Neural processing-type optical wavelength division multiplexing demultiplexer using a grating

A neural processing-type optical wavelength division multiplexing (WDM) demultiplexer (NP-ODEMUX), which consists of diffraction grating, detector array, and neural network LSI, is proposed in order to improve receiver sensitivity while maintaining flexible demultiplexing characteristics. In previous work, a neural processing-type optical WDM demultiplexer using a multimode optical waveguide was proposed. This conventional NP-ODEMUX and the proposed grating-type NP-ODEMUX are compared using power penalty calculation simulations. The results show that receiver sensitivity is improved using the proposed grating-type NPODEMUX. Also, a standard diffraction grating demultiplexer without neural processing is compared with the proposed NP-ODEMUX using power penalty and it is shown that the proposed NP-ODEMUX can demultiplex WDM signals without degrading receiver sensitivity. Next, the results of the demultiplexing experiment for six-channel WDM signal which has been directly modulated at 150 Mb/s using nonreturn to zero (NRZ) pseudorandom signal by grating-type NP-ODEMUX prototype are described. Bit-error rate measurement for both the conventional diffraction grating demultiplexer and the grating-type NP-ODEMUX using neural processing for the unequally spaced WDM signals revealed that the receiver sensitivity can be greatly improved and that it can flexibly compensate wavelength change.