Practical Robust Localization over Large-Scale 802.11 Wireless networks (MobiCom 2004)

Localize by office. Collected two minutes of data per office. Using half the data are able to predict the correct location in over 95% of attempts. Supports both static localization and tracking.

Use Naive Bayes. Use both a Gaussian fit sensor model and a Histogram sensor model.

The signal space has 33 dimensions (# of APs)

Experiments: Chose 5 scans at random and removed from the training set. Repeated it for 100 times.

Show that there is a linear way to callibrate signal strength from two different devices.