Physical-Layer Security and Privacy for Future Wireless Systems: From Signal Processing to Artificial Intelligence

Future wireless systems are expected to support not only high-rate communications but also sensing and intelligent services, raising growing concerns regarding the security and privacy of in- formation carried by or embedded in wireless signals. This dissertation investigates physical-layer security and privacy from two complementary perspectives: signal-processing-based protection and artificial-intelligence-enabled wireless design. First, the security of directional modulation implemented by time-modulated arrays (TMAs) with orthogonal frequency-division multiplexing (OFDM) waveforms is investigated from the eavesdropper’s perspective. The TMA-induced scrambling is formulated as an independent com- ponent analysis (ICA) problem, and a constant-modulus ICA method, together with a signal- structure-aware ambiguity-resolution scheme, is developed to recover the scrambled information. The resulting analysis reveals potential vulnerabilities and motivates corresponding mechanisms for strengthening TMA-based security. The dissertation then extends physical-layer information protection from communication security to sensing privacy. For radar-based vital sign monitor- ing, a nonlinear frequency-modulated continuous-wave waveform is designed to protect sensitive physiological information while enabling an authorized receiver to recover the desired vital signs. The second part of the dissertation investigates artificial intelligence for secure wireless sys- ii tem design. A generative flow network (GFlowNet)-based framework is developed to optimize the phase and time-modulation parameters of a time-modulated intelligent reflecting surface for integrated sensing and communication security. By learning to generate high-reward configura- tions, the proposed framework provides an efficient approach to exploring the large design space while jointly considering communication, sensing, and security objectives. To further address user mobility and environmental variation, a Meta-GFlowNet framework is developed by integrating meta-learning with GFlowNet-based optimization, enabling rapid adaptation to previously unseen directional-modulation tasks without retraining the model from scratch. Together, these contributions demonstrate how physical-layer security and privacy can be en- hanced through both careful exploitation of wireless signal structures and increasingly intelligent design methodologies, providing a progression from conventional signal processing to generative and adaptive artificial intelligence for future wireless systems.