ANALYSIS OF THE N-TO-N VOLUME-HOLOGRAPHIC NEURAL INTERCONNECT

A theoretical study of a weighted N-to-N volume-holographic interconnect is made. Coupled-wave solutions are derived that describe the behavior of an idealized version of the interconnect. System efficiency and fidelity are studied. For a restricted set of weight values, accurate implementation of the desired interconnect pattern is shown to be possible, apart from a simple error term dependent on position. However, significant departures from fidelity can occur for general weightings. Emphasis is placed on a prescribed (off-line) learning architecture, although the theory lays the groundwork needed for the analysis of adaptive networks using volume interconnects. The paper also complements studies of 1-to-N and N-to-1 holographic systems.