A nation-wide wi-fi RSSI dataset : Statistical analysis and resulting insights.

Abstract

We present a dataset collected during ten months from a network comprising approximately 9500 double-band Access Points (APs), corresponding to Uruguay’s nation-wide one-to-one computing program’s internet provider. The dataset includes the transmission power, used channel and measured RSSI (Radio Signal Strength Indicator) that each AP senses every other AP in sight, with a granularity of an hour. This results in a total of more than 750 million measurements, one of the largest Wi-Fi datasets to date. In the study of this dataset we have first focused on a linklevel analysis. Our contributions are fourfold. We verify that approximately only half of the RSSI time-series are actually stationary, and that in that case, they present strong time correlations. Moreover, the typical assumption that the channel is symmetrical is not true, even in the long-term, and we show that interference plays an important role on this asymmetry. Finally, we study attenuation in the 5 GHZ band and show that its upper section is prone to larger attenuation than what is predicted by classic models. The practical consequences of these observations are discussed throughout the article. We also present networklevel indicators of the system (such as number of neighbors per AP and interference level). These are particularly useful for simulating a planned network such as the one discussed here.