Air-to-ground path loss prediction using ray tracing and measurement data jointly driven DNN

With the wide application of unmanned aerial vehicle (UAV), air-to-ground (A2G) channel characterization is important for efficient and stable UAV-related communications. In this paper, a novel deep neural network (DNN) based path loss (PL) prediction model is presented for A2G communications. The new model considers the effects of path delay, carrier frequency, and reflection angle (RA) of non-line-of-sight (NLoS) paths. Specifically, a measurement-optimization-matrix (MOM) based DNN with multiple input neurons is designed. In order to solve the problem of insufficient measurement data, the new DNN is divided into two parts, i.e., initial-trained network and optimized network. The massive ray tracing (RT) simulation data is used to initially train the network and a little measurement data is used to further optimize the network. Finally, the proposed model is simulated and validated under a campus scenario at 2 GHz, 26 GHz, and 39 GHz, respectively. The predicted results are in good agreement with the validation set of RT simulation and measurement data. Moreover, our proposed model is consistent with the close-in (CI) model and 3rd Generation Partnership Project (3GPP) model under traditional terrestrial scenarios with small RAs, but it is also applicable for A2G scenarios with all frequency bands and large RAs.