无线电智能感知仪的设计与实现  

Design and realization of radio intellectual sensing instrument

作　　者：王睿奇 付丁一 马鹏 陈熙来 侯长波[1] WANG Ruiqi;FU Dingyi;MA Peng;CHEN Xilai;HOU Changbo(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《应用科技》2024年第1期136-142,共7页Applied Science and Technology

摘　　要：为解决电磁环境日益复杂下传统电磁感知设备体积大、人工操作繁琐、识别准确率低和运算量庞大等问题,本文设计了一款无线电智能感知仪,提出了一种基于滑动窗的信号频谱能量检测算法,以提高识别精度;其次为降低模型与运算量大小,提出了一种基于轻量化神经网络的识别算法;最后设计搭建了无线电智能感知仪硬件模块,部署算法。实验结果表明,所设计的无线电智能感知仪在信号感知任务中有较好的性能,在0 dB及以上的信噪比环境下,调制识别的准确率可达到95%以上,对不同的信号调制类型的召回率和精准度均在93%以上,模型轻量化部署后降低了99.07%的浮点运算量,而准确率仅下降了0.25%。试验结果可用于指导在复杂电磁环境下无线电频谱感知设备的设计与制作过程。In an electromagnetic environment that is more and more complex,in order to solve the problems of traditional electromagnetic perception devices such as large volume,cumbersome manual operations,low recognition accuracy,and large computational complexity,this study designed a wireless intelligent perception instrument and proposed a signal spectrum energy detection algorithm based on sliding windows to improve recognition accuracy;And then,a recognition algorithm based on lightweight neural network was proposed to reduce the size of the model and computation;Finally,the hardware module of the wireless intelligent perception instrument was designed and built,and the algorithm was deployed.The experimental results show that the designed wireless intelligent perception instrument has good performance in signal perception tasks.In signal-to-noise ratio environments of 0dB or above,the accuracy of modulation recognition can reach over 95%,with the recall and accuracy for different signal modulation types are above 93%.After lightweight deployment,the model reduces the floating-point computational load by 99.07%,while the accuracy only decreases by 0.25%.This conclusion can be used to guide the design and production process of wireless spectrum sensing equipment in a complex electromagnetic environment.

关 键 词：边缘计算 信号检测 信号识别 轻量化部署 无线电智能感知 嵌入式 深度学习 卷积神经网络 

分 类 号：TN98[电子电信—信息与通信工程]