Abstract:
Wirelessly connected devices play a vital role in people's daily life, especially with the rapid increase in the number of devices connected to the internet and the huge data being generated everyday. However, the open nature of the wireless channels makes them vulnerable to several threats. Two major threats are: eavesdropping attacks in which an adversary side tries to capture the transmitted data between two communicating parties, and jamming attacks in which an adversary side tries to disrupt the reception of useful signals by a legitimate receiver. The concept of physical layer security is gaining more and more attention from the research community nowadays. It makes use of the open nature and randomness of the wireless channel to achieve security in the communications system. In this thesis, the problem of combating passive eavesdroppers and jammers is studied. Multiple different scenarios are considered and security solutions based on the
physical layer are proposed. The proposed solutions are based on the usage of massive planar antenna arrays. First, a solution for combating a single passive eavesdropper at a known location is proposed. It is based on dedicating some antenna sub-arrays for the communication link between the legitimate parties, while simultaneously sending jamming signals towards the eavesdropper. Next, a look-up table-based physical layer solution is proposed to perform anti-jamming against a single jammer at a known location. The solution is based on performing beamforming by maximizing the receiver's gain towards the transmitter, and simultaneously placing a null in the direction of the jammer, thus, achieving a high signal-to-interference-plus-noise-ratio (SINR). Finally, the proposed anti-jamming technique is extended, and machine learning (ML) and deep learning (DL) algorithms are deployed in order to build a robust, and cognitive anti-jamming system. The dataset on which the models were trained and tested was generated, and efficient anti-jamming performance was achieved. The proposed models are scalable and can be extended to scenarios with multiple jammers. Insights for an effective method for detecting the location of the jammer are drawn from the proposed ML/DL models and left for future investigation.