摘要
无线通信易受到干扰攻击,为此,基于深度强化学习的抗干扰跳频(Frequency Hopping, FH)通信一直是近几年来通信中的一个活跃研究课题。针对跳频智能抗干扰通信场景,设计了一种基于注意力机制与长短时记忆(Long Short-Term Memory, LSTM)网络相结合的抗干扰深度Q网络(Deep Q Network, DQN)算法决策神经网络,该神经网络基于感知的短时频谱瀑布信号输入做出下一跳的跳频频点决策。所提出的决策神经网络通过引入注意力机制模块和LSTM处理模块可以快速提取短时频谱瀑布图中的时频结构信息,从而实现决策神经网络的在线训练加速。仿真结果表明,在梳状干扰与扫频干扰下,该决策神经网络具有快速收敛特性,只需训练一轮即可收敛,与单纯的DQN算法和其他深度决策神经网络相比具有更加优异的收敛性能,适于抗干扰动态决策。
Since military wireless communications are susceptible to interference attacks,anti jamming Frequency Hopping(FH)communications have been an active research topic in communications in recent years.In this paper,an anti-interference Deep Q Network(DQN)decision-making network based on attention mechanism and Long Short-Term Memory(LSTM)is designed for intelligent FH communication scenarios.The input of this decision network is a perceived short-term spectral waterfall signal,and the output is the next-step working frequency.The proposed decision-making network can extract time-frequency structure information in short-term spectrum waterfall diagram more efficiently by introducing an attention mechanism module and a LSTM processing module.Simulation results show that under the comb interference and sweep interference,the decision network can converge after only one round of training,which has better decision performance than other deep decision networks such as simple DQN algorithm.
作者
靳越
吴晓富
张剑书
JIN Yue;WU Xiaofu;ZHANG Jianshu(College of Communication and Information Engineering,Nanjing University of Posts and Telecommunication,Nanjing 210003,China;School of Computer Engineering,Nanjing Institute of Engineering,Nanjing 211167,China)
出处
《无线电通信技术》
2023年第6期1059-1066,共8页
Radio Communications Technology
基金
国家自然科学基金(61771256)~~。
