Geolocation of Internet hosts: Accuracy limits through Cramér-Rao lower bound

With active IP geolocation, the position of an Internet host is estimated by measuring the network delay from a number of other hosts with known position (usually called landmarks). In particular, after having converted delays into distances, geometrical techniques like trilateration are used to provide the estimated position on a global reference system. In this paper, we derive the Cramér-Rao lower bound (CRLB) of IP geolocation. The CRLB defines a bound on the minimum mean squared error that affects any unbiased estimator. From a practical point of view, the CRLB provides insights about the maximal theoretical accuracy that can be achieved by IP geolocalization methods. The CRLB also provides conceptual tools useful to understand how the position of landmarks and their distribution affect localization performance. Results show that to obtain accuracy levels in the order of a few tens of kilometers, the number of landmarks to be involved can be relevant and/or their distance from the target cannot be too large.