基于可解释网络解耦表征的低成本雷达定位解算方法  

Low-cost radar localization solution based on interpretable network decoupling representation

作　　者：刘磊[1,2] 林杰 Liu Lei;Lin Jie(School of Management,University of Shanghai for Science&Technology,Shanghai 200093,China;School of Optical-Electrical,University of Shanghai for Science&Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学管理学院,上海200093 [2]上海理工大学光电学院,上海200093

出　　处：《计算机应用研究》2024年第2期563-568,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(72071130);上海市自然科学基金资助项目(22ZR1443300)。

摘　　要：为降低调频连续波(FMCW)雷达成本,同时提高定位精度,设计可解释解耦表征模型。该模型由网络解算器、虚假信号生成器以及可解释潜变量三部分组成。首先处理雷达信号获得中频频谱;然后输入到网络解算器中生成位置潜变量;再通过物理机制对潜变量进行转换,生成虚假中频信号频谱;最后,设计局部光滑损失函数对模型进行自监督训练,实现潜变量的解耦物理表征。实验结果表明:所提算法能对雷达系统频谱信号的粗粒度进行超分辨率细化,其机理能有效应对雷达系统的硬件公差、环境噪声、安装误差等问题,并可自动地训练出雷达的解算网络,从而具有大规模室内、机载联网定位的应用潜力。In order to reduce the cost of frequency modulated continuous wave(FMCW)radar,while improving the positioning accuracy,this paper designed an interpretable decoupled characterization model.This model consisted of a network solver,a fake signal generator and interpretable latent variables.Firstly,it processed the radar signal to obtain the intermediate frequency spectrum.Then,it input processed signal into the network solver to generate position latent variables.Furthermore,it transformed these latent variables through physical mechanisms to produce a fake intermediate frequency spectrum.Finally,it designed a local smooth loss function to conduct self-supervised training on the model,achieving the decoupled physical representation of latent variables.Experimental results show that the proposed method can perform super-resolution refinement on the coarse spectrum signals of the radar system.The method can effectively addresses hardware tolerances,environmental noise,installation errors and other issues in the radar system.Moreover,it can train the radar’s solver network relatively automatically,demonstrating the application potentials for large-scale indoor and airborne network positioning.

关 键 词：调频连续波雷达定位 可解释深度网络 自监督学习 超分辨率细化 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]