Dr Mike Dodds from the OU’s Biology Department has been investigating the accuracy of GPS devices for ecology surveying and has helped us with testing the GPS accuracy of our phones. One of Mike’s approaches is to compare the accuracy of a range of GPS devices over a few hours when placed at a (known) fixed location. Interestingly, the accuracy of the GPS devices changes throughout the day as different satellites can be picked up by the devices. Given this variation, recording the location data from different devices for a known location over a few hours enables us to compare their accuracy.
In this case we set up two Android phones at a known location on the roof of the Biology Department’s building, namely an HTC Desire and an HTC Wildfire. Both phones used the Big Brother GPS (http://bk.gnarf.org/creativity/bigbrothergps/) application to send their GPS location to a web server once every minute. We started collecting data at around 10:45am and finished at 3:00pm. Unfortunately, the Desire dropped its data connection at 11:00am and was not restarted until 12:42pm. The Wildfire collected four and a quarter hours of data, and the Desire recorded two and half hours of data – giving us a sufficient dataset for a reasonable comparison (see Figure 1).
Figure 1. Scatterplot of error versus time for nothings and eastings measures from an HTC Desire and an HTC Wildfire phone (click on the plot for full-size image).
The above chart does not contain the first six data points, as these include moving the devices into place. The locations were recorded as latitude and longitude, and then converted to Ordinance Survey eastings and northings to aid comparison between their known and recorded locations (in meters). Both the Desire and Wildfire phones have very similar results. Generally (as with other device’s Mike has used), the difference between the known and recorded locations are greater for northings than eastings. For both phones the accuracy was better in the second half of the day, this may be due to the position of the satellites – further tests are needed. Most remarkably, the actual difference is much less than expected – ranging from just over 4 meters at the start of the day down to less than 1 meter towards the end.
For comparison Mike showed us a similar plot for a Garmin 60Cx, taken at the same location over a similar time period on the 20th of January 2011 (see Figure 2). The difference between the recorded and actual easting and northing positions appears to vary more with the Garmin than the two Android phones we tried.
Figure 2. Scatterplot of error versus time for nothings and eastings measures from a Garmin CX60 GPS (click on the plot for full-size image).
These findings certainly justify further work – first of all to repeat the fixed location measures, then to compare accuracy over a known path, and finally accuracy in more challenging environments such as forests. Understanding the accuracy characteristics of these devices will help inform how they may be used appropriately in survey or other fieldwork contexts.
A very interesting post Trevor, a thank you to Dr Mike Dodd for sharing. It is very encouraging to see that the Androids could maintain at least 4m accuracy, as you mentioned the availability of the satellites can play a large part in the results, something we also have to live with… Im quite impressed that they managed less than 1m accuracy during the test, a specification usually only boasted by higher end ‘GPS only’ hardware. Were you intending to monitor satellite availability in further tests? It would be interesting to see plots of accuracy against satellite availability.
For reference, in our own fieldwork applications in GPS assisted technology enhanced learning I found that anything around 2-3m accuracy is generally suitable for most points of interest (waypoints). I also find high density of tree canopy can be problematic with variable signal degradation. Proximity to cliff faces can sometimes cause issues with satellite accuracy, similar effects can seen in high density cityscapes, with position wandering.
Thanks Dr Mike Dodd for your research. In normal GPS units used in power system protection systems having the accuracy of -+ 5m. with my knowledge they are the most accurate GPS ever. But knowing about 2-3m accuracy on Androids amazed me..Thanks..