基于GRU网络的毫米波波束跟踪和阻碍判断联合预测方案

为了提高移动通信场景下，毫米波(millimeter wave，mmWave)大规模多输入单输出(multiple input single output，MISO)系统传输的稳定性，针对快速准确跟踪和阻碍判断问题，提出将波束跟踪和阻碍判断的联合预测问题定义为一个波束状态时间序列预测问题，设计了相应的联合预测数据集;基于门控循环单元(gated recurrent unit，GRU)模块设计未来波束状态预测方案，提出分布式固定输入门控循环单元(gated recurrent unit fixed input，GRU-FIN)训练方案，用来提高模型预测能力。通过仿真实验与3种基线方案进行对比，分析了迭代次数、天线数量、信噪比和神经网络参数设置对预测模型性能的影响。仿真结果表明，在不需要信道状态信息的情况下，该方案能够拟合移动用户非线性波束状态的变化，并且在观察范围较小的情况下，通过GRU-FIN方案和模型参数设计，能够有效提高波束状态的预测能力。

Joint prediction scheme for millimeter wave beam tracking and obstruction determination based on GRU network

In order to improve the stable transmission of the millimeter wave (mmWave) massive multiple input single output (MISO) system in the mobile communication scenarios, we target fast and accurate beam tracking and obstruction judgment problems. Firstly, we propose to define the joint prediction problem of beam tracking and obstruction judgment as a beam state time series prediction problem, and design the corresponding joint prediction data set. Then a future beam state prediction scheme is proposed based on the gated recurrent unit (GRU) module. A distributed fixed input gated recurrent unit (GRU-FIN) training scheme is proposed to improve the model prediction capability. Finally, simulation experiments are compared with three baseline schemes to analyze the effects of the number of iterations, the number of antennas, the signal-to-noise ratio, and the neural network parameter settings on the prediction model performance. The simulation results show that the scheme can fit the variation of the nonlinear beam state of mobile users without channel state information. The GRU-FIN scheme and the model parameters design can effectively improve the beam state prediction in the case of a small observation range.