A Linear Algebraic Framework for Autofocus in Synthetic Aperture Radar

Synthetic Aperture Radar;Autofocus;Electrical Engineering;Engineering;Electrical Engineering: Systems

Liu, Kuang-HungWiesel, Ami ;

University of Michigan

关键词: Synthetic Aperture Radar;    Autofocus;    Electrical Engineering;    Engineering;    Electrical Engineering: Systems;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/86443/khliu_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Synthetic aperture radar (SAR) provides a means of producing high-resolution microwaveimages using an antenna of small size. SAR images have wide applicationsin surveillance, remote sensing, and mapping of the surfaces of both the Earth andother planets. The defining characteristic of SAR is its coherent processing of datacollected by an antenna at locations along a trajectory in space. In principle, we canproduce an image of extraordinary resolution. However, imprecise position measurementsassociated with data collected at each location cause phase errors that, in turn,cause the reconstructed image to suffer distortion, sometimes so severe that the imageis completely unrecognizable. Autofocus algorithms apply signal processing techniquesto restore the focused image.This thesis focuses on the study of the SAR autofocus problem from a linear algebraicperspective. We first propose a general autofocus algorithm, called Fourier-domainMultichannel Autofocus (FMCA), that is developed based on an image supportconstraint. FMCA can accommodate nearly any SAR imaging scenario, whetherit be wide-angle or bistatic (transmit and receive antennas at separate locations). Theperformance of FMCA is shown to be superior compared to current state-of-the-artautofocus techniques.Next, we recognize that at the heart of many autofocus algorithms is an optimizationproblem, referred to as a constant modulus quadratic program (CMQP). Currently,CMQP generally is solved by using an eigenvalue relaxation approach. We propose analternative relaxation approach based on semidefinite programming, which has recentlyattracted considerable attention in other signal processing applications. Preliminaryresults show that the new method provides promising performance advantages at theexpense of increasing computational cost.Lastly, we propose a novel autofocus algorithm based on maximum likelihood estimation,called maximum likelihood autofocus (MLA). The main advantage of MLA isits reliance on a rigorous statistical model rather than on somewhat heuristic reverse engineeringarguments. We show both the analytical and experimental advantages ofMLA over existing autofocus methods.