Proposal of frequency-domain multiplexing in optical neural networks

Abstract

Frequency-domain multiplexing in optical neural networks is proposed for achieving higher parallelism. The neural system forms coherent-type optical circuits utilizing self-homodyne detection. Optical-frequency-dependent learning process is realized by adjusting delay time and transparency of optical neural connections. A simulation experiment demonstrates that the network learns desired output signals depending on the optical frequencies. This result provides the neural network systems with new degree of parallelism; i.e. multiplexing in optical carrier-frequency domain.