Research On New Method Of Adaptive Target Detection In Clutter Environment Of MIMO Radar

MIMO radar with widely separated antennas employes multiple widely spreaded transmit and receive antennas.Its geometry gain and diversity gain can significantly improve the performance of radar system in many aspects,making it one of the popular research topics in the radar society.Clutter often exists in the real world environment,which causes difficulties to radar signal processing such as target detection.How to optimize the detection scheme adaptively according to the clutter to better detect the target is another research interest.This thesis considers the adaptive target detection problem using MIMO radar in clutter environment.The MIMO detector based on statistical model is studied,and adaptive MIMO detector based on deep learning is proposed.We further propose adaptive MIMO detectors combining the statistical method and deep learning.Classical statistical analysis is employed to construct the statistical model of the MIMO radar received signal under clutter,assuming that the target with unknown reflection coefficients moves with a known speed in compound Gaussian clutter with unknown parameters.Based on the statistical model,a binary hypothesis testing problem is formulated for detection,and the generalized maximum likelihood ratio test(GLRT)detector is derived.Then we consider the case where the speed of the target is assumed unknown,and the associated GLRT detector is investigated.The effects of radar and environmental parameters,such as the number of antennas,target reflection coefficients,clutter texture components,and clutter covariance matrix,on the detection performance are analyzed.In case no sufficient statistical information is available,we propose to construct a MIMO radar deep neural network(DNN)to solve for the target detection problem.We use the received data to train the network,from which the network learns the best way to detect the target in the current clutter environment to realize adaptive target detection.The detection performance of the proposed DNN-based adaptive detector is analyzed via numerial examples.Considering that there are cases where both statistical aprior and radar receive data are available,we propose to design adaptive MIMO detectors based on joint statistical processing and deep learning.Initially,we consider the known target velocity case for simplicity.By defining a target detection parameter,the problem of MIMO radar target detection is reformulated as an equivalent estimation problem of the target detection parameter.A neural network loss function accounting for statistical analysis under the Neyman-Pearson(NP)criterion is developed.Taking into account both the statistical NP detection and deep learning network,an MIMO radar detector scheme with joint statistical and deep learning processing(called NPnet)is proposed.Numerical study shows that when the statistical model matches the actual observation,NPnet has similar performance as the GLRT.In the case of random target reflection coefficients,the NPnet outperforms the DNN method.If the statistical model mismatched the acutal observation,the performance of the NPnet is way much better than the GLRT detector.For the more general case with unknown target velocity,both the velocity estimation and NP detection are considered in the construction of the neural network loss function,such that the target velocity and detection parameter can be obtained simultaneously at the output of the neural network.Incoproating building blocks for the estimate of the target velocity into the NPnet,the scheme of the joint MIMO radar target detector and velocity estimator is proposed(called JNPnet).From numerical studies,results similar to the NPnet were found,which again show the benifits of the proposed joint processing.