/RadioUNet

Convolutional neural network for estimating radio maps in urban environments

Primary LanguageJupyter NotebookMIT LicenseMIT

RadioUNet

RadioUNet is a highly efficient and very accurate method for estimating the propagation pathloss from a point x to all points y on the 2D plane, in realistic propagation environments characterized by the presence of buildings. RadioUNet generates pathloss estimations that are very close to estimations given by physical simulation, but much faster.

For more information see the paper RadioUNet: Fast Radio Map Estimation with Convolutional Neural Networks.

Usage Examples

Download and extract the RadioMapSeer dataset to the folder of the Jupyter Notebooks.

For training without samples see RadioWNet_c_DPM_Thr2.ipynb.

For training with measurements and perturbed city map see RadioWNet_s_randSim_miss4build_Thr2.ipynb.

For training with simulated cars, measurements, and input car locations, see RadioWNet_s_DPMcars_carInput_Thr2.ipynb.