A comparison of spotlight synthetic aperture radar image formation techniques

Spotlight synthetic aperture radar images can be formed from the complex phase history data using two main techniques: (1) polar-to-cartesian interpolation followed by two-dimensional inverse Fourier transform (2DFFT), and (2) convolution backprojection (CBP). CBP has been widely used to reconstruct medical images in computer aided tomography, and only recently has been applied to form synthetic aperture radar imagery. It is alleged that CBP yields higher quality images because (1) all the Fourier data are used and (2) the polar formatted data is used directly to form a 2D Cartesian image and therefore 2D interpolation is not required. This report compares the quality of images formed by CBP and several modified versions of the 2DFFT method. We show from an image quality point of view that CBP is equivalent to first windowing the phase history data and then interpolating to an exscribed rectangle. From a mathematical perspective, we should expect this conclusion since the same Fourier data are used to form the SAR image. We next address the issue of parallel implementation of each algorithm. We dispute previous claims that CBP is more readily parallelizable than the 2DFFT method. Our conclusions are supported by comparing execution times between massively parallel implementations of both algorithms, showing that both experience similar decreases in computation time, but that CBP takes significantly longer to form an image