Introducing a data-driven Self-Awareness (SA) module in Cognitive Radio (CR) can support the system to establish secure networks against various attacks from malicious users. Such users can manipulate the radio spectrum in order to make the CR learn wrong behaviours and take mistaken actions. A basic SA module includes the ability to learn generative models and detect abnormalities inside the radio spectrum. In this work, we propose and implement Artificial Intelligence (AI)-based Abnormality Detection techniques at the physical (PHY)-layer in CR enabled by learning Generative Models. Specifically, two real-world practical applications related to different data dimensionality and sampling rates are presented. The first application implements the Conditional Generative Adversarial Network (C-GAN) investigated on generalized state vectors extracted from spectrum representation samples to study the dynamic behaviour of the wideband signal. While the second application is based on learning a Dynamic Bayesian Network (DBN) model from a generalized state vector which contains sub-bands information extracted from the radio spectrum. Results show that both of the proposed methods are capable of detecting abnormal signals in the spectrum and pave the road towards Self-Aware radio.

The integration of Cognitive Radio (CR) with Unmanned Aerial Vehicles (UAVs) is an effective step towards relieving the spectrum scarcity and empowering the UAV with a high degree of intelligence. The dynamic nature of CR and the dominant line-of-sight links of UAVs poses serious security challenges and make the CR-UAV prone to a variety of attacks as malicious jamming. Joint jammer detection and automatic jammer classification is a powerful approach against the physical layer threats by identifying multiple jammers attacking the network that realize a crucial stage towards efficient interference management. This paper proposes a novel method for joint detection and automatic classification of multiple jammers attacking with different modulation schemes. The method is based on learning a representation of the radio environment encoded in a Generalized Dynamic Bayesian Network (GDBN) whilst multiple GDBN models represent various jamming signals under different modulation schemes. The CR-UAV performs multiple predictions online in parallel and evaluates multiple abnormality measurements based on a Modified Markov Jump Particle Filter (M-MJPF) to select the best-fit model that explains the detected jammer and recognize the modulation scheme accordingly. The simulated results demonstrate that the proposed GDBN-based method outperforms Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and Stacked Autoencoder (SAE) in terms of classification accuracy and achieves a higher degree of explainability of its own decisions by interpreting causes and effects at hierarchical levels under the Bayesian learning and reasoning processes.

The development of efficient and reliable trajectory-planning strategies for swarms of unmanned aerial vehicles (UAVs) is an increasingly important area of research, with applications in surveillance, search and rescue, smart agriculture, defence operations, and communication networks. This article provides a comprehensive and critical review of the various techniques available for UAV swarm trajectory planning, which can be broadly categorised into three main groups: traditional algorithms, biologically inspired metaheuristics, and modern artificial intelligence (AI)-based methods. The study examines cutting-edge research, comparing key aspects of trajectory planning, including computational efficiency, scalability, inter-UAV coordination, energy consumption, and robustness in uncertain environments. The strengths and weaknesses of these algorithms are discussed in detail, particularly in the context of collision avoidance, adaptive decision making, and the balance between centralised and decentralised control. Additionally, the review highlights hybrid frameworks that combine the global optimisation power of bio-inspired algorithms with the real-time adaptability of AI-based approaches, aiming to achieve an effective exploration–exploitation trade-off in multi-agent environments. Lastly, the article addresses the major challenges in UAV swarm trajectory planning, including multidimensional trajectory spaces, nonlinear dynamics, and real-time adaptation. It also identifies promising directions for future research. This study serves as a valuable resource for researchers, engineers, and system designers working to develop UAV swarms for real-world, integrated, intelligent, and autonomous missions.

This work proposes a novel resource allocation strategy for anti-jamming in Cognitive Radio using Active Inference (AIn), and a cognitive-UAV is employed as a case study. An Active Generalized Dynamic Bayesian Network (Active-GDBN) is proposed to represent the external environment that jointly encodes the physical signal dynamics and the dynamic interaction between UAV and jammer in the spectrum. We cast the action and planning as a Bayesian inference problem that can be solved by avoiding surprising states (minimizing abnormality) during online learning. Simulation results verify the effectiveness of the proposed AIn approach in minimizing abnormalities (maximizing rewards) and has a high convergence speed by comparing it with the conventional Frequency Hopping and Q-learning.

In this paper, we propose to introduce an emergent Self-Awareness (SA) module at the physical layer (PHY) in Cognitive Unmanned Aerial Vehicle (UAV) Radios to improve PHY security, especially against jamming attacks. SA is based on learning a hierarchical representation of the radio environment by means of a proposed Hierarchical Dynamic Bayesian Network (HDBN). It is shown how the acquired knowledge from previous experiences facilitate the radio spectrum perception and allow the radio to detect abnormal behaviours caused by jamming attacks. Detecting abnormalities realize a fundamental step towards growing up incrementally the radio’s long-term memory. Deviations from predictions estimated during abnormal situations are used to characterize jammers at multiple levels and discover their dynamic behavioural rules. Besides, a proactive consequence can be drawn after estimating the jammer’s signal to act efficiently by mitigating its effects on the received stimuli. Simulation results show that the introduction of the novel SA functionalities with the proposed HDBN framework provides the high accuracy of characterizing, detecting and predicting the jammer’s activities.

Classical imitation learning methods suffer substantially from the learning hierarchical policies when the imitative agent faces an unobserved state by the expert agent. To address these drawbacks, we propose an online active learning through active inference approach that encodes the expert’s demonstrations based on observation-action to improve the learner’s future motion prediction. For this purpose, we provide a switching Dynamic Bayesian Network based on the dynamic interaction between the expert agent and another object in its surrounding as a reference model, which we exploit to initialize an incremental probabilistic learning model. This learning model grows and matures based on the free-energy formulation and message passing of active inference dynamically at discrete and continuous levels in an online active learning phase. In this scheme, generalized states of the learning world are represented as distance-vector, where it is the learner’s observation concerning its interaction with a moving object. Considering the distance vector entail intentions, it enables action prediction evaluation in a prospective sense. We illustrate these points using simulations of driving intelligent agents. The learning agent is trained by using long-term predictions from the generative learning model to reproduce the expert’s motion while learning how to select a suitable action through new experiences. Our results affirm that a Dynamic Bayesian optimal approach provides a principled framework and outperforms conventional reinforcement learning methods. Furthermore, it endorses the general formulation of action prediction as active inference.