Reservoir Computing for Few-Mode Fiber Channel OSNR Monitoring in Deep Learning Frameworks

Abstract

This paper presents a novel optical performance monitoring (OPM) scheme based on reservoir computing (RC) and Resnet network for monitoring the optical signal-to-noise ratio (OSNR) of few-mode transmission channels, without the need for any demodulation process. By using 300 RC nodes, the number of floating-point operations (FLOPs) is reduced by 75%, with almost no change in prediction accuracy compared to a network without RC. In a system ranging from 0 to 30 dB, 40 simulation runs yield an OSNR prediction accuracy band around 0.9900. We also investigate the prediction accuracy when using different spectral information, with results showing that frequency-domain information effectively captures the characteristics of both signal and noise. Additionally, we examine the impact of different mode combinations on OSNR prediction accuracy and find that the prediction performance is nearly unaffected by the mode combination.

Keywords

• Optical performance monitoring (OPM)
• optical signal-to-noise ratio (OSNR)
• reservoir computing (RC)
• Resnet
• deep learning
• mode coupling