Beamforming Optimization for Intelligent Reflecting Surface Assisted MISO: A Deep Transfer Learning Approach
Beamforming Optimization for Intelligent Reflecting Surface Assisted MISO: A Deep Transfer Learning Approach
复制标题
DOI：
10.1109/tvt.2021.3062870
复制
发表时间：
2021-04-01
影响因子：
6.8
通讯作者：
Fan, Jiancun
中科院分区：
文献类型：
--
作者：
Ge, Yimeng;Fan, Jiancun
研究了智能反射面辅助多输入单输出无线通信系统的波束形成优化问题。我们建立了一个基于深度迁移学习（DTL）的框架，以学习如何优化IRS侧的相移。在此基础上，我们还设计了一个损失函数来实现无监督训练，而无需大量的标记数据样本。最后，我们将优化问题扩展到离散相移约束，以解决硬件限制。仿真结果表明，提出的基于DTL的方法可以实现类似的性能相比，上限，同时大大降低了计算复杂度。
This article studies the beamforming optimization for intelligent reflecting surface (IRS) assisted multiple-input single-output (MISO) wireless communication system. We establish a deep transfer learning (DTL)-based framework to learn how to optimize the phase shifts at the IRS side. Based on it, we also design a loss function to implement unsupervised training without a large number of labeled data samples. Finally, we extend the optimization problem to discrete phase shift constraint to solve the hardware limitation. The simulation verifies that the proposed DTL-based approach can achieve similar performance compared with upper bound while substantially reducing the computational complexity.