe that God is real, that you will have to be accountable for your life when you meet Him, you have to do some things in this life that you may not feel like doing at the time, e.g get 8 kids to church by 9 AM and be there for 3 hours every week, but you find that even those things turn our to be for your benefit long-term in a way that you possibly could not have thought of or imagined. And there are many things that are required that I would do anyway. I find that simplicity inspired, and it witnesses to be of the truthfulness of the Book of Mormon and everything that follows from that.

A.M. Ran with Rob - 4 miles. Jacob was the only non-Benjamin kid that was somewhat healthy enough to try running and he did a mile with difficulty. Everybody else was afflicted with a cough. I did 8 with Benjamin afterwards including a short tempo run, which went better than on Saturday. I still did not feel 100% fit, but I was able to do 5:28-5:29 to get 10:57 for 2 miles down the canyon. Benjamin ran the first mile with me. When I got home I ran a mile with Sarah. Total of 13 miles. 

A.M. Total of 12. Even Benjamin started getting sick, so I had him cut the run down to 4 miles. Jenny did a mile. Everybody else rested. I ended up with a total of 12 miles.

Worked on the Fast Running Friend trying to come up with a good GPS correction algorithm. I tried several variants of bad point elimination using a simulator and plotting the resulting corrected path on the map. Naive approach of "if the angle is too big get rid of the point" kind of worked, but failed to eliminate a detour. This was a good educational experience as I learned a lot about what kind of data you get from a GPS.  Most of the time your position is wrong, but most of the time it is close enough to your route that it will not affect the total distance computed over an interval as short as 0.05 mile. However, there are minority nasty outlier points that take you on abrupt detours, and that is the main source of the annoying GPS watch distance inaccuracy. So my next approach is going to be this - break the route into samples, for each sample compute the mean and the standard deviation of the route bend angles, adjust the outlier points so their angle is within the standard deviation, and recompute the distance and pace. I have hopes for this algorithm. If not, I'll think of something else.

Once I figure out how to correct it, things are going to be very good, I might consider Fast Running Friend to be a true friend indeed when doing speed workouts or racing.

A.M. Everybody still sick. Had to cut the mileage for some, and others had to rest. Benjamin did 3, Jenny, Julia, Joseph, and Jacob did 0.5, William 0.3. I ran total of 12. Did a pickup for 0.75 in 4:08 (5:30 pace) and could feel the effects of the cough I had picked up as well. Rob ran 5 miles. Sarah did 1.

A.M. A hard day. Matthew got a cough and did not sleep. I got a cough too. So I decided to cut the mileage a bit to survive. Ran total of 10. Benjamin did just 1 mile. William 0.3. Jacob, Joseph, Jenny, and Julia did 0.5. Worked on refining the Fast Running Friend GPS algorithm.

A.M. Another rough day. In survival mode. Ran 10 miles. Benjamin did 2, Jenny and Julia 1, Joseph and Jacob 1 as well. William 0.5.

A.M. Plowed snow. Then ran with Chad and the kids. We are still not 100%. So Jacob ran 1, Joseph 2, Benjamin 2, Jenny 2, and Julia 1. Chad and I did a pickup for 1.5 on a surface that started out being good enough for 5:40 pace, but then deteriorated to make 6:40 the best we could do. At first I felt like I was going to choke, but then my throat cleared out and I felt better. Our time for 1.5 was 9:19. 

P.M. 0.5 with Benjamin.

Day of rest. Went to church.

A.M. We are still recovering from the sickness, but I ran 12 miles today. Benjamin did 3, Jenny 2, Julia 1, Jacob 1, Joseph 2, William 0.5.

A.M. Another day of recovering from the cough. 12 miles total. Benjamin did 4, Jenny and Julia 2, Joseph 2, Jacob 1, William 0.5.

A.M. Finally found a good algorithm for GPS correction in the Fast Running Friend. It worked great in the profiler on my test case which was rather nasty. Started implementing it in the app itself.

 Total of 12 miles today again. Benjamin did 5, Jenny 3, Julia 2, Joseph 2, Jacob 1, William 0.5.

A.M. We are slowly recovering. I ran 12 again today. Benjamin did 5, Jenny 3, Julia 2, Joseph 1, Jacob 1, William 0.5. I did a pickup for 0.25 and felt like I was choking in my mostly dry cough, if there is such a thing as choking in dry cough.

