A deep learning approach for LoS/NLoS identification via PRACH in UAV-assisted public safety networks

The high mobility of Unmanned Aerial Vehicles (UAVs) and their capability to rapidly deploy Aerial Base Stations (ABS) in areas where the terrestrial network becomes unavailable is a key enabler for Public Safety Networks. In our work we introduce a model in order to identify Line of Sight (LoS) and Non-Line of Sight (NLoS) conditions for User Equipments (UEs) that attempt a connection to an ABS through the Physical Random Access Channel (PRACH) based on Convolutional Neural Networks (CNNs). Our method limits the number of antennas employed with respect to other methods that were developed for traditional approaches, while achieving higher than 80% accuracy for SNR of -20 dB. Finally, we study the impact of UAV's height on the accuracy of our method and we compare it with typical computationally efficient methods based on the delay spread with and without the aid of beamforming.