Correlated shadow-fading in wireless networks and its effect on call dropping

We discuss a statistical model to generate correlated shadow-fading patterns for wireless systems in the absence of detailed propagation and landscape information. The currently available autocorrelation models result in anomalous effects that depend on traffic density and mobility, as they propose independent random processes for each mobile. Our approach involves generating a pre-computed fading map with the right marginal distributions and spatial correlations, which avoids inconsistencies such as providing widely differing values for mobiles close to each other. The correlations are introduced via a Gaussian random field, which has a covariance structure that depends on a set of parameters which can be computed from local measurements. The model is efficiently implemented using standard linear-algebra methods. We conclude by describing a simulation experiment to study the effect of correlations on call dropping. The experiment reveals a strong relationship between call dropping and the correlation length of the fading pattern, and indicates circumstances under which dropping may be relatively high.