Abstract

Real-time classification of wide instantaneous bandwidth temporal signals such as radar range profiles or wideband communications signals present a challenging application to demonstrate the capabilities of adaptive, feature-based optical processing systems. In this paper we discuss progress on a nonlinearly cascaded optical neural classifier for this application. The system uses a holographic optical learning subsystem to classify time-shift invariant features computed from input wideband signals. The optical subsystems are cascaded using an optically addressed spatial light modulator (OASLM) implementing a saturating square-law nonlinearity. Through the use of two-dimensional, optically-computed trilinear input feature we take full advantage of the high space-bandwidth and high throughput processing capabilities specific to optical architectures to solve an otherwise intractable real-time signal-processing task such as wideband signal recognition. In what follows an experimental demonstration of this cascaded system is shown.

© 1995 Optical Society of America