SICNN: Spatial Interpolation with Convolutional Neural Networks for Radio Environment Mapping

This paper addresses the spatial interpolation problem in measurement-based radio environment estimation. For accurate interpolation, we need to extract global and local radio features, i.e., path loss and shadowing features. To this end, we propose a novel convolutional neural network structure, referred to as SICNN, for interpolation problems. We also investigate the impact of kernel design in SICNN on the interpolation accuracy to demonstrate that there exists an optimal kernel size. Extensive simulations show that SICNN learns the global and local features to construct an accurate radio environment map from a few measurement data.