基于深度学习的仅相位变化的波束形成算法  

Phase-Only Beamforming Algorithm Based on Deep Learning

作　　者：丛阳 CONG Yang(National Key Laboratory of Electromagnetic Space Security,Jiaxing,Zhejiang 314033,China;No.36 Research Institute of CETC,Jiaxing,Zhejiang 314033,China)

机构地区：[1]电磁空间安全全国重点实验室,浙江嘉兴314033 [2]中国电子科技集团公司第三十六研究所,浙江嘉兴314033

出　　处：《通信对抗》2024年第4期12-16,共5页Communication Countermeasures

摘　　要：仅相位变化的波束形成可以减少功率损失,减少运算时间,在工程应用中尤其关键。但由于加权因子恒等于一,问题是非凸的,难以解决。深度学习具有强大且鲁棒的高维函数模型拟合能力,提出将深度学习用于仅相位变化的波束形成设计。首先,仅将波束形成的相位作为深度学习网络的输入进行优化。然后,基于深度学习中的多层全连接(FC)网络和反向传播网络(BP)构造了一个离线学习的优化训练网络(OTN),无需松弛就能求解这个非凸问题,得到优化后的相位。仿真表明,所提出的方法具有很好的收敛性,在不同的场景下在干扰方向都具有较好的凹口性能,抗干扰能力较强。Phase-only beamforming reduces power losses and computation time,which is especially critical in engineering applications.But since the weighting factor is always equal to one,the problem is non-convex and difficult to solve.Deep learning has a strong and robust ability to fit high-dimensional function models,so this paper proposes to use deep learning for phase-only beamforming design.Firstly,the weighted factor design of constant mode constraints is transformed into a phase design problem.Then,based on the multi-layer Fully Connected(FC)network and back-propagation network(BP)in deep learning,we construct an ofline learning Optimized Training Network(OTN)to solve the non-convex problem without relaxation.After computer simulation,the comparison results show that the proposed method has good notch performance and good convergence in different scenarios.

关 键 词：仅相位变化 波束形成 离线学习 非凸优化 

分 类 号：TN911.7[电子电信—通信与信息系统]