| 摘要: |
| 针对传统智能反射面（Intelligent Reflecting Surface，IRS）辅助多输入单输出（Multiple-Input Single-Output，MISO）无线通信系统信息安全传输算法计算复杂度高、泛化能力不足的缺点，提出了一种基于图神经网络（Graph Neural Network，GNN）的联合波束成形优化方案。通过建立IRS反射系数与基站（Base Station，BS）发射波束成形的协同优化模型，在满足BS发射功率约束及IRS反射单元模一约束条件下，实现系统最小保密速率最大化目标。具体而言，首先构建无线信道环境的图结构表征，将导频信号映射为节点特征向量，随后基于消息传递机制实现多层图卷积运算，最终生成BS波束成形向量与IRS反射系数的最优解。仿真实验表明，在非理想信道状态信息（Channel State Information，CSI）条件下，相较传统交替优化（Alternative Optimization，AO）算法，所提方案可提升约20%的保密速率性能，同时计算时延降低约98.2%，且在不同用户数量和传输功率条件下表现出更好的鲁棒性，显著优于现有方法。 |
| 关键词: 多输入单输出(MISO) 智能反射面 图神经网络 物理层安全 |
| DOI：10.20079/j.issn.1001-893x.240926001 |
|
| 基金项目:国家自然科学基金资助项目(61831002) |
|
| A Security Rate Optimization Algorithm for IRS-assistedMISO System Based on Graph Neural Network |
| TANG Chongxin,XI Bing,DENG Bingguang |
| (School of Communications and Information Engineering，Chongqing University of Posts and Telecommunications，Chongqing 400065，China) |
| Abstract: |
| To address the shortcomings of high computational complexity and insufficient generalization capability in traditional intelligent reflecting surface(IRS)-assisted multiple-input single-output(MISO) wireless communication systems for secure information transmission，a graph neural network(GNN)-based joint beamforming optimization scheme is proposed.By establishing a collaborative optimization model between IRS reflection coefficients and base station(BS) transmit beamforming，the objective of maximizing the system’s minimum secrecy rate is achieved under BS transmit power constraints and IRS reflection unit modulus constraints.Firstly，a graph-structured representation of the wireless channel environment is constructed and pilot signals are mapped to node feature vectors.Subsequently，multi-layer graph convolution operations are implemented through a message-passing mechanism，ultimately generating optimal solutions for BS beamforming vectors and IRS reflection coefficients.Simulation results demonstrate that under imperfect channel state information(CSI) conditions，compared with traditional alternative optimization(AO) algorithms，the proposed scheme achieves approximately 20% improvement in secrecy rate performance while reducing computational latency by about 982%.Moreover，it exhibits superior robustness across varying user numbers and transmission power conditions，significantly outperforming existing methods. |
| Key words: multiple-input single-output(MISO) intelligent reflector surface graph neural network physical layer security |
《电讯技术》编辑部 