【作者】 闫静；

【导师】 金明录；

【作者基本信息】 大连理工大学 ， 信息与通信工程， 2023， 硕士

【摘要】 分布式多输入多输出(Multiple-Input Multiple-Output,MIMO)雷达是一种天线阵元间隔很远的雷达系统,其可以从不同角度对目标进行全面的探测,从而获得空间分集增益和几何分集增益,显著提高雷达运动目标探测性能,是国际雷达信号处理领域的热点研究方向之一。在现实环境中,雷达会受到各种干扰源的影响,这些干扰源导致雷达接收到的信号中包含大量的杂波。由于复杂电磁环境下杂波的统计特性呈现出非高斯、非线性以及非平稳的特点,给运动目标检测带来了极大的挑战。本文针对杂波环境下分布式MIMO雷达的动目标检测问题,将统计模型和深度神经网络相结合,构建了基于模型驱动的深度展开检测网络。首先,考虑在目标速度已知,但杂波纹理分量和目标反射系数未知的情况下,分析分布式MIMO雷达在杂波环境中的目标回波模型。在该回波模型的基础上,推导基于统计模型的广义似然比检测器(Generalized Likelihood Ratio Test,GLRT)、Rao和Wald检测器。仿真结果表明,雷达采样点数越多、接收天线数量越多或者杂波越尖锐时,三种检测器的检测性能越好。同时,Wald检测器的性能较好,GLRT次之,Rao较差。其次,针对已有基于统计模型或数据的目标检测方法的技术瓶颈,通过引入一个二进制离散变量,将基于GLRT准则的二元假设检验问题转化为离散变量的估计问题,充分结合统计模型和数据信息,基于深度展开网络构建了改进的纽曼皮尔逊神经网络(Improved Neyman Pearson network-GLRT,INPnet-GLRT),并提出了一种新的网络训练方法。仿真结果表明,相较于已有的目标检测方法,基于INPnet-GLRT的检测方法有明显的性能提升,并且能够保证恒定的虚警概率。最后,将基于GLRT准则的纽曼皮尔逊神经网络构建框架拓展到Wald检验准则,构建了基于深度展开网络的INPnet-Wald检测网络。仿真结果表明,相较于传统的Wald检测器,INPnet-Wald的检测性能较好。针对杂波环境中分布式MIMO雷达动目标检测问题,本文基于两种不同的检验准则构建了基于模型驱动的深度神经网络,在一定程度上丰富了雷达运动目标检测理论与方法,对航空航天、军事防御等具有重要意义。更多还原

【Abstract】 The distributed multiple-input multiple-output(MIMO)radar is a radar system with widely separated antennas.It can comprehensively detect targets from different angles,thereby obtaining spatial diversity gain and geometric diversity gain,which significantly improve the target detection performance of radar.The distributed MIMO radar target detection has become one of the focus research directions in the field of radar signal processing internationally.In real-world environments,radar can be affected by various sources of interference.These sources of interference can result in a significant amount of clutter in the received radar signals.Due to the non-Gaussian,nonlinear and non-stationary statistical characteristics of clutter in complex electromagnetic environments,it poses a great challenge for moving target detection.Aiming at the moving target detection problem of distributed MIMO radar in clutter environment,this paper combines statistical model and deep neural network to construct a deep unfolding detection network.Firstly,the target echo model of distributed MIMO radar in clutter environment is analyzed,assuming that the target velocity is known while the clutter texture component and target reflection coefficient are unknown.Based on this echo model,the generalized likelihood ratio test(GLRT),Rao and Wald detector based on statistical model are derived.Simulation results indicate that the detection performance of the three detectors is better when the number of radar sampling points is more,the number of receiving antennas is more,or the clutter is sharper.Meanwhile,the performance of the Wald detector is better,followed by the GLRT,and Rao is poor.Secondly,aiming at the technical bottleneck of existing target detection methods based on statistical model or data,by introducing a binary discrete variable,the binary hypothesis testing problem based on the GLRT criterion is transformed into the estimation problem of discrete variable.An Improved Neyman-Pearson network-GLRT(INPnet-GLRT)is constructed based on the deep unfolding network by fully combining statistical model and data information,and a new network training method is proposed.The simulation results show that,compared with the existing target detection methods,the detection method based on INPnet-GLRT has obvious performance improvement,and can guarantee a constant false alarm probability.Finally,the Newman-Pearson neural network based on the GLRT criterion is further extended to the Wald test criterion,and the INPnet-Wald detection network based on the deep unfolding network is constructed.The simulation results show that,compared with the traditional Wald detector,the detection performance of INPnet-Wald is better.Aiming at the problem of distributed MIMO radar moving target detection in clutter environment,this paper constructs two model-driven deep neural network based on two different test criterions,which enrich the theory and method of radar moving target detection to a certain extent,and are of great significance to aerospace,military defense and so on.更多还原