Reliability and Validity of Wi-Fi Visualization

This image was produced as part of reliability and validity tests to answer the question: Do the standard, inexpensive tools widely available for the visualization of unlicensed spectrum actually produce valid results? The answer was no. Siddhartha Raja and Christian Sandvig observed a network with a known location using kismet. These pictures are part of an effort to assess the error in kismet/gpsmap's visualizations of the network. We found that convex hulls were remarkably more accurate at estimating coverage areas than the standard method (standard distance around the mean center), and that in some situations estimated mean centers could be dramatically affected by the driving path of the observer collecting the data. However, even covex hulls produce reports of large areas of coverage that were not reliably measured in repeated attempts.

In the above image, ten attempts were made to locate a Wi-Fi access point. Each attempt is visualized by a light gray semi-transparent box. When there is agreement between attempts as to the location of the Wi-Fi access point, the box is darker. The actual location of the point was then plotted using a cross. Note that the actual location is correct (the "X" is within all ten boxes -- the darkest area), but notice that the area of coverage estimated in each attempt varies dramatically.

For research on this topic, see How do we decide what the electromagnetic spectrum looks like?

--Christian Sandvig

Downloads:

• all networks, multiple observations
• one network, multiple observations
• reliability/validity of standard distance vs. convex hull