Abstract

We presented a complete statistical analysis of a generic optical linear algebra processor previously, determining the signal and noise statistics with no reference to the system component characteristics. In this paper we introduce device statistics into the analysis and evaluate output signal probability densities and power spectra for certain practical device combinations. Based on these statistics, we then apply M-ary hypotheses testing and Wiener filtering techniques to improve the processor accuracy. Both approaches are aided by proper normalizing transforms. Analytical and computer simulation results are presented to quantify the improvements, and optimum approaches are suggested for particular architectures and devices.

© 1992 Optical Society of America