Opto-electronic machine learning network for Kramers-Kronig receiver linearization

Abstract

We numerically demonstrate the use of an opto-electronic network comprising a photonic reservoir and an electronic feedforward equalizer (FFE) to linearize a Kramers-Kronig (KK) receiver. The KK receiver is operated under stringent conditions, with restricted sampling rates and low carrier powers, resulting in a nonlinear behavior. We propose two different network configurations, varying in the placement of the FFE component, and evaluate their ability to linearize the KK receiver. By training these networks on back-to-back systems, we arrive at a generic solution that significantly enhances the receiver performance, independent of specific link characteristics. The trained networks are tested in a plug-and-play manner across diverse short-reach links that employ standard digital signal processing blocks for the qualization. Our results show significant improvement in receiver linearity, resulting in a reduction in bit error rate of up to a factor of four.