Deep learning techniques for enhanced security and privacy in 6G terrestrial–nonterrestrial network architecture

Abstract

The seamless integration of terrestrial networks (TNs) and non-terrestrial networks (NTNs) is a cornerstone in achieving the vision of ubiquitous connectivity in 6G systems. This native integration of TN and NTN components enables global coverage, ultrareliable low-latency communications, and massive device connectivity, but it also presents significant challenges. These challenges include ensuring robust security in distributed architectures, safeguarding privacy during cross-domain data handling, and managing complex interference scenarios in shared and adjacent spectrum environments. Artificial intelligence (AI) serves as a transformative enabler in overcoming these challenges, providing advanced capabilities for real-time threat detection, adaptive authentication, privacy-preserving mechanisms, and dynamic spectrum management. By leveraging advanced techniques such as bidirectional long short-term memory (BiLSTM) networks, multi-agent reinforcement learning (MARL), and federated learning (FL), the proposed solutions enable robust anomaly detection, dynamic spectrum allocation, and scalable privacy-preserving model training. These techniques collectively improve key performance metrics, including a 30% enhancement in signal-to-interference-plus-noise ratio (SINR), over 90% spectrum efficiency, and a privacy budget of ϵ<1, while reducing latency to below 50 ms. This paper explores the vulnerabilities inherent to TN–NTN architectures, identifies limitations of traditional approaches, and examines cutting-edge AI-driven solutions tailored for integrated TN–NTN systems. Key contributions include insights into AI’s role in addressing security, privacy, and interference challenges, and the development of decentralized intelligence frameworks for scalable and efficient network operations.