Josef “Jeff” Sipek

Google Traffic, iOS edition

Several years ago, I wrote about how Google gets traffic information and how to turn off this location reporting on Android phones. Since then I’ve switched to iPhones. While I normally use the built-in Maps app, I keep Google Maps installed as a fallback—just in case.

I upgraded my phone recently and so I spent some time going through all the apps and making sure they worked and didn’t have more access than necessary. This is when I discovered that the Google Maps app for iOS defaults to collecting location information and sharing it with Google. Given my previous post, this isn’t really surprising.

Turning it off

Anyway, as with the Android post, here are the steps to limit Google’s collection of location information.

First of all, in Settings → Privacy → “Location Services”, I changed Google Maps’s permission to “While Using the App”. If I’m not using the app, then it doesn’t need to know where I am.

Second, in the app itself, go to: Settings → “About, terms & privacy” → “Location data collection”. That’s right, this setting is buried in what appears to be a page with the boring legal notices.

And then turn off the the toggle:

That should do it…at least assuming that Google honors the settings in its own app.

Google Traffic

Ever wonder how Google gets its traffic information?


Apparently, there are two sources. The first is the Department of Transportation. The second consists of Android users.

You can always check Google Location History to see what sort of data Google has. (Of course, they may always have more than they show.) Seeing the data can be a bit unnerving. Since I’m not really into giving Google more data than they already have to begin with, and I see no reason for Google to know exactly where I spend my time, I decided to turn this feature off.

Turning it off

You can find the setting by running the “Google Settings” app. That’s right, not “Settings”. Once there, select “Location”.


As you can see, I want to treat Google apps like any other vendor’s apps. As an added bonus, it looks like my GPS is on way less often.

Garmin G1000 Data Logging: Cross-Country Edition

About a week ago, I talked about G1000 data logging. In that post, I mentioned that cross-country flying would be interesting to visualize. Well, on Friday I got to do a mock pre-solo cross country phase check. I had the G1000 logging the trip.

First of all, the plan was to fly from KARB to KFPK. It’s a 51nm trip. I had four checkpoints. For the purposes of plotting the flight, I had to convert the pencil marks on my sectional chart to latitude and longitude.

> xc_checkpoints
          Name Latitude Longitude
1      Chelsea 42.31667 -84.01667
2       Munith 42.37500 -84.20833
3       Leslie 42.45000 -84.43333
4 Eaton Rapids 42.51667 -84.65833

First of all, let’s take a look at the ground track.

ground track

In addition to just the ground track, I plotted here the first three checkpoints in red, the location of the plane every 5 minutes in blue (excluding all the data points near the airport), and some other places of interest in green.

As you can see, I was always a bit north of where I was supposed to be. Right after passing Leslie, I was told to divert to 69G. I figured out the true course, and tried to take the wind into account, but as you can see it didn’t go all that well at first. When I found myself next to some oil tanks way north of where I wanted to be, I turned southeast…a little bit too much. Eventually, I made it to Richmond which was, much like all grass fields, way too hard to spot. (I’m pretty sure that I will avoid all grass fields while on my solo cross countries.)

So, how about the altitude? The plan was to fly at 4500 feet, but due to clouds being at about 3500, Wikipedia article: pilotage being the purpose of this exercise, and not planning on going all the way to KFPK anyway, we just decided to stay at 3000. At one point, 3000 seemed like a bit too close to the clouds, so I ended up at 2900. Below is the altitude graph. For your convenience, I plotted horizontal lines at 2800, 2900, 3000, and 3100 feet. (Near the end, you can see 4 touch and gos and a full stop at KARB.)


While approaching my second checkpoint, Munith, I realized that it will be pretty hard to find. It’s a tiny little town, but sadly it is the biggest “landmark” around. So, I tuned in the JXN Wikipedia article: VOR and estimated that the 50 degree radial would go through Munith. While that wouldn’t give me my location, it would tell me when I was abeam Munith. Shortly after, I changed my estimate to the 60 degree radial. (It looks like 65 is the right answer.)

