Abstract

We have borrowed a few concepts from neural-network modeling to implement a class of guided wave frequency demultiplexers and switches that adapt themselves to their input signals. The adaptation feature is useful in demultiplexing cases where the signal carrier frequencies are a priori unknown, or where the frequencies are subject to drifts. Given two carrier frequencies ω₁ and ω₂ copropagating along a fiber, for example, the demultiplexer's task is to separate the two channels. In the present case, the demultiplexing is done through a nonlinear competitive process that takes place in a multimode photorefractive ring resonator. Each of (in this case) two modes associates itself with one of the carrier frequencies while rejecting the other. In a slightly different configuration, the demultiplexer becomes a switch for which the destination is determined by a spatial pattern imposed by a transmitter using a multimode waveguide. Although our experimental work uses bulk optical components, the schemes are particularly well suited to integrated approaches.

© 1991 Optical Society of America