The use of millimeter waves and frequencies in the THz band paves the way for an accurate sensing of the electromagnetic environment, i.e., the localization of passive objects (such as humans, cars, and other static and moving objects) that reflect electromagnetic signals emitted by transmitters (e.g., user equipments) and captured by receivers (e.g., the base station). Such sensing capabilities make the 6G network very similar to a radar system.

Security is relevant in any communication system, and new evolutions of security call for a physical layer security approach, where the radio signals are used to obtain security mechanisms. In this thesis the focus will be on the authentication of users, i.e., the verification that a message received by the base station and presumably coming from the user A, is truly coming from user A, and thus it is authentic. An attacker user B instead aims at transmitting messages to the base station impersonating user A, i.e., making the base station believe that the message is coming from A.

To detect such attack, exploiting the sensing capabilities, when receiving a message, the base station can estimate the propagation channel from the received signal and compare it with the propagation environment sensed before (or using other transmit sources). In fact, the propagation of signals transmitted by user A are influenced by the same objects that are sensed using the base station antennas as a radar. However, checking the compatibility of the sensing environment and the propagation characteristics is not trivial, since both the sensing and the propagation estimation procedures are subject to noise but also to changes in the environment over time. Therefore, machine learning solutions that learn how to perform the comparison taking into account the several imperfections can be very efficient to detect the attacks. In such detection process, the temporal evolution of the propagation environment could also taken into account when designing the machine learning model.

The thesis will first understand the main features of the sensing and propagation estimation procedures, and then propose a machine learning solution to compare the to information and detect possible attacks. The solution will then be tested on a suitably obtained database and using python software.

For more information on this topic, please contact Prof. Stefano Tomasin (stefano.tomasin@unipd.it)