Range Dependent Phase Gradient Autofocus - Geoscience and Remote Sensing Symposium Proceedings, 1998. IGARSS '98. 1998 IEEE International

The Phase Gradient Autofocus (PGA) algorithm has been widely used in Spotlight Synthetic Aperture Radar (SAR) to remove motion-induced blurs in the images. The PGA algorithm has been proven to be a superior autofocus method. PGA assumes a narrow beam, which is valid for most SAR systems. However, lower altitude SA& have large range dependencies that cannot be ignored. A new phase estimator for PGA is introduced and extended to allow range dependence. An ERS-1 image of Death Valley is used in simulations comparing the new estimator to the widely used maximum likelihood approach and in demonstrating the range-dependent PGA algorithm. INTRODUCTIONANDBACKGROUND Full focusing of SAR images requires some type of autofocus routine. The Phase Gradient Autofocus (PGA) algorithm has proven to be a superior method for higher order autofocus because it does not assume a model for the phase error. The standard PGA model assumes a small beamwidth in range, which results in a phase error constant in the range direction. Most satellites and other high altitude systems fit this model. However, a low-altitude SAR like YSAR [l] will have range-dependent phase errors. In this paper we extend the algorithm by dropping the narrow beam assumption and introducing range dependencies in the phase error. There are four main steps in the PGA algorithm. The four steps are center shifting, windowing, phase estimation and iteration. These steps are described in detail in [2, 3, 41. The Phase Weighted Estimation PGA (PWE-PGA) proposed here differs from earlier algorithms only in the phase estimation step. A few methods have been proposed for the phase estimation step, with different criteria for optimality. The original PGA algorithm used a linear unbiased minimum variance [2]. The same authors later proposed a method using a maximum likelihood (ML) estimator [3]. This paper proposes a new phase estimation technique which allows extension to a range-dependent algorithm.