Research On Related Technology Of Compressive Sensing Based SAR

Compressive sensing (CS) is a novel theory in digital signal processing. If certain conditions are met, the original signal can be recovered nearly perfectly with a very high probability from the data, of which the sampling rate is much lower than the Nyquist sampling rate. And the data rate of the entire system, even the sampling rate of the system front-end, can be reduced to some extent. The theory of CS has brought about a revolution for data acquisition. It has very important practical significance. In this paper, the measurement matrix of the CS is studied. And for the application of synthetic aperture radar (SAR) imaging, the theory and method of CS is taken into consideration.. A processing method using CS is then investigated with emphasis on the one-dimensional range imaging about the targets of linear characteristics.In the first two chapters, the introduction and overview of the basic principles and applications of CS are given in this paper. The three important aspects in the procedure of CS, namely, sparsity of the signal, measurement matrix and the recovery algorithms, are all discussed. General examples of the three important aspects are also presented.Chapter3is focused on the measurement matrix. Several general measurement matrices are introduced. And the self-correlation of eight kinds of measurement matrices is studied according to the requirements of CS,. The figures of the relation between the number of measurements and reconstruction probabilities are presented by simulation, and further convincing comparisons are given in terms of the matrices’ performances. To meet the practical needs of less storage, decreased computation complexity and performance improvements, several possible structures of matrices are proposed. And each structure’s performance is verified through simulation experiments.In chapter4, a design recommendation of the measurement matrix is proposed from the view point of recovery algorithms. Firstly, the orthogonal matching pursuit (OMP) which is the most classic recovery algorithm is introduced, and the requirement that a recovery matrix should meet is given from the perspective of OMP. Then following the relation between the recovery matrix and the measurement matrix, the binarization of the measurement matrix, which can improve the performance, is derived, and the comparison with the original matrix is presented. After binarization, the recovery performance of measurement matrices can be improved to a certain extent. And most importantly the storage of the measurement matrices is greatly reduced. So the binarization is very significant for practical application.A procedure of SAR imaging based on compressive sensing is given in chapter5, and investigations focuing on the anti-noise performance of the method are carried out. After the mathematical model is written in the form of linear equations, CS is used in the procedure with the consideration of the sparsity of the problem. Then during the data processing, the traditional matched filtering method in azimuth is combined with the CS algorithm in range direction. The experiments verify the good anti-noise performance of the method. The sidelobe in the range direction has distinct improvement, while the amount of data of the system is also reduced.In the final chapter, a set of bases, which is based on the linear features of the target in the range profile, is given for the data processing in the one-dimensional range imaging, and an adaptive recovery method is presented for reconstruction. The proposed sparse domain makes the problem become more sparse. Compared to the conventional domain, it has a better performance in solving the linear target problem. But the bases need too much information about the targets, so it is difficult for practical use. In order to leverage the difficulty, an adaptive recovery method is proposed for reconstruction. The sparse domain is revised every time by using the matching information contained in every recovery results, so as to find the more appropriate sparse domain, and to solve the recovery issue. The simulation results verify that the method has very good performance on dealing with linear targets, especially when the lengths of the targets are very long. Furthermore, it can also resolve the multi-target issue.