> summary(factor(data$NAV1))
109.6 114.3 
 3192  1406 
> summary(factor(data$CRS))
  36   37   42   44   47   48   49   50   52   57   59   60 
1444    1    1    1    1    1    1  135    1    1    1 3010 
> head(subset(data, HSIS=="NAV1")$Time, 1)
[1] "2013-05-31 09:43:23 EDT"
> head(subset(data, NAV1==109.6)$Time, 1)
[1] "2013-05-31 09:43:42 EDT"
> head(subset(data, CRS==50)$Time, 1)
[1] "2013-05-31 09:44:26 EDT"
> head(subset(data, CRS==60)$Time, 1)
[1] "2013-05-31 09:46:48 EDT"

When I got the plane, the NAV1 radio was tuned to 114.3 (SVM) with the 36 degree radial set. At 9:43:25, I switched the input for the HSI from GPS to NAV1; at 9:43:42, I tuned into 109.6 (JXN). 44 seconds later, I had the 50 degree radial set. Over two minutes later, I changed my mind and set the 60 degree radial, which stayed there for the remainder of the flight.

In my previous post about the G1000 data logging abilities, I mentioned that the engine related variables would be more interesting on a cross-country. Let’s take a look.

engine RPM

As you can see, when reaching 3000 feet (cf. the altitude graph) I pulled the power back to a cruise setting. Then I started leaning the mixture.

fuel flow

Interestingly, just pulling the power back causes a large saving of fuel. Leaning helped save about one gallon/hour. While that’s not bad (~11%), it is not as significant as I thought it would be.


Since there was nowhere near as much maneuvering as previously, the fuel quantity graphs look way more useful. Again, we can see that the left tank is being used more.

The cylinder head temperature and exhaust gas temperature graphs are mostly boring. Unlike the previous graphs of CHT and EGT these clearly show a nice 30 minute long period of cruising. To be honest, I thought these graphs would be more interesting. I’ll probably keep plotting them in the future but not share them unless they show something interesting.

cylinder head temperature exhaust gas temperature

Same goes for the oil pressure and temperature graphs. They are kind of dull.

oil pressure oil temperature

Anyway, that’s it for today. Hopefully, next time I’ll try to look at how close the plan was to reality.

Garmin G1000 Data Logging

About a month ago I talked about using R for plotting GPS coordinates. Recently I found out that the Wikipedia article: Cessna 172 I fly in has had its G1000 avionics updated. Garmin has added the ability to store various flight data to a CSV file on an SD card every second. Aside from the obvious things such as date, time and GPS latitude/longitude/altitude it stores a ton of other variables. Here is a subset: indicated airspeed, vertical speed, outside air temperature, pitch attitude angle, roll attitude angle, lateral and vertical G forces, the NAV and COM frequencies tuned, wind direction and speed, fuel quantity (for each tank), fuel flow, volts and amps for the two buses, engine RPM, cylinder head temperature, and exhaust gas temperature. Neat, eh? I went for a short flight that was pretty boring as far as a number of these variables are concerned. Logs for cross-country flights will be much more interesting to examine.

With that said, I’m going to have fun with the 1-hour recording I have. If you don’t find plotting time series data interesting, you might want to stop reading now. :)

First of all, let’s take a look at the COM1 and COM2 radio settings.

> unique(data$COM1)
[1] 120.3
> unique(data$COM2)
[1] 134.55 120.30 121.60

Looks like I had 3 unique frequencies tuned into COM2 and only one for COM1. I always try to get the Wikipedia article: ATIS on COM2 (134.55 at KARB), then I switch to the ground frequency (121.6 at KARB). This way, I know that COM2 both receives and transmits. Let’s see how long I’ve been on the ATIS frequency…

> summary(factor(data$COM2))
 120.3  121.6 134.55 
     1   3303     70 

It makes sense, between listening to the ATIS and tuning in the ground, I spend 70 seconds listening to 134.55. The tower frequency (120.3 at KARB) showed up for a second because I switched away from the ATIS only to realize that I didn’t tune in the ground yet. Graphing these values doesn’t make sense.

I didn’t use the NAV radios, so they stayed tuned to 114.3 and 109.6. Those are the Salem and Jackson VORs, respectively. (Whoever used the NAV radios last left these tuned in.)

