Empowering V2X Networks with Cognitive and Self-aware Models for secure Radio Resource Management

Cognitive Radio (CR), originally envisioned by Mitola, augmented with embodied self-awareness (SA) and cognition through Artificial Intelligence (AI), provides a robust framework to address critical challenges in secure Radio Resource Management (RRM) for vehicular communications. In Vehicle-to-Everything (V2X) communication systems, RRM is pivotal in ensuring reliable and efficient communication among vehicles, infrastructure, and other road entities. It encompasses key components such as spectrum allocation, adaptive modulation and coding, power control, resource optimization, interference management, and security. These components collectively form the backbone of safe and reliable communication in dynamic environments, where timely and accurate information exchange significantly enhances road safety and traffic management. This thesis explores three key domains: (1) vehicular-infrastructure radio link security threat detection at the base station (BS), (2) Automatic Modulation Conversion and Classification (AMCC) as an enhancement to existing Automatic Modulation Classification (AMC) modules, and (3) network and link- level abnormality detection at the Physical (PHY) layer and victim node localization in V2X scenarios. The integration of AI-enabled CR into BS for physical layer security was analyzed, focusing on LTE-based V2I physical links compromised by a mobile jammer vehicle. Using a Bayesian filtering approach, the SA framework facilitates utilization of knowledge gained from prior experiences, effectively enabling radio spectrum perception and real-time detec- tion of jamming attacks within practical SNR thresholds. Furthermore, the implementation of the AMCC module complemented RRM with a data-driven SA module at the physical layer. This solution allows AI-enabled radios to dynamically adapt between lower and higher-order modulation schemes based on ambient conditions, integrating transport planning (TP) to guarantee secure data transmission and optimal data rates for the Internet of Vehicles (IoVs). The AMCC module is validated through a real-time automatic modulation classification case study and is capable of further allowing spectrum-efficient data relay services to the surrounding radio infrastructure while ensuring error-free communications. Followed this, a novel brain-inspired SA framework was developed to enable intelligent radios at the BS to learn computationally efficient hierarchical representations of the environment and incrementally update its respective long-term memory, thereby provisioning cognitive services as envisioned for 6G networks. An interactive Bayesian Generative model is devised for V2X network and link level (PHY-layer) abnormality detection scheme. The model leverages semantics embedded in coupled Multi-Generalized Dynamic Bayesian Networks (M-GDBN) to monitor and secure IoV links in real-time. High detection probabilities were achieved by matching predicted with observed network connectivity maps. Additionally, link-level disruptions were tracked, and victim nodes of attacks, such as jamming and spoofing, were localized within the cell served by the cognitive BS. The reasoning capabilities were effectuated via messages across the hierarchical levels of the coupled M-GDBN structure. Moreover, experimental results demonstrate superior performance in the online detection and localization of attacked nodes and link-level abnormalities in V2X networks of varying sizes. The proposed framework under high variability is shown to outperform state-of-the-art deep Auto-encoder-based anomaly detection, which also incorporates similar latent hierarchies within the data. Evaluated scenarios include multiple jamming attacks targeting different vehicles within the network, inter-vehicular connectivity variations from unobserved abnormal channel characteristics, and spoofing events. Insights from the detected anomalies may inform and guide mitigation actions and facilitate updates to the CR’s autobiographical memory through new model learning and meta-state labels where needed. The studies in this thesis conclude that the use of graph-based probabilistic representation learning, inspired by the Bayesian Brain hypothesis, offers a causal, interpretable, and explainable framework for PHY-layer anomaly detection in V2X systems. This approach, leveraging Bayesian generative models, demonstrated clear advantages over deep learning techniques, particularly in explainability and causality. Therefore, these findings contribute to advancing secure and efficient communication in V2X networks, setting a foundation for future developments in intelligent vehicular systems.