UN APPROCCIO DEEP LEARNING PER IL RILEVAMENTO DI ATTACCHI DI JAMMING CONTRO I VEICOLI AEREI SENZA EQUIPAGGIO

In the last years the use of Unmanned Aerial Vehicle (UAV), or simply drones, is significantly increased: they are are becoming very popu- lar enabling technology for several applications including monitoring, surveillance and shipping. Thanks to their extreme versatility drones are used in very different sectors ranging from agriculture, environ- ment, military but also civil with various models of drones for amateur use. UAVs offer an avenue for businesses to become more efficient in the way they work. As the drones use increases, are increasing the number of enemies too. One of the drone’s worst enemies is interference radio jamming attacks; in a civil scenario, these are interferences provided by other wireless communication, in a military scenario these are interferences provided by devices specially designed to disrupt communications, and known as jammer. In this thesis, we are going to investigate an aspect regarding the drone security linked to the unreliability of the wireless medium and we will provide a high-level description for a new Risk-based Access Control (AC) system. Risk-based AC Model is one of the most dynamic meth- ods that utilize the security risk value related to each access request as a criterion to determine access decisions. Indeed, such a protocol has the advantage of being extremely flexible and adaptable to the context. In particular, in the first part of this thesis, we tackle the problem of detecting interference produced by other Bluetooth and/or Wi-Fi signals, in the drone-to-controller communication with a Deep Learning (DL) approach, specifically through the use of a Convolutional Neural Network (CNN). In the end part, we go to analyze the Risk-Adaptable Access Control (RAdAC) and provide a high-level architectural design.