Using STAR-IRS to Secure Indoor Communications Through Symbol-Level Random Phase Modulation

Basic Information

• Paper ID: 2510.11925
• Title: Using STAR-IRS to Secure Indoor Communications Through Symbol-Level Random Phase Modulation
• Authors: Yanan Du, Zeyang Sun, Yilan Zhang, Sai Xu, Beiyuan Liu
• Category: eess.SP (Signal Processing)
• Publication Date: October 13, 2024
• Paper Link: https://arxiv.org/abs/2510.11925

Abstract

This paper proposes a secure indoor communication scheme based on Simultaneous Transmitting and Reflecting Intelligent Reflecting Surface (STAR-IRS). Specifically, a transmitter (Alice) sends confidential information to an indoor target user (Bob), while multiple eavesdroppers (Eves) lurk outdoors. To protect the transmission against eavesdropping, STAR-IRS is deployed on walls or windows. Electromagnetic waves incident on STAR-IRS are dynamically divided into two components, enabling surface transmission and reflection. The reflected signal is controlled to enhance Bob's reception, while the transmitted signal reduces the signal quality for Eves through symbol-level random phase modulation. Based on this setup, a secrecy rate maximization problem is formulated. To solve this problem, a Graph Neural Network (GNN)-based approach is developed. Additionally, a Field-Programmable Gate Array (FPGA)-based GNN accelerator is designed to reduce computational latency. Simulation results demonstrate that the proposed strategy outperforms conventional schemes and pure reflection schemes in terms of secrecy performance.

Research Background and Motivation

Problem Background

1. Surging 6G Network Demands: Sixth-generation networks are expected to bring unprecedented 1000-fold growth in data traffic, driven primarily by data-intensive applications such as massive machine-type communications, virtual reality, and Internet of Things devices.
2. Indoor Communication Security Challenges: Since most wireless data transmission occurs in indoor environments, the inherent openness of wireless propagation makes indoor transmission particularly vulnerable to interception by external eavesdroppers, posing significant risks to information confidentiality.
3. Limitations of Existing Methods:
   • Traditional physical layer security (PLS) strategies such as beamforming and artificial noise are primarily designed for general scenarios
   • Few studies focus on external eavesdropping through physical barriers (such as walls or windows)
   • Existing research on STAR-IRS-assisted indoor communications is relatively limited

Research Motivation

STAR-IRS, as a sophisticated planar metasurface, can simultaneously handle reflection and transmission of incident electromagnetic energy, providing additional degrees of freedom for full-space coverage. STAR-IRS deployed on walls or windows in indoor venues can simultaneously affect indoor and outdoor users, making it a promising solution for addressing potential outdoor eavesdroppers.

Core Contributions

1. First Proposal of Symbol-Level Random Phase Modulation-Based STAR-IRS Security Strategy: Intelligently divides incident signals into two components: information-bearing signals reflected to indoor target users and transmitted artificial noise signals for degrading the performance of potential outdoor eavesdroppers.
2. Formulation of Secrecy Rate Maximization Optimization Problem: Mathematically models the secrecy rate maximization problem under total power budget constraints and reflection/transmission coefficient constraints.
3. Development of GNN-Based Deep Learning Algorithm: Designs a GNN algorithm to simultaneously optimize the transmitter's beamforming vector and STAR-IRS's transmission and reflection matrices.
4. Implementation of FPGA Hardware Accelerator: Leveraging hardware parallelism and structural optimization, the FPGA accelerator aims to minimize inference latency and power consumption, meeting the stringent real-time requirements of indoor communication systems.

