Multi-Agent-Driven Cognitive Secure Communications in Satellite-Terrestrial Networks

TL;DR

This work tackles secure communication in STNs under intelligent eavesdroppers by formulating an NP-hard optimization that maximizes secrecy probability $P_s$ while enforcing reliable transmission probability $P_u$. It introduces a two-layer defense: a Foundation Layer using centralized training with multi-agent DRL (MADRL) for time-frequency scheduling ($\mathbf{S}$ and $\mathbf{X}$), and a Protection Layer using GANs to produce adversarial matrices $\mathbf{A}$ plus learning-aided power control $\mathbf{p}$ to degrade eavesdropper inference. The approach is implemented via CTDE-MADRL, a Wasserstein GAN with gradient penalty to align adversarial patterns, and DDQN-based power control, with performance gains shown against AN-based FH, game-theoretic, and GAN-based baselines. Results demonstrate higher SP and lower power overhead across varying reliability targets, UE counts, and eavesdropper capabilities, highlighting the method’s practical potential for cognitive secure communications in heterogeneous STNs.

Abstract

Satellite-terrestrial networks (STNs) have emerged as a promising architecture for providing seamless wireless coverage and connectivity for multiple users. However, potential malicious eavesdroppers pose a serious threat to the private information via STNs due to their non-cooperative behavior and ability to launch intelligent attacks. To address this challenge, we propose a cognitive secure communication framework driven by multiple agents that coordinates spectrum scheduling and protection through real-time sensing, thereby disrupting the judgment of eavesdroppers while preserving reliable data transmission. On this basis, we formulate an optimization problem to maximize the secrecy probability of legitimate users, subject to a reliable transmission probability threshold. To tackle this problem, we propose a two-layer coordinated defense system. First, we develop a foundation layer based on multi-agent coordination schedule to determine the satellite operation matrix and the frequency slot occupation matrices, aiming to mitigate spectrum congestion and enhance transmission reliability. Then, we exploit generative adversarial networks to produce adversarial matrices, and employ learning-aided power control to set real and adversarial signal powers for protection layer, which actively degrades the inference capability of eavesdroppers. Simulation results demonstrate that the proposed method outperforms benchmark methods in terms of enhancing security performance and reducing power overhead for STNs in the cognitive secure communication scenario.

Figures (8)

• Figure 1: Illustration of the STNs. SAT$^1$: LEO satellite $1$; SAT$^n$: LEO satellite $n$; BS$^1$: base station $1$; BS$^m$: base station $m$; sUE$^1_1$: the legitimate UE $1$ associated with the SAT$^1$; sUE$^1_2$: the legitimate UE $2$ associated with the SAT$^1$; sUE$^n_3$: the legitimate UE $3$ associated with the SAT$^n$; tUE$^1_4$: the legitimate UE $4$ associated with the BS$^1$; tUE$^m_5$: the legitimate UE $5$ associated with the BS$^m$.
• Figure 2: Illustration of the communication channel. UE$_1$: the legitimate UE $1$; UE$_2$: the legitimate UE $2$; UE$_k$: the legitimate UE $k$; and they are associated with the same node.
• Figure 3: Comparison between the simulation and theoretical results on (a) satellite links (b) terrestrial links.
• Figure 4: Illustration of the proposed multi-agent-driven decision-making framework. Our method builds a two-layer defense system. On the left, the foundation layer employs MADRL to decide the matrix $\bf{S}$ and $\bf{X}$. The protection layer has two parts. First, given $\bf{S}$ and $\bf{X}$, GANs are trained to produce an adversarial matrix $\bf{A}$. Second, with $\bf{S}$, $\bf{X}$ and $\bf{A}$ fixed, we determine the optimal power within the feasible region. The red arrows can illustrate the flow of training data across modules.
• Figure 5: Performance comparisons vs. different reliability.