A Reconstruction-Inversion Method of Atmospheric Duct Parameters Using ADS-B Signals Based on Autoencoder and ResNet18

Low-altitude atmospheric ducts are anomalous atmospheric phenomena in the troposphere, significantly affecting the operation of microwave or ultrashort-wave radio systems. Therefore, an accurate operation crucially relies on the real-time acquisition of these atmospheric duct parameters. Based on Automatic Dependent Surveillance-Broadcast (ADS-B) signals and deep neural networks, this paper initiates a reconstruction-inversion method. ADS-B signals are for the first time applied to the inversion of atmospheric duct parameters. First, the PL field is reconstructed utilizing the Propagation Loss Field Reconstruction Network (PLFRN) with an autoencoder architecture. Then, the inversion of the duct parameters is conducted by means of the Duct Parameter Inversion Network (DPIN) based on ResNet18. According to experimental results in actual maritime scenarios, the Mean Absolute Percentage Error (MAPE) of less than 13% is obtained between the inverted duct parameters and the measured parameters. In addition, the MAPE between the reconstructed PL field and the results calculated by the traditional Parabolic Equation (PE) method is only 3%. The algorithm accuracy is well verified in its applications and points out greater future potentials.