Methodology Details

Task Definition

Input: Channel State Information (CSI) for Alice-to-Bob, Alice-to-Eves, STAR-IRS-to-Bob, STAR-IRS-to-Eves, and Alice-to-STAR-IRS links

Output:

• Transmitter beamforming vector w ∈ C^(N×1)
• STAR-IRS reflection coefficient matrix Ω_r
• STAR-IRS transmission coefficient matrix Ω̃_t

Objective: Maximize secrecy rate R_s = log₂(1 + γ_b) - max_{k∈K} log₂(1 + γ_k)⁺

System Model

Signal Transmission Model

Signal received by Bob:

y_b = h_b^H w s + f_b^H Ω_r G w s + n_b

Signal received by the k-th Eve:

y_k = h_k^H w s + f_k^H Ω̃_t G w z + n_k

where z is artificial noise generated through symbol-level random phase modulation.

STAR-IRS Element Model

The reflection and transmission coefficients of each STAR-IRS element are √(β_l^r)e^(jθ_l^r) and √(β_l^t)e^(jθ_l^t) respectively, satisfying the energy conservation constraint: β_l^r + β_l^t = 1.

SINR Expressions

SINR for Bob:

γ_b = |(h_b^H + f_b^H Ω_r G)w|² / σ_b²

SINR for the k-th Eve:

γ_k = |h_k^H w|² / (|f_k^H Ω̃_t G w|² + σ_k²)

Optimization Problem Formulation

(P1) max_{w,Ω_r,Ω̃_t} R_s
s.t. C1: Tr(ww^H) ≤ P_max
     C2: β_l^r + β_l^t = 1, ∀l ∈ L
     C3: 0 ≤ β_l^r ≤ 1, ∀l ∈ L  
     C4: 0 ≤ β_l^t ≤ 1, ∀l ∈ L
     C5: 0 ≤ θ_l^r ≤ 2π, ∀l ∈ L
     C6: 0 ≤ θ̃_l^t ≤ 2π, ∀l ∈ L

GNN Architecture Design

Graph Representation

Nodes: STAR-IRS, Bob, and all Eves Node Features: Include real and imaginary parts of direct and cascaded channels Adjacency Matrix: Reflects the connectivity between STAR-IRS and other nodes

Network Structure

1. Two GCN Layers: Capture complex interactions and dependencies among nodes
2. Two Independent FC Layers: Separately output beamforming vector w and STAR-IRS coefficients {Ω_r, Ω̃_t}

Feature Propagation

H^(i+1)' = Ã H^(i) F^(i)
H^(i+1) = f(H^(i+1)')

where à is the normalized adjacency matrix and f(·) is the activation function.

FPGA Hardware Acceleration

Design Features

1. GCN Layer Implementation: Pipelined matrix-vector multiplication engine supporting simultaneous computation for multiple output nodes
2. FC Layer Optimization: Pipelined matrix-vector multiplication units supporting batch processing
3. Memory Optimization: Weights stored in on-chip BRAM with feature streaming using burst transfers
4. Fixed-Point Arithmetic: Balances precision and resource utilization

Experimental Setup

Simulation Parameters

• Number of Antennas: Alice has N=8 antennas
• STAR-IRS Elements: L=80 elements
• Number of Eavesdroppers: K=2 Eves
• Transmit Power: P=18 dBm
• Noise Variance: -90 dBm
• Rician Factor: κ=0.3
• Distance Configuration: Alice-to-Bob, STAR-IRS-to-Bob, and Alice-to-STAR-IRS are all 8m; Alice-to-Eve and STAR-IRS-to-Eve are randomly generated in the 4-8m range

Comparison Schemes

1. AN-GNN: Proposed symbol-level random phase modulation scheme
2. CONV-GNN: Conventional STAR-IRS scheme
3. IRS-GNN: Pure reflection scheme
4. AN-MRT/ZF/MMSE: Schemes using traditional optimization methods

Evaluation Metrics

• Secrecy rate (bits/s/Hz)
• Convergence performance
• Computational latency

Experimental Results

Main Results

Convergence Performance

The GNN scheme converges rapidly within approximately 300 iterations, with AN-GNN consistently achieving the highest secrecy rate, validating the effectiveness of symbol-level random phase modulation.

