低过采样数字调制信号的多尺度一维卷积神经网络解调器  

MULTI-SACLE 1D-CNN DEMODULATOR FOR LOW OVERSAMPLING DIGITAL MODULATION SIGNAL

作　　者：陈显敏 符杰林[1] Chen Xianmin;Fu Jielin(Key Laboratory of Cognitive Radio&Information Processing of the Ministry of Education,Guilin University of Electronic Technology,Guilin 541004,Guangxi,China)

机构地区：[1]桂林电子科技大学认知无线电与信息处理教育部重点实验室,广西桂林541004

出　　处：《计算机应用与软件》2024年第5期113-117,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61761014)。

摘　　要：针对应用深度学习方法对数字调制信号进行解调时过采样要求较高的问题,设计低过采样的多尺度一维卷积神经网络数字解调器。该解调器可以在与传统解调器相同的过采样条件下,对BPSK、4-QAM、8-QAM、16-QAM四种数字调制信号进行解调,并能保证传统解调方法相同的误码性能。仿真结果表明,在高斯和Rayleigh衰落信道下,给出的数字调制信号解调器可以在保证解调误码性能的同时,减少了对采样倍数的要求,降低了神经网络结构的复杂性。Aiming at the problem of high oversampling requirements when applying deep learning methods to demodulate of digital modulation signals,this paper designs a multiscale onedimensional convolutional neural network digital demodulator with low oversampling.It could demodulate the four digital modulation signals of BPSK,4QAM,8QAM,and 16QAM under the same oversampling conditions as the traditional demodulator,and could ensure the same error performance of the traditional demodulation method.Simulation results show that under Gaussian and Rayleigh fading channels,the provided digital modulation signal demodulator can not only ensure the performance of demodulation error codes,but also reduce the requirement of sampling multiple,and also reduce the complexity of neural network structure.

关 键 词：低采样倍数 解调 多尺度一维卷积神经网络 BPSK和M-QAM 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]