RF-based Direction Finding of Small Unmanned Aerial Vehicles Using Deep Neural Networks

This paper presents a sparse denoising autoencoder-based deep neural network (SDAE-DNN) for the direction estimation of small unmanned aerial vehicles. It is motivated by practical challenges associated with classical direction finding algorithms such as MUSIC and ESPRIT. The proposed scheme is robust and low-complex in the sense that phase synchronization mechanism, calibration procedure, or any knowledge about the antenna radiation pattern is not required. Also, the direction estimation can be accomplished using a single-channel implementation. The paper validates the proposed architecture experimentally as well.