Study On Target Enumeration Methods In MIMO Radar Based On Random Matrix Theory And Modified MDL Criterion

MIMO(Multiple-Input Multiple-Output) radar, as a new radar emerging in recent years, has attracted a lot of attentions among the scholars at home and aboard since it has large advantages in the areas of target detection, identification and localization. In MIMO radar signal processing, multi-target parameter estimation and localization are always the hot research fields. However, their prerequisite is to detect the number of the targets in certain noise environment. Therefore, target enumeration is one of the important parts in MIMO radar signal processing.Commonly, target enumeration in MIMO radar can refer to traditional source enumeration methods. However, these methods depend on a prerequisite that the number of the transmitting and receiving array elements is small whereas the number of snapshots is large, in which case the statistical covariance matrix can be replaced by sample covariance matrix. However, when the number of snapshots is limited and the number of array elements is large such that they are in the same magnitude, we can’t replace the statistical covariance matrix by the sample covariance matrix. The traditional signal processing methods are no longer valid.In array signal processing, MDL(Minimum Description Length) criterion is a source enumeration method based on Information theory. MDL criterion has been widely used for its many advantages such as consistent estimation of the number of the targets, no requirement of decision threshold and small calculation load. In this thesis, the signal model of bistatic MIMO radar is used, and the traditional MDL criterion is improved. Considering the development tendency of MIMO radar towards large arrays, the methods of target enumeration in large arrays MIMO radar are deeply investigated based on random matrix theory(RMT) and modified MDL criterion in order to solve the problem caused by the small number of snapshots and the large number of array elements. The work is supported by the project “Robust targetdetection and estimation for MIMO radar based on large dimensional random matrix theory”(No. 61371158), which is supported by the National Natural Science Foundation of China. The creative works of this paper are as follows:For insufficient number of snapshots, a target enumeration method is proposed for MIMO radar based on LS-MDL criterion. The linear shrinkage(LS) optimization technology is used to deal with the sample covariance matrix in the condition that the product of the numbers of the transmitting and receiving array elements is in the same level with the number of the snapshots, so that the sample covariance matrix can replace the stationical covariance matrix again. Then its eigenvalues are substituted into the MDL Criterion to improve the performance of target enumeration.When the numbers of transmitting and receiving elements are large, the eigenvalues of large dimensional sample covariance matrix are modified based on the asymptotic spectrum theory in RMT. Combined with the MDL criterion, a target enumeration method is proposed for large array MIMO radar based on RMT-MDL criterion. The performance of the target detection has been improved.Further, a target enumeration method is proposed for MIMO radar based on MDLr criterion.This method uses the rth order moment of the eigenvalues to build a test statistic so that the expression of the MDL criterion has been improved. By simulating the method when the order is 2, 3, 4,5, respectively, it shows that the new method has greatly improved the performance of the multi-target detection in large array MIMO radar.