Study On Ultra-high Resolution SAR Imaging Algorithms

Synthetic Aperture Radar (SAR) is an imaging system which could achieve fine-resolution images in two dimensions. It has been widely used in military and civil application. Improving the resolution is meaningful to targe identification, feature extraction, and so on. Consequently, there is a growing demand for SAR systems with an ultra-high resolution in the recent couple of years.The core of SAR signal processing is image formation algorithm. In ultra-high resolution imaging, a very long aperture is required to achieve the high resolution in azimuth. Since motion errors are inevitable in such a long aperture, convolution backprojection algorithm (CBP) and polar format algorithm (PFA) are superior to other algorithms for their nonplanar motion compensation capability. Unfortunately, they have shortcomings respectively. Traditionally, CBP is too computationally expensive. The classic PFA requires frequency-domain interpolation, and due to wavefront curvature, it has limits on the scene size. On the other hand, large bandwidth is required to achieve the high resolution in range. Wide bandwidth management is a great technological problem in current ultra-high resolution SAR system, and this problem can be solved by adopting stepped chirps and applying synthetic bandwidth approach. However, traditional synthetic bandwidth technique could not completely avoid phase error during pulse synthesis. In this dissertation, the existing algorithms are further improved and extended to solve the high resolution imaging problems presented above.Chapter 1 is the introduction. The history of SAR is introduced firstly. Then the conception of ultra-high resolution SAR is given, including its latest developments in China and abroad. Lastly, the key problems during research are presented and the main contents of this thesis are outlined.In chapter 2, the fast implementation of CBP is investigated. Firstly, the foundation of CBP, the image reconstruction principles in computer-aided tomography (CAT) is introduced. Then, the CBP algorithm as applied in SAR is interpreted. There is no scene size limitation since CBP could compensate the wavefront curvature exactly. On the other hand, CBP could also adopt planar-wavefront assumption when the scene is rather small. To raise the processing efficiency of CBP, the fast implementation methods are respectively investigated under those two conditions. When considering wavefront curvature, a fast implementation of CBP based on recursively decomposing the image like a quadtre is presented. It uses sub-image truncation and FFT to implement the low-pass filtering and sampling decimation during the recursive decomposition. When the scene is small enough to apply the planar-wavefront assumption, a simple fast backprojecting method in the image domain is proposed, which uses a series of FFTs and multiplications to realize the resampling in'backprojection', and could substantially avoid the interpolation operation. These two methods both result in a great speed-up for the standard convolution backprojection.In chapter 3, a novel efficient implemention of PFA based on the principle of chirp scaling (PCS) is investigated, which could totally avoid interpolation. Firstly, the principle of PFA is briefly introduced, and it is induced that the polar format transform could be implemented via a range scaling and an azimuth scaling. Based on the scaling transform interpretation, the processing flow of the PCS based PFA is derived in detail. Finally, the conventional interpolation based PFA and PCS based PFA are compared from the viewpoint of image quality and processing speed.Due to the wavefront curvature, the classic PFA has limits on the scene size since defocus would appear beyond the limited scene size. Based on the idea of digital spotlighting, the scene size limitation could be broken down by using the sub-image strategy. In chapter 4, two sub-image based methods are proposed to make PFA applicable in the broad-scene situation. Firstly, an implementation of PFA based on recursively decomposing the image like a quadtre is presented. Then, a more efficient and more precise method, the two-hierarchy PFA algorithm, is proposed. As long as the sub-image extracted from the whole image is controlled in the limited scene size of traditional PFA, there would be no defocus. Via geometric distortion correction, each sub-image could be seamed together to produce a full image with no defocus, achieving the goal of enlarging the valid scene size of PFA. In addition, to apply PCS based PFA here to avoid interpolation and raise the processing efficiency, several important problems which we need to pay attention to during the implementation are picked out and interpreted briefly.The application of synthetic bandwidth technique in PFA is investigated in chapter 5. The waveform model of stepped chirps is given firstly. There are three classic synthetic bandwidth methods, including a time domain method, a frequency domain method, and a time domain method applied to deramping receiving mode. This chapter applies these three methods in PFA, solves the motion compensation problem between the sub-pulses within a burst, and demonstrates the signal processing flows respectively in detail. Nevertheless, some minor phase error is inevitable during the pulse synthesis. To eliminate this disadvantage, a modified PFA is presented, which does not need pulse synthesis while the bandwidth synthesis is embedded in the resampling of polar format transform. Furthermore, we are dedicated in utilizing PCS to realize the resampling in this modified PFA, obtaining a more efficient solution, in which only FFT's and complex vector multiplications are involved. In the end, the research work of this thesis is concluded and some other aspects to be further studied are pointed out as well.