Abstract

The accuracy of matrix computations performed on high-speed analog optical associative processors (OAPs) is limited by the noise and spatial errors. There exist two different approaches for alleviating this limitation: (i) postprocessing with a bimodal system [1] and (ii) preprocessing with a preconditioner [2,3]. In this talk we show that these two approaches can be combined to develop an "intelligent" optical processor that can adapt the computational steps depending on the data and produce accurate solutions at a high speed.

© 1989 Optical Society of America