Research On Angle Estimation Algorithm In MIMO Radar Based On Tensor Algebra

Recently, Multiple-input Multiple-output radar has become one of research hits at home and abroad, and angle estimation is one of the important research contents. However, the current research on MIMO radar mostly use uniform linear array as the array configuration,which can only estimate one-dimensional angle of the target, while MIMO radar-based L-shaped array configuration can estimate two-dimensional angle information of the target,and has more practical application values. Therefore, based on the existing angle estimation algorithms of L-shaped MIMO radar, some DOA estimation methods combined with tensor algebra and non-circular signals, which have more superior performance, are proposed in this dissertation.Firstly, the array structure and signal model of L-shaped MIMO radar are constructed,and the principle of virtual array and the maximum resolution target numbers are analysed.Then the basic principle of traditional angle estimation algorithms are introduced,including MUSIC algorithm, Root-MUSIC algorithm and ESPRIT algorithm, and study the pros and cons of these algorithms..Secondly, the basic computation of tensor algebra are introduced, including the unfolding matrix of tensor, the multiplication between tensor and matrix, the outer product and scalar product of tensor, and the higher order singular value decomposition(HOSVD) and canonical decomposition of tensor; Then exploiting the structural characteristics of MIMO radar matched filter output data, signal model based the tensor is established for L-shaped array and the angle estimation algorithms based on tensor decomposition are proposed. For MUSIC class-based angle estimation algorithm, the 2D-MUSIC algorithm using higher order singular value decomposition is analyzed, but the algorithm brings a huge computational burden. In order to solve this problem, the reduced-dimension MUSIC algorithm based on the higher order singular value decomposition of tensor structure data of L-shaped MIMO radar is proposed, and the two-dimensional spectral peak searching of 2D-MUSIC algorithm is reduced to one-dimensional search, so the computational complexity is reduced and due to the exploiting of the structural information of L-shaped MIMO radar output data, the angle estimation performance is more superior, especially in the case of small samples. For ESPRIT class-based algorithm, the original complex-valued tensor are transformed to real-value by unitary transformation process, then the angles are estimated by tensor higher-order singular value decomposition, this algorithm can not only effectively reduce the complexity of computation, but also achieve an improved performance of angle estimation.Again, the tensor canonical decomposition algorithm based on the L-shaped MIMO radar is proposed. Firstly, canonical decomposition is applied to the tensor structure data, then the signal model is established and the uniqueness of canonical decomposition is analyzed,and the factor matrixs are computed by the alternately least squares algorithm. For the computation problem in calculation process, the factor matrices are estimated by the core tensor which has a lower dimension and then the original matrices are recovered based on the estimated factor matrixs, and finally the target angles are estimated by the Vandermonde property of factor matrixs.Finally, with the advantages of non-circular signals which can extend the virtual array aperture, the angle estimation algorithms based on tensor decomposition and non-circular signal in L-shaped MIMO radar are proposed. On the basis of L-shaped MIMO radar transmitting non-circular signals, the received tensor structure data are extended, and ESPRIT and UESPRIT algorithms are applied to estimate the two-dimensional angle informations.