Impact of Model Uncertainties on the Accuracy of Spatial Interpolation Based Coverage Estimation

Improved propagation prediction has become important due to emerging 5G and ultra-dense wireless networks. One important aspect in this domain is to understand and model better the spatial correlations of the underlying shadowing fields. In this work, we investigate the sources of uncertainty on the predictions made by employing the optimal linear predictor, namely \emph{kriging}. We show that kriging is robust to the estimation errors of the spatial correlation structure. Moreover, we show that the amount of the training data (known data points) has great impact on the prediction error. Our results help to quantify the trade off between number of data points collected and accuracy that can be reached by interpolation. The results especially help to design and optimize expensive measurement and test drive campaigns. Another contribution of the paper is to explicitly show the capabilities of kriging for coverage prediction.