A.M. We are slowly recovering from the cough. I still have one - it is quite violent after I finish the run, but it is dry. I ran 12 miles this morning.  Benjamin did 6, Jenny 3, Julia 2, Joseph 2, Jacob 1, William 0.5. Did a pickup for 0.5 uphill in 2:51. Felt like I was going to choke from coughing as I went.

P.M. Finished the main coding of the Fast Running Friend new GPS algorithm and started testing it. I was too lazy to make a testing simulator with pre-loaded GPS data, so my development cycle was fix->build -> go for a jog -> fix ... At first I was getting some interesting results and found myself having traveled the distance to the sun in a matter of seconds. After bug fixes it worked better, correctly reported the distance and pace for 0.25 miles, but then bugged out around 0.3. I ended up doing about a mile of jogging during the testing cycle.

A.M. 12.5 miles total. This morning was the Fast Running Friend test/debug/fix morning. I ran some with the kids and some alone. Benjamin did 6, Jenny 3, Julia 1, Joseph 2, Jacob 1, William 0.5. Once the bugs were fixed (it took 5 miles), the Fast Running Friend produced near perfect readings, just about as good as it gets for a GPS-only distance/pace estimator. During the last 4 miles with Benjamin it reported 3.99 miles for the distance that I believe to be 4 miles from my calibrated pace-based estimates, and the immediate pace (sampled over 0.08-0.15 mile segments) never contradicted my intuition. Then I dropped him off and went for 3 more. It reported 2.99 on my course, and again the reported immediate pace  never contradicted my intuitive estimate. I was quite impressed with how it noticed that I slowed down on the uphills and snow-covered parts, and sped up on the downhills.

P.M. Added a time of day sync with the GPS clock, as well as a separate debugging log to the Fast Running Friend, and ran another mile for the final pre-release test. Then made the official release. You can download it from the official Fast Running Friend site (see link on the left in my profile). 

Day of rest. Went to church. It was Benjamin's 14th birthday, so I ordained him to the office of a teacher in the Aaronic Priesthood. The full-time missionaries spoke in the Sacrament meeting. Interestingly enough, both of them are from Northern California and both grew up on a farm. One comes from a family of 11 children. The other was a convert to the Church, his family joined when he was 12 years old. He told about how his father was an alcoholic and his parents were in the process of divorce when they started meeting with the missionaries. His father was quite stubborn but eventually decided he wanted to know for himself if the Book of Mormon was true. He prayed and received a strong positive answer. They joined the Church, he overcame his addition, and he and his wife are still together.

Then we had a great lesson in Sunday school on revelation, and a continuation of that of sorts in the Elder's Quorum - the topic was the Holy Ghost. The lesson manual mentions how President Lorenzo Snow and other missionaries were thrown off a boat during a storm in Hawaii in 1864 and when they found President Snow he was not breathing. They did what they knew how to do, and he was still not breathing. Then they had a prompting to do what we know as mouth-to-mouth resuscitation, which was successful. It was interesting to read their account of it - they did not know the word for the procedure and they just described it.

A.M. Total of 12 miles. We are feeling better although not 100%. Benjamin did 7, Jenny 3, Julia 0.5, William 0.5, Joseph 2, Jacob 1. Tested the Fast Running Friend again with perfect results both on the distance and immediate pace.

I am going to announce a "special offer". If you bring a MotoACTV or any Android-based GPS-capable device that you are willing to run with to my house I will install Fast Running Friend on it. If you bring your laptop, I will configure it so that it has the tools for me to do remote maintenance on your device. If Fast Running Friend bugs out on your device I will investigate and fix the problem. As insurance, if I brick your device and cannot repair it in time, I will buy it from you for a price you consider fair. All services are free of charge. The offer lasts until I am tired of people coming to my house. If you are interested, send me a private message to make the arrangements.

