嵌入式GPU平台下雷达调制信号分析识别  

Analysis and Recognition of Radar Modulation Signal under Embedded GPU Platform

作　　者：杨博溢 汪向阳 陈涛[1] 李君[1] YANG Bo-yi;WANG Xiang-yang;CHEN Tao;LI Jun(Harbin Engineering University,Harbin 150001,China;Unit 63861 of PLA,Baicheng 137001,China)

机构地区：[1]哈尔滨工程大学,黑龙江哈尔滨150001 [2]解放军63861部队,吉林白城137001

出　　处：《舰船电子对抗》2022年第1期72-76,共5页Shipboard Electronic Countermeasure

摘　　要：针对近些年雷达脉内信号识别方式由传统的特征提取不断地向着深度学习方向发展,且雷达信号侦察设备的需求在深度学习的基础上逐渐地向便携、低功耗方向发展,而对于一些特定的研究往往对嵌入式平台又具有一定要求。基于此,研究了基于嵌入式图形处理器(GPU)平台的雷达信号分析识别系统的设计。将近些年提出的轻量级网络MobileNet V3移植到嵌入式GPU平台,采用基于SPWVD时频分析结合MobileNet V3网络,实现了低信噪比下对雷达信号的侦测识别,且与传统卷积神经网络在训练效率和训练准确率方面进行比较,最后完成了信号分析的图形用户界面(GUI)软件设计,为深度学习雷达信号识别在嵌入式平台的应用提供了基础。In recent years,the radar intra pulse signal recognition method has been continuously de-veloped from the traditional feature extraction to the direction of deep learning,and the demand for radar signal reconnaissance equipment has been gradually developed to the direction of portability and low power consumption based on deep learning.Several specific researches often have certain requirements for the embedded platforms.Based on the above,this paper studies the design of radar signal analysis and recognition system based on embedded graphics processing unit(GPU)plat-form.The lightweight network MobileNet V3 proposed in recent years is transplated to the embed-ded GPU platform,and SPWVD-based time-frequency analysis combined with MobileNet V3 net-work are used to realize the detection and recognition of radar signals under low signal-to-noise ra-tio.The method is compared with traditional convolutional neural networks,finally the graphical user interface(GUI)software design of signal analysis is completed,which provides the foundation for the application of deep learning radar signal recognition to the embedded platform.

关 键 词：深度学习 嵌入式图形处理器平台 雷达信号识别 MobileNet V3 时频分析 

分 类 号：TN971.1[电子电信—信号与信息处理]