To keep track of one’s altitude, one must set the Wikipedia article: altimeter to what a nearby weather station says. The setting is in Inches of Mercury. The ATIS said that 30.38 was the setting to use. The altimeter was set to 30.31 when I got it. You can see that it took me a couple of seconds to turn the knob far enough. Again, graphing this variable is pointless. It would be more interesting during a longer flight where the barometric pressure changed a bit.

> summary(factor(data$BaroA))
30.31 30.32 30.36 30.38 
  262     1     1  3110 

Ok, ok… time to make some graphs… First up, let’s take a look at the outside air temperature (in °C).

> summary(data$OAT)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
    4.0     6.8    12.2    11.5    16.0    18.5 


In case you didn’t know, the air temperature drops about 2°C every 1000 feet. Given that, you might be already guessing, after I took off, I climbed a couple of thousand feet.


Here, I plotted both the altitude given by the GPS (Wikipedia article: MSL as well as Wikipedia article: WGS84) and the altitude given by the altimeter. You can see that around 12:12, I set the altimeter which caused the indicated altitude to jump up a little bit. Let’s take a look at the difference between the them.

altitude difference

Again, we can see the altimeter setting changing with the sharp ~60 foot jump at about 12:12. The discrepancy between the indicated altitude and the actual (GPS) altitude may be alarming at first, but keep in mind that even though the altimeter may be off from where you truly are, the whole air traffic system plays the same game. In other words, every aircraft and every controller uses the altimeter-based altitudes so there is no confusion. In yet other words, if everyone is off by the same amount, no one gets hurt. :)

Ok! It’s time to look at all the various speeds. The G1000 reports indicated airspeed (IAS), true airspeed (TAS), and ground speed (GS).


We can see the taxiing to and from the runway — ground speed around 10 Wikipedia article: kts. (Note to self, taxi slower.) The ground speed is either more or less than the airspeed depending on the wind speed.

Moving along, let’s examine the lateral and normal accelerations. The normal acceleration is seat pushing “up”, while the lateral acceleration is the side-to-side “sliding in the seat side to side” acceleration. (Note: I am not actually sure which way the G1000 considers negative lateral acceleration.)


Ideally, there is no lateral acceleration. (See Wikipedia article: coordinated flight.) I’m still learning. :)

As you can see, there are several outliers. So, why not look at them! Let’s consider an outlier any point with more than 0.1 G of lateral acceleration. (I chose this values arbitrarily.)

> nrow(subset(data, abs(LatAc) > 0.1))
[1] 41
> nrow(subset(data, abs(LatAc) > 0.1 & AltB < 2000))
[1] 28

As far as lateral acceleration goes, there were only 41 points beyond 0.1 Gs 30 of which were below 2000 feet. (KARB’s pattern altitude is 1800 feet so 2000 should be enough to easily cover any deviation.) Both of these counts however include all the taxiing. A turn during a taxi will result in a lateral acceleration, so let’s ignore all the points when we’re going below 25 kts.

> nrow(subset(data, abs(LatAc) > 0.1 & GndSpd > 25))
[1] 26
> nrow(subset(data, abs(LatAc) > 0.1 & AltB < 2000 & GndSpd > 25))
[1] 13

Much better! Only 26 points total, 13 below 2000 feet. Where did these points happen? (Excuse the low-resolution of the map.) You can also see the path I flew — taking off from runway 6, making a left turn to fly west to the practice area.


The moment I took off, I noticed that the Wikipedia article: thermals were not going to make this a nice smooth ride. I think that’s why there are at least three points right by the highway while I was still climbing out of KARB. The air did get smoother higher up, but it still wasn’t a nice calm flight like the ones I’ve gotten used to during the winter. Looking at the map, I wonder if some of these points were due to abrupt power changes.

Here’s a close-up on the airport. This time, the point color indicates the amount of acceleration.


There are only 4 points displayed. Interestingly, three of the four points are negative. Let’s take a look.

