一种基于机器学习的卫星干扰信号DOA估计算法

A Machine Learning Based DOA Estimation Algorithm for Satellite Interference Signals

随着现代通信技术的发展,波达方向(Direction of Arrival,DOA)估计在卫星通信抗干扰中越来越重要,通过对干扰信号的DOA实时和精确估计,为后续抗干扰提供依据。传统的估计算法计算任务重且耗费时间,不利于对干扰信号进行实时定位,如MUSIC,ESPRIT算法等。在此背景下,提出了基于机器学习的DOA估计方法,采用神经网络学习方位特征样本,在空间信号和方位角之间建立非线性映射关系,利用训练后的网络估计方位角度,可减小运算量和提高估计精度。在分析了机器学习算法特点的基础上,提出基于BP神经网络和RBF神经网络的DOA估计算法,并从算法复杂度、信噪比、相干性和信号类型等方面评价了估计性能,通过仿真结果分析,得出RBF网络DOA估计性能优于BP网络的结论。

With the development of modern communication technology,DOA estimation becomes more and more important in anti-jamming of satellite communication.Through real-time and accurate estimation of DOA of interference signal,the basis for subsequent anti-jamming is provided.Traditional algorithms are complex and time-consuming,which are disadvantageous to the real-time location of interference signals,such as MUSIC,ESPRIT algorithm and so on.In this context,a DOA estimation method based on machine learning is proposed.The neural network learns the azimuth feature samples and establishes a non-linear mapping relationship between the spatial signal and azimuth.Using the trained network to estimate azimuth can reduce the computational complexity and improve the estimation accuracy.Based on the analysis of the characteristics of machine learning algorithm,a DOA estimation algorithm based on BP neural network and RBF neural network is proposed,and the algorithm complexity,signal-to-noise ratio,coherence and signal type are analyzed.