A fast implementation of the complex singular value decomposition on the Connection Machine

An algorithm suitable for signal processing applications of the complex singular value decomposition (SVD), and its implementation on the Connection Machine, are described. Many approaches to high-resolution array signal processing require the computation of a complex SVD. For small matrices, the SVD may be computed in a reasonable amount of time on very fast uniprocessors. For larger matrices, a special-purpose systolic array may be designed to compute the SVD, at the expense of programmability. As an alternative, massively parallel machines such as the Connection Machine may be candidates for performing high-resolution signal processing on large arrays