Abstract

We describe the implementation of a holographic heteroassociative memory and the modeling of the nonlinear dynamics of the system. The system uses thermoplastic plates for storing associative images and a Hughes liquid crystal light valve (LCLV) for image amplification and thresholding. Planar holographic interconnections are used for interconnecting patterns from one layer to another. The problem of degenerate readouts in planar holographic interconnections is analyzed and the technique of using sampling grids for specifying unique interconnections is described. The optical system is arranged as a feedback loop. It locks the recognized input and its associative image as a stable state in the loop. We show that the main factors determining the dynamics of the system are the optical gain and the similarity between input and the stored images. The optical system is equivalent to a two-layer network with feedback. Equations describing the dynamics of the system are derived, including the effects of the square-law detection by the neural plane. A geometric method is introduced which allows us to analyze how the system state evolves and how the parameters influence the evolution.

© 1989 Optical Society of America