Error analysis for fingerprint-based localization

In this paper, we derive the theoretical error Probability Density Function (PDF) and Region of Confidence (RoC) conditioned on the on-line signal parameter vector, for a generalized fingerprint-based localization system. As the computations of these terms require the exact expression of the joint PDF for both the device location and the on-line signal parameter vector, which is often not available practically, we propose to approximate this joint PDF by Nonparametric Kernel Density Estimation techniques using the training fingerprints.