Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graph Neural Networks for Physical-Layer Security in Multi-User Flexible-Duplex Networks

Tharaka Perera, Saman Atapattu, Yuting Fang, Jamie Evans

TL;DR

This work studies physical-layer security in multi-user flexible-duplex networks with coordinated eavesdroppers. It introduces a novel system model and a graph-based representation that enables a GNN to jointly optimize transmission direction and power in an unsupervised, scalable manner, achieving higher sum secrecy rates and lower complexity than classical iterative methods. A no-eavesdrop CSI extension uses distance-based proxies to approximate channels, maintaining competitive performance. The results demonstrate FlexD’s superiority over half-duplex systems and show the GNN approach outperforms classical methods in both secrecy performance and computational efficiency, with practical implications for secure wireless design.

Abstract

This paper explores Physical-Layer Security (PLS) in Flexible Duplex (FlexD) networks, considering scenarios involving eavesdroppers. Our investigation revolves around the intricacies of the sum secrecy rate maximization problem, particularly when faced with coordinated and distributed eavesdroppers employing a Minimum Mean Square Error (MMSE) receiver. Our contributions include an iterative classical optimization solution and an unsupervised learning strategy based on Graph Neural Networks (GNNs). To the best of our knowledge, this work marks the initial exploration of GNNs for PLS applications. Additionally, we extend the GNN approach to address the absence of eavesdroppers' channel knowledge. Extensive numerical simulations highlight FlexD's superiority over Half-Duplex (HD) communications and the GNN approach's superiority over the classical method in both performance and time complexity.

Graph Neural Networks for Physical-Layer Security in Multi-User Flexible-Duplex Networks

TL;DR

This work studies physical-layer security in multi-user flexible-duplex networks with coordinated eavesdroppers. It introduces a novel system model and a graph-based representation that enables a GNN to jointly optimize transmission direction and power in an unsupervised, scalable manner, achieving higher sum secrecy rates and lower complexity than classical iterative methods. A no-eavesdrop CSI extension uses distance-based proxies to approximate channels, maintaining competitive performance. The results demonstrate FlexD’s superiority over half-duplex systems and show the GNN approach outperforms classical methods in both secrecy performance and computational efficiency, with practical implications for secure wireless design.

Abstract

This paper explores Physical-Layer Security (PLS) in Flexible Duplex (FlexD) networks, considering scenarios involving eavesdroppers. Our investigation revolves around the intricacies of the sum secrecy rate maximization problem, particularly when faced with coordinated and distributed eavesdroppers employing a Minimum Mean Square Error (MMSE) receiver. Our contributions include an iterative classical optimization solution and an unsupervised learning strategy based on Graph Neural Networks (GNNs). To the best of our knowledge, this work marks the initial exploration of GNNs for PLS applications. Additionally, we extend the GNN approach to address the absence of eavesdroppers' channel knowledge. Extensive numerical simulations highlight FlexD's superiority over Half-Duplex (HD) communications and the GNN approach's superiority over the classical method in both performance and time complexity.
Paper Structure (21 sections, 1 theorem, 13 equations, 4 figures)

This paper contains 21 sections, 1 theorem, 13 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Related work
  3. Problem statement and contribution
  4. System Model
  5. Network Model
  6. Analytical Model
  7. Optimization Problem
  8. Sum Secrecy Rate Maximization with Full CSI
  9. Classical Approach
  10. Graph Neural Network (GNN) Approach
  11. GNNs
  12. Proposed Graph Representation of the FlexD Network
  13. Proposed GNN Architecture
  14. Loss Function in the Unsupervised Learning Setting
  15. Sum Secrecy Rate Maximization without Eavesdroppers' CSI
  16. ...and 6 more sections

Key Result

Theorem 1

Given the distance matrix $\bm{D}\in\mathbb{R}^{2N\times K}$ to eavesdroppers, there exists an MLP-based approximation $f_{\operatorname{MLP}}~:~\bm{D}~\mapsto~\mathbb{C}^{2N \times K}$ such that, for any accuracy threshold $\varepsilon > 0$, the channel matrix $\bm{G}\in\mathbb{C}^{2N \times K}$ ca

Figures (4)

  • Figure 1: FlexD network with $2N$ users and $K$ coordinated eavesdroppers.
  • Figure 2: The proposed graph representation of the FlexD network.
  • Figure 3: Average sum secrecy rate (ASSR) performance comparison of GNN model, classical algorithm, Max Power, and HD communication.
  • Figure 4: (a) Average running time of algorithms with $K=2$ eavesdroppers and a variable number of users. (b) Average running time of algorithms with 2 users and a variable number of eavesdroppers. (c) Average sum secrecy rate performance comparison of GNN models with and without eavesdropper channel knowledge when $K=2$ eavesdroppers are present in the network.

Theorems & Definitions (1)

  • Theorem 1