Deep Learning-Based Secrecy Performance of UAV-IRS NOMA Systems With Friendly Jamming

The Internet of Things (IoT) landscape is rapidly evolving driven by recent advances in key emerging technologies. In this paper, we explore the integration of the intelligent reflecting surface (IRS), non-orthogonal multiple access (NOMA), and unmanned aerial vehicle (UAV) to enhance spectral and energy efficiencies, as well as wireless connectivity in IoT applications. However, such applications are vulnerable to eavesdropping threats that can compromise both their integrity and privacy. To circumvent this, we incorporate a friendly jamming technique into the proposed UAV-borne IRS NOMA system. Moreover, we analyze the secrecy performance of the underlying system with friendly jamming by deriving analytical expressions of the and the secrecy outage probability (SOP) and the probability of strictly positive secrecy capacity. In addition, we obtain explicit and simplified expressions to approximate the SOP at high signal-to-noise ratio values and to highlight useful insights of various parameters into the system design. Subsequently, the accuracy of the proposed theoretical framework is validated through comprehensive Monte Carlo simulations. We also propose an algorithm that determines an optimal power allocation. Finally, a fully optimized deep neural network model is developed to predict the SOP under dynamic conditions. Numerical results demonstrate the efficacy of the proposed jamming-enabled system over its non-jamming counterpart. Specifically, it shows that the proposed system outperforms its non-jamming counterpart in terms of SOP by about 72%, and the runtime of the deep neural network model is 5.7 times faster than Monte Carlo simulations while maintaining a good accuracy.