Impact of Transmit Power

• At low transmit power, performance differences among schemes are minimal
• At high transmit power, AN-GNN significantly outperforms CONV-GNN and IRS-GNN
• GNN schemes overall outperform MRT, ZF, and MMSE schemes

Impact of Number of Eavesdroppers

• Increasing eavesdroppers leads to decreased secrecy rate
• AN-GNN performs best in multi-eavesdropper scenarios
• CONV-GNN shows the most significant performance degradation

Impact of STAR-IRS Element Count

• Increasing element count brings near-linear improvement in secrecy rate
• AN-GNN effectively utilizes additional elements to enhance Bob's signal while degrading Eves' signal quality

Impact of Channel Estimation Error

The GNN scheme demonstrates strong robustness under Gaussian-distributed channel estimation errors, maintaining high secrecy rates even with significant estimation errors.

FPGA Accelerator Performance

• Hardware Platform: Xilinx Zynq-7000 XC7Z100-2FFG900 SoC FPGA
• Inference Latency: 8.992 ms (10 ns clock period)
• Precision Loss: Fixed-point computation results are highly consistent with floating-point results
• Meets real-time requirements of indoor communication systems

Related Work

STAR-IRS-Assisted Physical Layer Security

Existing research primarily focuses on weighted sum secrecy rate maximization, key generation, and power minimization, but research specifically targeting indoor communications is relatively limited.

Deep Learning-Based IRS/STAR-IRS Optimization

• DRL Methods: While effective, suffer from long training times, high computational complexity, and limited generalization capability
• GNN Methods: Can effectively model network topology and scale efficiently, but their application in STAR-IRS-assisted secure indoor communications has not been explored

Conclusions and Discussion

Main Conclusions

1. Symbol-level random phase modulation strategy can significantly enhance secrecy performance in indoor communications
2. GNN scheme outperforms traditional methods in jointly optimizing beamforming and STAR-IRS coefficients
3. FPGA accelerator achieves low-latency inference, meeting real-time application requirements

Limitations

1. Assumes quasi-static flat fading channel model; actual environments may be more complex
2. Assumes eavesdropper locations are known; in practice, accurate acquisition may be difficult
3. Hardware implementation complexity of symbol-level random phase modulation is insufficiently discussed

Future Directions

1. Extension to more complex channel models and mobile scenarios
2. Consideration of robust optimization with unknown eavesdropper locations
3. Investigation of practical constraints and costs in hardware implementation

In-Depth Evaluation

Strengths

1. Strong Technical Innovation: First application of symbol-level random phase modulation to STAR-IRS-assisted secure indoor communications
2. Reasonable Methodology Design: GNN architecture fully leverages network topology characteristics with significant optimization results
3. Comprehensive Experiments: Validates scheme effectiveness and robustness from multiple dimensions
4. Hardware Implementation: FPGA accelerator design and implementation enhance practical applicability

Weaknesses

1. Scenario Limitations: Primarily targets specific indoor-outdoor separation scenarios with limited applicability
2. Insufficient Complexity Analysis: Lacks in-depth analysis of GNN training complexity and STAR-IRS hardware complexity
3. Practical Deployment Considerations: Lacks discussion of practical issues such as channel acquisition and synchronization in actual deployment

Impact

1. Academic Value: Provides new perspectives for STAR-IRS applications in indoor secure communications
2. Practical Value: FPGA implementation demonstrates engineering feasibility
3. Reproducibility: Detailed simulation parameters provided, but lacks open-source code

Applicable Scenarios

1. Indoor confidential communications requiring protection against external eavesdropping
2. Secure communications in smart buildings and office environments
3. Indoor wireless network security for military and government institutions

References

The paper cites 27 relevant references covering multiple domains including 6G networks, physical layer security, STAR-IRS technology, and deep learning optimization, with comprehensive and authoritative citation coverage.

Overall Assessment: This is a high-quality paper with innovation in the field of STAR-IRS-assisted secure communications. The technical approach is novel, experimental validation is comprehensive, and hardware implementation enhances practical value. While there are certain limitations in applicable scenarios and complexity analysis, the paper provides valuable contributions to the development of this field.