Geo-statistical analysis of wireless spectrum occupancy using extreme value theory

In this paper, we present preliminary investigations into modelling the geographical variation of spectrum utilization. While spectrum occupancy has typically been characterized by a parameterized subset of probability distributions, through extreme value theory (EVT) the tails of non-degenerate distributions can be estimated by the generalized pareto distribution. Spectrum usage parameters were determined from data collected between 88 MHz-2686 MHz at twenty geographically separated sites across four mid-sized cities located within the United States. EVT is then used to approximate the tail distributions for the 20 sites, illustrating the suitability of the generalized pareto distribution. The relationship between overall spectrum occupancy and the tail estimates is then investigated.