Abstract

We describe a parametric model for simulation of an experimental lenslet-array processor. The model parameters succintly quantify the performance of the optical system and are estimated from experimental data. The parametric model specifically examines nonuniformity, nonlinearity, local (physically adjacent) element cross talk, global (not physically adjacent) element cross talk, and time-variation of output. The parameters can be set to yield ideal performance or can approximate the experimental optical system. Characterization data from a previously described experimental lenslet-array optical neural network was used in an analytical and numerical best-fit estimation of the parameters. In particular, we consider the results of an optical interconnection as an overdetermined linear system of equations, with the model parameters serving as unknowns. A linear least-squares numerical procedure then yields minimum-error estimates of the parameters of the model. The errors of the model and the validity of underlying assumptions are examined. Additional techniques of optical processor characterization are described. Suggestions for improving the accuracy and increasing the level of detail of the model follow. Generalization of the model to other optical systems is proposed.

© 1992 Optical Society of America