基于DQN的探测干扰一体化波形优化设计  

Integrated waveform optimization design of detection and jamming based on DQN

作　　者：陈涛[1,2] 张颖[1,2] 胡学晶 肖易寒[1,2] CHEN Tao;ZHANG Ying;HU Xuejing;XIAO Yihan(College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China;Key Laboratory of Advanced Marine Communication and Information Technology of Ministry of Industry and Information Technology,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 [2]哈尔滨工程大学先进船舶通信与信息技术工业和信息化部重点实验室,黑龙江哈尔滨150001

出　　处：《系统工程与电子技术》2023年第3期638-646,共9页Systems Engineering and Electronics

基　　金：上海航天科技创新基金(SAST2022-063);国防科技基础加强计划(2019-JCJQ-ZD-067-00)资助课题。

摘　　要：由于侦察干扰机设备具有发射功能,为使发射的干扰信号还具有探测的效果,考虑将探测信号隐藏在干扰信号中,提出一种基于非均匀间歇采样重复转发的探测干扰一体化信号波形。首先,建立一体化信号模型,并利用非均匀间歇采样重复转发技术实现幅度编码调制;然后,在优化过程中,从模糊函数以及雷达检测环节分析一体化信号的特征,根据距离、速度分辨率以及脉压后幅度的均值与标准差之比,构造相应的目标函数;最后,利用深度Q学习算法求解目标函数,获取最优的幅度编码方式。仿真结果表明,当编码状态量小时,深度Q网络(deep Q-network,DQN)算法与强化学习算法收敛效果一致。与遗传算法相比,DQN算法最优解的质量提高了13.10%;当编码状态量增大时,相对于遗传算法和强化学习算法,DQN算法的收敛值更优,最优解更稳定。In view that the reconnaissance jammer has the transmitting function,with the aim that the transmitted interference signal can also have the detection effect.The detection signal can be hidden in the jamming signal.A detection and jamming integrated signal waveform based on non-uniform intermittent sampling and repeated forwarding is proposed.Firstly,an integrated signal model is established,and then the non-uniform intermittent sampling and repeating forwarding technology is used to realize amplitude coding modulation.Secondly,in the optimization process,the characteristics of the integrated signal are analyzed from the fuzzy function and the radar detection link,and the corresponding objective function is constructed according to the distance,velocity resolution,and the ratio of the mean value to the standard deviation value of the post-pulse compression amplitude.Finally,the deep Q-learning algorithm is used to solve the objective function to obtain the optimal amplitude coding.The simulation results show that when the number of coding states is small,the convergence effect of deep Q-network(DQN)algorithm is consistent with that of the reinforcement learning algorithm.Compared with the genetic algorithm,the quality of optimal solution of DQN algorithm is improved by 13.10%.When the number of coding states increases,compared with genetic algorithm and reinforcement learning algorithm,the convergence value of DQN algorithm is larger and the optimal solution is more stable.

关 键 词：探测干扰一体化信号 非均匀间歇采样重复转发 模糊函数 脉冲幅度编码 深度Q学习 

分 类 号：TN97[电子电信—信号与信息处理]