                    Time LatAc  AltB  E1RPM
2594 2013-05-25 12:52:10 -0.11 879.6 2481.1
2846 2013-05-25 12:56:31 -0.13 831.6  895.8
2847 2013-05-25 12:56:32  0.18 831.6  927.4
2865 2013-05-25 12:56:50 -0.13 955.6 2541.5

The middle two are a second apart. Based on the altitude, it looks like the plane was on the ground. Based on the engine RPMs, it looks like it was within a second or two of touchdown. Chances are that it was just nose not quite aligned with the direction of travel. The other two points are likely thermals tossing the plane about a bit — the first point is from about 50 feet above ground the last is from about 120 feet. Ok, I’m curious…

> data[c(2835:2850),c("Time","LatAc","AltB","E1RPM","GndSpd")]
                    Time LatAc  AltB  E1RPM GndSpd
2835 2013-05-25 12:56:20 -0.02 876.6 1427.9  66.71
2836 2013-05-25 12:56:21  0.01 873.6 1077.1  65.71
2837 2013-05-25 12:56:22  0.01 864.6  982.4  64.21
2838 2013-05-25 12:56:23  0.04 861.6  994.1  62.77
2839 2013-05-25 12:56:24  0.01 858.6  982.6  61.54
2840 2013-05-25 12:56:25  0.01 852.6  988.2  60.18
2841 2013-05-25 12:56:26 -0.02 845.6  959.0  58.91
2842 2013-05-25 12:56:27  0.00 846.6  945.5  57.73
2843 2013-05-25 12:56:28  0.01 844.6  930.9  56.53
2844 2013-05-25 12:56:29  0.10 834.6  908.0  55.16
2845 2013-05-25 12:56:30 -0.01 827.6  886.6  54.16
2846 2013-05-25 12:56:31 -0.13 831.6  895.8  52.71
2847 2013-05-25 12:56:32  0.18 831.6  927.4  51.49
2848 2013-05-25 12:56:33 -0.06 831.6  982.0  50.21
2849 2013-05-25 12:56:34  0.05 840.6 1494.0  49.39
2850 2013-05-25 12:56:35 -0.07 833.6 2249.7  48.76

The altitudes look a little out of whack, but otherwise it makes sense. #2835 was probably the time throttle was pulled to idle. Between #2848 and #2849 throttle went full in. Ground was most likely around 832 feet and touchdown was likely at #2846 as I guessed earlier.

Let’s plot the engine related values. First up, engine RPMs.


It is pretty boring. You can see the ~800 during taxi; the 1800 during the runup; the 2500 during takeoff; 2200 during cruise; and after 12:50 you can see the go-around, touch-n-go, and full stop.

Next up, cylinder head temperature (in °F) and exhaust gas temperature (also in °F). Since the plane has a 4 cylinder engine, there are four lines on each graph. As I was maneuvering most of the time, I did not get a chance to try to lean the engine. On a cross country, it be pretty interesting to see the temperature go up as a result of leaning.



Moving on, let’s look at fuel consumption.

fuel quantity

This is really weird. For the longest time, I knew that the plane used more fuel from the left tank, but this is the first time I have solid evidence. (Yes, the fuel selector was on “Both”.) The fuel flow graph is rather boring — it very closely resembles the RPM graph.

fuel flow

Ok, two more engine related plots.

oil temperature

oil pressure

It is mildly interesting that the temperature never really goes down while the pressure seems to be correlated with the RPMs.

There are two variables with the vertical speed — one is GPS based while the other is barometer based.

vertical speed

As you can see, the two appear to be very similar. Let’s take a look at the delta. In addition to just a plain old subtraction, you can see the 60-second moving average.

vertical speed: GPS vs. Barometer

Not very interesting. Even though the two sometimes are off by as much as 560 feet/minute, the differences are very short-lived. Furthermore, the differences are pretty well distributed with half of them being within 50 feet.

> summary(data$VSpd - data$VSpdG)
     Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
-559.8000  -49.2800    0.4950    0.8252   53.0600  563.4000 
> summary(SMA(data$VSpd - data$VSpdG),2)
     Min.   1st Qu.    Median      Mean   3rd Qu.      Max.      NA's 
-240.2000  -22.2200    0.6940    0.8226   25.4700  226.7000         9 

Ok, last but not least the CSV contains the pitch and roll angles. I’ll have to think about what sort of creative analysis I can do. The only thing that jumps to mind is the mediocre S-turn around 12:40 where the roll changed from about 20 degrees to -25 degrees.



