毫米波MIMO系统中基于CNN的混合波束成形设计

Hybrid beamforming design based on CNN in millimeter-wave MIMO system

混合波束成形是毫米波多输入多输出(MIMO)系统中的关键技术,提出了一种基于深度学习的方法来克服复杂性问题且提升系统性能。首先,利用无约束波束成形的相互正交性,对基带波束成形加以正交性的约束后确定等效波束成形器从而通过相位提取来获得模拟波束成形器的相位,将获得的输出插入到基于卷积神经网络(CNN)的混合波束形成(HBCN),HBCN使用的可行集中;其次,HBCN是将天线选择和混合波束成形器设计作为CNN的分类、预测问题,在天线选择上将信道矩阵作为输入,找出最优子阵列,合成的子阵信道矩阵再反馈给CNN来获得模拟和基带波束成形。最后,仿真结果显示,对比传统算法,能够得到更好的频谱效率和更低的复杂度。

Hybrid beamforming is a key technology in mm-wave multiple-input-multiple-output(MIMO)system.A deep-learning-based approach is proposed to overcome the complexity and improve performance of the system.Firstly,based on the mutual orthogonality of unconstrained beamforming,the equivalent beamforming is determined by applying the orthogonality constraint of baseband beamforming,and the phase of analog beamforming is obtained by phase extraction.The outputs are inserted into the feasible set which is used by hybrid beamforming via convolutional neural network(CNN)(HBCN).Secondly,HBCN takes antenna selection and hybrid beamforming design as the classification and prediction problems of CNN.In antenna selection,channel matrix is used as input to find the optimal subarray,and the synthesized subarray channel matrix is fed back to CNN in order to obtain simulation and baseband beamforming.Finally,the simulation results show that compared with the traditional algorithm,the spectral efficiency is better and the complexity is lower.