一种基于神经网络的LFMCW雷达目标距离提取方法

A METHOD BASED ON NEURAL NETWORK FOR LFMCW RADAR SYSTEM OBJECT RANGE SUPER-RESOLUTION ESTIMATION

根据 L FMCW系统的特点 ,提出了一种通过对信号自相关矩阵进行秩 - 1分解的方法来替代传统的 FFT方法 ,以实现对目标距离的超分辨提取 .通过应用 Hopfield神经网络能量函数变换 ,将分解问题转化为一个简单的迭代问题来求解 .文中通过计算机仿真和硬件系统的实际测试研究了它的性能 ,并与 MUSIC、最大熵等其它谱估计方法做了比较 ,结果表明该方法具有更好的信噪比和分辨性能 .

To replace the traditional FFT method to realize the object range super resolution estimation, an artificial neural network method was proposed to decompose signal's auto correlation matrix into the summation of rank 1 matrices, and convert the decomposition problem to an iterative one by using Hopfield neural network. The property of this method was investigated both theoretically and experimentally. And the results were compared with five other typical super resolution algorithms including MUSIC, etc., and it was found that the present method has a lower SNR threshold and higher range resolution.