Fiber-induced nonlinearity compensation in coherent optical systems by affinity propagation soft-clustering

TL;DR

This paper introduces a novel, system-agnostic fiber nonlinearity compensator using affinity propagation clustering, significantly improving signal quality in coherent optical systems without additional training data.

Contribution

It presents the first experimental demonstration of a training-data-free, system-agnostic NLC method based on affinity propagation clustering, outperforming existing algorithms and schemes.

Findings

Increases signal quality-factor by up to 5 dB in 16-QAM systems

Outperforms benchmark clustering algorithms and complex deterministic schemes

Extends power margin by up to 4 dB

Abstract

There is a tremendous need to maximize signal capacity without sacrificing energy consumption and complexity in optical fiber communications due to the rise of network traffic demand and critical latency-aware services such as telemedicine and the Internet-of-Things. In this work, the first system-agnostic and training-data-free fiber nonlinearity compensator (NLC) using affinity propagation (AP) clustering is experimentally demonstrated. It is shown that compared to linear equalization, AP can increase the signal quality-factor up to about 5 dB in 16-quadrature-amplitude-modulated coherent optical systems by effectively tackling deterministic and stochastic nonlinearities. AP outperforms benchmark clustering algorithms and complex deterministic schemes such as K-means, fuzzy-logic, digital-back propagation and Volterra-based NLC, offering a power margin extension of up to 4 dB.