I completely ignored the volts and amps variables (for each of the two busses), all the navigation related variables (waypoint identifier, bearing, and distance, Wikipedia article: HSI source, course, Wikipedia article: CDI/Wikipedia article: GS deflection), wind (direction and speed), as well as ground track, magnetic heading and Wikipedia article: variation, GPS fix (it was always 3D), GPS horizontal/vertical alert limit, and WAAS GPS horizontal/vertical protection level (I don’t think the avionics can handle WAAS — the columns were always empty). Additionally, since I wasn’t using the autopilot, a number of the fields are blank (Autopilot On/Off, mode, commands).


A while ago I learned about CloudAhoy. Their iPhone/iPad app uses the GPS to record your flight. Then, they do some number crunching to figure out what kind of maneuvers you were doing. (I contacted them a while ago to see if one could upload a GPS trace instead of using their app, sadly it was not possible. I do not know if that has changed since.) I think it’d be kind of cool to write a (R?) script that’d take the G1000 recording and do similar analysis. The big difference is the ability to use the great number of other variables to evaluate the pilot’s control of the airplane — ranging from coordinated flight and dangerous maneuvers (banking too aggressively while slow), to “did you forget to lean?”.

Update (2016-10-10): Out of the blue, I got an email from the CloudAhoy guys letting me know that a lot has changed since I wrote this post three and a half years ago and that they support uploading of lots of different flight data file formats. They seem to have some very interesting ways of visualizing the data. I think I’ll have to play with it in the near future.

Plotting with ggmap

Recently, I came across ggmap package for R. It supposedly makes for some very easy plotting on top of Google Maps or OpenStreetMap. I grabbed a GPS recording I had laying around, and gave it a try.

You may recall my previous attempts at plotting GPS data. This time, the data file I was using was recorded with a USB GPS dongle. The data is much nicer than what a cheap smartphone GPS could produce.

> head(pts)
        time   ept      lat       lon   alt   epx    epy mode
1 1357826674 0.005 42.22712 -83.75227 297.7 9.436 12.755    3
2 1357826675 0.005 42.22712 -83.75227 297.9 9.436 12.755    3
3 1357826676 0.005 42.22712 -83.75227 298.1 9.436 12.755    3
4 1357826677 0.005 42.22712 -83.75227 298.4 9.436 12.755    3
5 1357826678 0.005 42.22712 -83.75227 298.6 9.436 12.755    3
6 1357826679 0.005 42.22712 -83.75227 298.8 9.436 12.755    3

For this test, I used only the latitude, longitude, and altitude columns. Since the altitude is in meters, I multiplied it by 3.2 to get a rough altitude in feet. Since the data file is long and goes all over, I truncated it to only the last 33 minutes.

The magical function is the get_map function. You feed it a location, a zoom level, and the type of map and it returns the image. Once you have the map data, you can use it with the ggmap function to make a plot. ggmap behaves a lot like ggplot2’s ggplot function and so I felt right at home.

Since the data I am trying to plot is a sequence of latitude and longitude observations, I’m going to use the geom_path function to plot them. Using geom_line would not produce a path since it reorders the data points. Second, I’m plotting the altitude as the color.

Here are the resulting images:

If you are wondering why the line doesn’t follow any roads… Roads? Where we’re going, we don’t need roads. (Hint: flying)

Here’s the entire script to get the plots:

#!/usr/bin/env Rscript


pts <- read.csv("gps.csv")

/* get the bounding box... left, bottom, right, top */
loc <- c(min(pts$lon), min(pts$lat), max(pts$lon), max(pts$lat))

for (type in c("roadmap","hybrid","terrain")) {
	map <- get_map(location=loc, zoom=13, maptype=type)
	p <- ggmap(map) + geom_path(aes(x=lon, y=lat, color=alt*3.2), data=pts)

	jpeg(paste(type, "-preview.jpg", sep=""), width=600, height=600)

	jpeg(paste(type, ".jpg", sep=""), width=1024, height=1024)

P.S. If you are going to use any of the maps for anything, you better read the terms of service.


I got myself a bike in early May. It has been a long time since the last time I biked, but Holly and I thought that it would be fun to get bikes and ride around. I am a geek and so I have to do some stats tracking. Data collection is pretty easy with a phone since it has a (mediocre) GPS. Here are my findings about various ways to keep track of the data collected while biking.

