Researchers have proposed a great number of optical matrix processing
architectures and devices, each having significant advantages and disa
dvantages. However, no general metrics have been established to compar
e processors, and there has been little work to implement optical matr
ix processors for specific applications. In this article, we examine t
he range of proposed optical matrix processors both to determine the m
ost promising classes of applications and, by establishing performance
criteria, to select the best architectures and algorithms for these a
pplications. This examination concludes that the greatest current pote
ntial is for analog optical matrix-vector processors to analyze slowly
varying or invariant systems with high data throughput rates. (C) 199
6 John Wiley & Sons, Inc.