A.M. Total of 12. Benjamin did 7, Jenny 3, Julia 1, Joseph 2, Jacob 1, William 0.5. We are slowly recovering. I did a little workout to test my health and see how much fitness I've lost. Did a mile up the canyon in 6:08, and after a little rest another mile back down in 5:34. It felt hard, both up and down. Fast Running Friend was perfect with one small exception. Coming out of the canyon it reported pace 5:53, and then 5:57 when the actual pace was around 5:45. Then immediately after it reported 5:18 when the actual pace was around 5:28. However that spot is set up very well for signal distortion with the trail squeezing through a narrow opening between two tall rocks. I have not yet analyzed the data, but I've seen a similar pattern in another test sample and I am suspecting the problem is that we pick up the signal with virtual delay - that is the signal keeps us on the trail, at least the points the Fast Running Friend considers legitimate are on the route, but they are shifted back. So the GPS driver in other words tells the Fast Running Friend that we are now where we were in reality one second ago. If that indeed is the problem, there is absolutely nothing we can do to correct for that short of making the sampling distance too long because in the absence of some other feedback we cannot tell if the runner legitimately slowed down by 10 seconds per mile and then sped back up or if he ran evenly. Such a change of pace is completely legitimate and the erroneous point lies on the course.

I am still working on refining the GPS part of the Fast Running Friend, mostly the usability stuff - configuration file reading and writing, then UI to edit it on the device and via web browser if the device can connect to WiFi, support for splits of various kind, ability to upload workouts to the Fast Running Blog, etc, but I already ordered Garmin Footpod and will start research once it gets here. I have high hopes for the footpod  - I think it will have more accuracy than GPS, and also less power utilization.

A.M. Total of 12. Benjamin did 7, Jenny 3, Julia 1, Joseph 2, Jacob 1, William 0.5. I did a pickup down the canyon for 1.25 in 6:46. Felt stronger than the day before. More sleep helped, I think.

A.M. Total of 12. Benjamin did 7, Jenny 3, Julia 1, Jacob 1, Joseph 2, William 0.5. Worked on adding wireless connectivity to the Fast Running Friend. So far it has been giving me good results distance/pace wise - has not bugged out yet.

A.M. Total of 12. Benjamin did 8, Jenny 3, Joseph 2, Julia 1, Jacob 1, William 0.5. Discovered a bug in the Fast Running Friend. If you pause, walk over to the new location, and then restart, it adds the displacement to the distance. Will fix it. Finished the configuration file reading/writing code, and worked on the wireless connectivity. The idea is that to configure it the user can just start wireless configuration daemon on the device using a menu function, and then connect to the device through a browser. The only issue is how to get the device to figure out the wireless authentication credentials initially - my plan at this time is that he just presses the "enable mass storage mode" on the device menu, then connects the devices to the desktop/laptop via USB, then puts his SSID and password in the config file, disconnects the device, and Fast Running Friend figures out the rest. One problem, of course, is that some users are too illiterate to open and edit a plain text file :-) If it is Microsoft Word, or PDF, they do not have a problem, but I've had accounting departments return my invoices that I sent as a plain text attachment with .txt extension complaining that they could not figure out how to open the file. Indeed Microsoft does not "delight in plainness", and definitely does not help its users in that cause. So maybe I should make that config file in Microsoft Word format :-)

A.M. Total of 15.5. Ran with Chad and the kids. Benjamin did 8.5, Jenny 3, Julia 0.5, Joseph 2, Jacob 1, William 0.5. 

Fast Running Friend had a chance to prove itself against Chad's Garmin 210. The course was of known length with marks. In the first 5.04 miles Fast Running Friend said 5.03, Garmin said 4.99. Garmin got off track on the distance over about a half mile stretch, then maintained the difference. Fast Running Friend was within 0.01 on all intermediate splits. Then over the next 2.04 miles Fast Running Friend added 2.02, while the Garmin added 2.05, so now the Garmin was closer to the truth that it was before. Then Chad accidentally disabled the GPS, so we were testing just the Fast Running Friend. Over the next 4.00 it reported 4.03, then after that it stayed on track until it ran out of battery with the total running time of around 1:53, but we started at 80% battery charge.

Fast Running Friend was without question more accurate on the immediate pace. This one is difficult to verify exactly, but both Chad and I agreed whenever we sampled it that Fast Running Friend was reasonable every single time, while Garmin made sense only 60% of the time or so.

Fast Running Friend also showed some resilience in a situation that I did not explicitly think through in my code. At around 1.4 mark I made a VPB stop and did not press the Stop button. When I was done and started running, the Fast Running Friend extrapolated my distance to be about where Chad was at the time. Then once it noticed that I've gone 0.15 from the last trusted point it corrected the distance moving me backwards, and eliminating the detour because the turn angles in the path were too high and disruptive. By the time I caught up to Chad the Fast Running Friend was showing the correct distance. It is a good sign when your code does the right thing when used in a way that you did not explicitly program for.

