Exploring spatial decomposition methods and linear regression in radio environment Maps

In this paper we propose the application of spatial decomposition methods for improving the prediction accuracy of linear regression models when used for establishing radio environment maps through large-scale pathloss estimation. We discuss two basic decomposition schemes, namely square and radial decomposition, and apply them in a reference scenario for a high-power TV transmitter. Our results show that spatial decomposition can improve the residual mean square error by up to 4 dB, but also indicate that the benefit of our proposed approach is very sensitive to an appropriate selection of the decomposition scheme and its parameters.