Abstract

Regardless of the particular architecture chosen, certain operations such as multiplication, addition, and the generation, propagation, and detection of light must be performed in an optical linear algebra processor. We show that each of these operations introduces fundamental noise limitations which affect the overall accuracy of the processor. These limitations are examined for various types of light including polarized thermal light, unpolarized thermal light, and laser light, as well as for low light levels. The effects of various spatial light modulators on the accuracy of the multiplication process are considered. Likewise, the effect of the type of addition used (i.e., space integrating or time integrating) on the resulting accuracy is considered.

© 1989 Optical Society of America