A Boltzmann machine to model the multipath environment for particle filtering based GPS navigation

In GPS navigation, an accuracy of about 10 m can be currently obtained, but this performance can be strongly degraded in a multipath environment. The multipath directly impact the distribution of the additive noise corrupting the distance measurements between the satellites and the GPS receiver. They are modeled either as variance jumps if there is a direct path between the satellites and the receiver or as mean-value jumps otherwise. The originality of our approach is to take into account the spatial dependencies between GPS measurements when modeling multipath occurrences. Indeed, if two signals from satellites have close directions of arrival, they are very likely to be simultaneously degraded by multipath. For that purpose, we suggest using a Boltzmann machine which provides a “natural” setting to define interactions between the GPS measurements. Then, as the proposed model is strongly non linear and non Gaussian, we jointly estimate the mobile location and perform the multipath detection/estimation by using particle filtering.