基于深度网络的太赫兹波束预判方法研究

针对太赫兹波波束较窄引起的伺服天线被动跟踪无法达到空间太赫兹通信链路性能要求的问题，提出了一种基于优化深度网络的太赫兹波束预判方法。首先通过分析方位、俯仰角误差大小随运行周期、外推初始时刻的变化，得到高轨对低轨卫星指向误差呈周期性发散的特征；然后采用粒子群算法优化长短时记忆网络参数，对未来时刻指向误差进行预测并修正，针对粒子群算法易陷入局部最优、全局搜索能力较差的问题，通过动态调整粒子群算法的惯性权重，以达到优化长短时记忆网络的目的。仿真结果表明：基于改进后粒子群算法优化的长短时记忆网络能够有效预测未来时刻指向误差，在同一链路场景中相比未改进网络平均绝对百分比误差降低13.08%。

Research on terahertz beam prediction method based on deep network

Aiming at the problem that passive tracking of servo antennas cannot meet the performance requirements of spatial terahertz communication links caused by narrow terahertz beams, a pre-judgment method of terahertz beams based on optimized deep networks is proposed. Firstly, by analyzing the variation of the azimuth and pitch angle errors with the operation period and the initial time of extrapolation, the characteristic of the periodic divergence of the pointing errors of high-orbit satellites to low-orbit satellites is obtained. Then, the long-short-term memory network parameters are optimized by particle swarm optimization, the pointing error of the future time is predicted and corrected. In view of the poor global search ability of the particle swarm optimization and easy to fall into the local optimum, the inertia weight of the particle swarm optimization is dynamically adjusted to achieve the purpose of optimizing the long-short-term memory network. The simulation results show that the long-short-term memory network optimized by the improved particle swarm optimization can effectively predict the pointing error in the future, and the average absolute percentage error is reduced by 13.08% compared with the unimproved network in the same scenario.