A statistical CSI model for indoor positioning using fingerprinting

Indoor positioning methods are still inaccurate, as they use Wi-Fi signals which are distorted by multipath propagation. A promising approach consists in using channel state information (CSI) for fingerprinting. We propose a new method for this approach. To do so, we start by building a statistical model of the CSI information of the room where positioning needs to be resolved. The key of our method lies on how we build this model. We do so by considering the room model as a realization of a random model, and building the posterior distribution of the room model given the measurements acquired during initialization. This permits obtaining an accurate interpolation of the CSI behavior of the room at points where CSI was not initially measured. Once the statistical room model is built, the position is estimated using the maximum likelihood criterion. We present experimental results showing the higher positioning accuracy of the proposed method, in comparison with other available alternatives.