MAP RSS based joint localization and model identification under spatially autocorrelated noise

This paper addresses the problem of model-based source localization using spatially autocorrelated received signal strength (RSS) measurements when the model parameters are not known a priori. This combined problem arises typically in situations in which a large number of observations are collected in positions close to each other in unknown environments. In our approach we model the RSS as a spatial Gaussian process characterized by an autocorrelation function. The position is then estimated by maximizing the log-likelihood after a previous analytical optimization over some of the model parameters. This technique allows the estimation of position without prior knowledge of these parameters. Although very convenient, we show experimentally that this method can be very sensitive to the geometry of the problem. As a solution we propose maximum a posteriori (MAP) estimation with indirect priors on the parameters affecting the mean of the predictions. Using data gathered from three different environments we demonstrate the effectiveness of the approach in real scenarios. On average, the localization errors achieved are 0.53, 0.99 and 0.86 m in a basketball field, a lobby and an office respectively.