On SAR Processing Using Pixel-wise Matched Kernels

The recent increase in computing resources allows the emergence of new algorithms, the reappraisal of conventional algorithms, and even the reinvention of formerly computationally infeasible algorithms. This holds true for synthetic aperture radar (SAR) which is nowadays a widely-employed technology for imaging at distance in diverse operating conditions. In this paper, we discuss a formulation for SAR image reconstruction that involves no approximations. A unique kernel function is produced for each pixel of the image. The proposed principle of the kernel formation allows using a direct integral transform from a measured polar domain to a Cartesian image domain; no additional processing is required for any measurement geometry or the size of the image. This approach provides a well-defined premise for the implementation of other essential techniques, related to e.g. using robust waveform and compensating unexpected motion, which are vital for high-quality SAR images in unideal operating conditions. We illustrate our approach with simulated and measured inverse synthetic aperture radar (ISAR) experiments.