Research On Distributed Radar Sparse Imaging Technologies

In this dissertation, the distributed radar imaging is defined as a novel technology which utilizes space expanded multiple transmitters and multiple receivers (or virtual transmitters and receivers) for efficient target observation. It integrates the real aperture imaging with the virtual aperture imaging, and generates the concepts of multiple-input multiple-output (MIMO) radar, then builds a unified description for imaging systems.Based on space spectrum theory, this dissertation analyzes three typical distributed radar systems, i.e. inverse synthetic aperture radar (ISAR), frequency diverse MIMO radar, and distributed passive radar, respectively. And it focuses on realizing the high resolution imaging methods in those systems under the situation of practical limitations, by combining informations of aperture, bandwidth, and target prior together. Besides, it is devoted to solving several problems which may be faced in practical application of distributed radar sparse imaging. The main works and contributions are presented as follows:Firstly, according to space spectrum theory, a unified imaging framework is established, and the concept of generalized aperture imaging is derived. Then, the relationship between aperture, bandwidth and imaging resolution is discussed, and the performances of three typical distributed radar systems are analyzed. Moreover, considering the practical limitations, the ways and the difficulties of applying the compressed sensing (CS) theory into the distributed radar sparse imaging are taken into account.Secondly, the distributed radar sparse optimization methods are studied. Since conventional sparse recovery methods have heavy complexities for real-time hardware implementation, we propose a novel homotopy DCD method with non-convex penalties, which has the advantages of high recovery performance and low calculation complexity, etc. In addition, conventional sparse recovery methods are not performing well in low SNR situation. To deal with this problem, firstly we propose a method based on Bayesian compressed sensing (VB-BCS) for single measurement vector (SMV) imaging, then for multiple measurement vector (MMV) imaging, a method based on sparse Bayesian learning with Laplace prior (Laplace-SBL) is provided. Simulation results show the performance improvement of the proposed methods.Thirdly, since the scatterers are distributed in a continuous scene, the off-grid problem inevitably exists in distributed radar sparse imaging. Therefore, we present three ways for solving the inherent calculating defects of CS-based imaging methods, which are listed as follow:1. sparse recovery techniques combined with adaptive mesh generation;2. MUSIC and modified Matrix Pencil imaging methods based on analog CS theory;3. two off-grid CS-based imaging methods, i.e. sparse autofocus calibration based on Bayesian compressed sensing (SAC-BCS), and the off-grid orthogonal matching pursuit (OG-OMP), respectively. Simulation results confirm that our methods are not sensitive to grid spacing, and exhibit a robust capability of target-information extraction.Fourthly, considering practical implementation of distributed radar system, the echo measurements are unavoidably perturbed by phase errors (usually caused by system errors), which would make sensing matrix unknown and lead to considerable performance degradation by conventional methods. Therefore, we propose two sparse autofocus imaging approaches in SMV and MMV imaging cases, including adaptive phase error calibration based on iterative optimization techniques (APEC-IOT) and adaptive phase error calibration based on Bayesian compressed sensing (APEC-BCS), respectively. Simulation results verify that our methods can achieve good autofocus qualities.Fifthly, the distributed radar sparse imaging for extended target are researched. For extended target, the corresponding scatterers in the imaging scene have a great number, and usually have regional distribution. Therefore, the target space is not sparse enough, which would result in the degradation of conventional CS-based methods. In addition to the sparsity constraint, we consider to utilize the structural information of extended target image further, and propose two novel imaging methods for extended target. The first one is by the combination of non-convex regularization and total variation (NCR-TV), and the second one is based on the distribution of extended target prior via Bayesian compressed sensing (ET-BCS). We demonstrate the effectiveness of the proposed methods through numerical simulations.