Binarized Coherent Optical Receiver Based on Opto-Electronic Neural Network

Abstract

The resolution of analog to digital converter (ADC) and the power consumption of digital signal processing (DSP) are becoming the bottleneck of optical communication system. To acquire and recover signals with low resolution ADCs, we propose a binarized coherent optical receiver based on opto-electronic neural network, which consists of three layers, the broadcast layer, the optical neural layer, and the electronic neural layer. The broadcast layer that composed of Mach–Zehnder interferometers (MZIs) is to duplicate and select the input optical signal into the opto-electronic neural network. The optical neural layer is constituted by specially designed structures to map the binarized weights into optical domain, which can be integrated on silicon together with the broadcast layer. The one-bit vertical resolution ADCs act as the binarized activation after the optical neural layer. This binarized coherent optical receiver is demonstrated in simulation of 50-Gb/s single polarization (SP) QPSK system and 100-Gb/s polarization multiplexing (PDM) QPSK system. The one-bit vertical resolution ADCs can be utilized to recover the complex modulation format signal successfully. The broadcast layer and the optical neural layer can be integrated based on silicon, which could increase the calculation speed and reduce the energy consumption of the total system.