I do need to do something about the battery life. From the tests I've done so far, things look ugly. If I tell Android to give me GPS updates at a higher interval, it does give them to me at a higher interval so the accuracy is lost, but the GPS driver (/system/bin/SiRFDrv) is still reading the GPS device (/dev/ttyS0) once a second no matter what and the battery drains just as fast. So I might need to bypass the Android GPS notification system, and read the GPS device myself. This is going to be an adventure, but it has some payback potential - first, improvement in the battery life. Second, quicker acquisition of the GPS signal - I suspect somewhere in the convoluted communication process that takes the signal for the device to the application there is a bug that withholds the legitimate signal. There are times when the app is not getting the signal, the GPS driver logs says it is waiting for signal, yet I catch SiRFDrv red-handed reading the correct GPS coordinate in plain text from /dev/ttyS0. Third, this can give me the flexibility of varying the read interval quickly which could allow better battery life without sacrificing accuracy. I just need to learn more about how the GPS works on a lower level.

Day of rest. Went to church. Took a long nap when I got home. It was much needed.

A.M. Ran with Rob and the kids. Total of 12. Benjamin did 8, Jenny 3, Julia 1, Joseph 1, Jacob 1, William 0.5. 

A.M. Total of 12. Benjamin did 8, Jenny 3, Julia 1, Joseph 1, Jacob 1. Worked on the fix of the pause repositioning bug in the Fast Running Friend, but it was buggy. On the positive side got the wireless connectivity to work.

P.M. 0.5 with William. Fixed the repositioning bug and ran 0.75 testing the fix.

A.M. 12 miles total. Rob and Caleb joined us. Benjamin did 8, Jenny 3, Julia 1, Joseph 1, Jacob 1, William 0.5. Benjamin, Caleb, and I did a pickup for 1 K in 3:14 down the canyon. Caleb had Garmin Forerunner 610, so we had a chance to test it against the Fast Running Friend. Total distance in reality was 8.00 and that is what FR 610 reported. Fast Running Friend said 7.99. However, FR 610 won hands down on immediate pace, which surprised me. It sampled it every second, and every single time got it right on - verified against land marks. Fast Running Friend was OK, but it sampled every 0.08-0.15 miles, so the data came with delay.

Even though we've got David fighting Goliath here - one hacker that rights code in his spare time vs a large business, I am still competitive when it comes to what matters to me - precision. Besides, when David vs Goliath analogy is used it is frequently forgotten that it was David who won. So I started researching. First we tested the position  reported FR 610. When we stood in place it did not change - Fast Running Friend (working through Android and MotoACTV GPS receiver/driver) cannot stay still - it wiggles. To make sure that FR 610 was not playing tricks we moved a few feet. It changed. We went back to where we started. The position went back to what it was originally. So it was apparent that FR 610 had the upper hand on the position itself.

I did some research and discovered that FR 610 uses SiRFstarIV chipset which has a lot of cool features including the ability to neutralize the noise. It was quite an adventure to figure out what GPS chipset was in MotoACTV. I knew it was some SiRF chip, otherwise Motorolla would not have named the driver SiRFDrv., but all of my diagnostic tricks as well as internet searches failed to unearth the info. I got really mad, replaced the driver in /etc/location.cfg with a dummy one to avoid unwanted restart of the real one, killed the driver, then manually turned on the GPS via /dev/gps_standby and dumped /dev/ttyS0. Downloaded the protocol description PDF for SiRFStarIII, and found an message that had an unsupported code. Then I found the manual for SiRFStartIV and sure enough that message was documented. So from this I think we can fairly safely assume that MotoACTV does have SiRFStarIV just like FR 610.

The difference, it appears, is in the quality of the driver. It looks like while Garmin did a thorough job using all of the features of the hardware, Motorolla quickly put something together and threw it out on the market. Thus the difference in the quality of position reporting. This is correctable in Fast Running Friend, though. I can write the driver to match the one used by Garmin. This will be an adventure and a great learning experience.

A.M. 11.5 miles total. Benjamin did 8, Jenny 3, Julia 1.5, Joseph 2, Jacob 1, William 0.5.