Map My Ride

While at the bike store, we were told about Map My Ride. When I got home, I checked it out. I liked the elevation over time (for workouts) and over distance (for tracks), the per-week summary conveniently attached right next to the week in the monthly calendar view, and the idea of a leader board for each track. I’d say the major difference between it and the other sites I tried is the split between a route and a workout. A route is just the path, while a workout is a route and timing information. At first that sounded great, but it turns out that this split is utterly useless and lead to data loss for me. (The phone app lets you ride a route and it will associate the route with your workout instead of the default “create a new route from this workout.” Sadly, if you deviate from the route, the whole system gets very confused. It still makes the association and when you view the workout everything looks right, but certain parts of the website believe that you did exactly the route — no more, no less. If you try to edit the route, it obliterates your detailed time information.) I was willing to call that a quirk of the system and work around it (by always making a new route), but then I found out that the Wikipedia article: GPX export functionality exports only the route and not the workout. In other words, it does not let you get the detailed time information out of it. That is a deal breaker. I searched and it turns out that other people “complained” about this before. While I think that many of the people there had the wrong attitude, I couldn’t help but notice that it took the staff over two and a half months to respond with a essentially useless comment. Ow! Since I liked the interface enough, I thought that maybe I could live with using My Tracks on my phone to record the data, export it as GPX, and then import it on the web.

Couple of days later, someone happened to post a way to get a JSON encoded latitude, longitude, elevation, and time for every workout. I quickly grabbed the data, and whipped up a small Python script to spit out a minimal GPX file.

Custom Software

Initially, I resisted the temptation to write my own custom software to read in the GPS data. When I found out that Map My Ride did not let you export time information, I thought to myself how hard could it be? I had some GPX files I created using My Tracks, so I went and hacked together a quick script which spit out a 100 kB HTML file that used HighCharts to display them. A few observations I made during this were…

  1. One has to do something smarter than dumping over 100 kB of data into a HTML file otherwise the browser will have unacceptably slow response times.
  2. Having the GPX file is a great first step, but some post processing needs to be done to account for the variable accuracy of GPS. (I haven’t experimented with better GPS receivers.) If you do not, you will invariably end up with holes in you data (I’ve seen several 30-second intervals without any location updates), and worse yet random jumps from place to place when the GPS receiver changes its mind about where you are. This results in spikes in distance traveled between two location updates leading to huge spikes in speed.
  3. I should have prototyped my algorithms in R. While running the script and refreshing the page took only a couple of seconds, python doesn’t quite lend itself to the same algorithmic tinkering that R does. (Matlab would have worked too.)

Needless to say, after about 3 hours of playing around, I more or less gave up on the idea of making my own visualization.


I first found Endomondo back in January, when I searched the Android Market Place for an application that would record my GPS location over time. Endomondo is a sports tracker. The free app is pretty simple and straight forward. I think the website is less interesting than Map My Ride’s, but that’s to be expected since Map My Ride caters to cycling. Endomondo makes it easier to find what your best kilometer (or mile) is, while Map My Ride has much better elevation graphing and analysis. While Endomondo felt a bit less feature-full, it lets you export your workouts as GPX with all 4 dimensions.

I was just about ready to switch to it fully (instead of importing My Tracks-generated GPX files), when I heard of Strava during bike-to-work week.


I thought I would give Strava a shot in case it had more features than Endomondo. I was surprised. The user interface is very clean, and it definitely caters to cyclists (even though it supports other activities). You can view the elevation, speed, and power of a ride with either distance or time on the x-axis with a single click. Strava also differentiates between time and time spent moving and understands that bikes have mass. Strava also has the concept of a “segment”. A segment is a certain path and everyone that takes that path at any point in time will be put on a leader board for others to see. It is interesting to see how you compare to other people in the area.

The Strava android app had issues with Samsung Galaxy Nexus (after you start recording, when the screen blanks as part of the power saving configuration, the app exits), but within a week or two an app update fixed the issues.


My recommendation is either Endomondo or Strava based on what you primarily care about. If you prefer many different sports, I feel that Endomondo will probably work out better for you, while if you are more of a cycling or running fiend Strava is the way to go. Then again, you can sign up for both, and decide which one you like more.

Finally, when I want to just a generic GPS track (e.g., for geo-tagging photos from my camera), I use My Tracks and then export as GPX.

Powered by blahgd