基于深度学习的非扩跳频通信体制窄带干扰抑制技术  

Deep Learning Based Narrowband Interference Suppression Technique in Non-Spread-Hopping Spectrum Communication System

作　　者：马松[1,2] 王家豪 张蔺 王军[1] MA Song;WANG Jiahao;ZHANG Lin;WANG Jun(National Key Laboratory of Wireless Communications,University of Electronic Science and Technology of China,Chengdu 611731,China;Southwest China Institute of Electronic Technology,Chengdu 610036,China;Shenzhen Institute for Advanced Study,University of Electronic Science and Technology of China,Shenzhen 518110,China)

机构地区：[1]电子科技大学通信抗干扰全国重点实验室,成都611731 [2]中国西南电子技术研究所,成都610036 [3]电子科技大学(深圳)高等研究院,深圳518110

出　　处：《电子科技大学学报》2024年第6期862-870,共9页Journal of University of Electronic Science and Technology of China

基　　金：电子科技大学通信抗干扰全国重点实验室基础科研创新基金(IFN20230102)。

摘　　要：复杂环境下的无线业务在近年来呈现数据密集化发展趋势,对无线通信系统的数据传输能力和干扰对抗能力都提出了更高的要求。现有扩跳频通信体制通过牺牲时频资源利用率换取干扰对抗能力,难以满足复杂环境下的高速数据传输需求。为此,提出了一种基于深度学习的非扩跳频通信体制窄带干扰抑制技术,在非扩跳频通信体制基础上,在接收端级联频域陷波模块和深度神经网络模块有效抑制窄带干扰,同时提升数据传输速率和干扰抑制能力。其中,深度神经网络用于从频域陷波后的失真信号中重构真实信号。实验结果表明,与传统频域陷波算法相比,所提出的算法具有更低的误码率,训练好的深度神经网络能够泛化到不同信号强度、干扰强度、干扰频段、干扰波形等场景。The wireless service in complex environment shows the trend of data intensive development in recent years,which puts forward higher requirements for the data transmission capability and interference countermeasure capability of wireless communication systems.The existing spread spectrum and frequency hopping communication systems achieve interference countermeasure capability by sacrificing time-frequency resource utilization,but can not meet the demand of high-speed data transmission in complex environment.Therefore,this paper proposes a deep learning based narrowband interference suppression technology of nonspread-hopping spectrum communication system.On the basis of non-spread-hopping spectrum communication system,the frequency-domain notch filtering module and deep neural network module are cascaded at the receiver end to effectively suppress narrowband interference,improving data transmission rate and interference suppression ability at the same time.In particular,the deep neural network is used to reconstruct the expected signal from the distorted signal by the frequency-domain notch filtering module.The experimental results show that the proposed algorithm has a lower bit error rate than the traditional frequency-domain notch filtering algorithms,and the welltrained deep neural network can generalize to the scenarios with differences of signal power,interference power,interference frequency band,interference waveform,etc.

关 键 词：深度学习 泛化能力 窄带干扰抑制 非扩跳频通信 

分 类 号：TN97[电子电信—信号与信息处理]