End-to-End Deep Learning in Phase Noisy Coherent Optical Link

TL;DR

This paper introduces an end-to-end deep learning framework based on autoencoders to mitigate laser phase noise in coherent optical OFDM systems, enabling high spectral efficiency and improved error rates.

Contribution

It presents a novel autoencoder-based model that effectively reduces phase noise effects in CO-OFDM, especially with low-cost lasers, simplifying traditional mitigation techniques.

Findings

Reduces bit error rate below FEC threshold under severe phase noise

Enables use of high-order modulation with large FFTs

Demonstrates effectiveness with low-cost lasers

Abstract

In coherent optical orthogonal frequency-division multiplexing (CO-OFDM) fiber communications, a novel end-to-end learning framework to mitigate Laser Phase Noise (LPN) impairments is proposed in this paper. Inspired by Autoencoder (AE) principles, the proposed approach trains a model to learn robust symbol sequences capable of combat LPN, even from low-cost distributed feedback (DFB) lasers with linewidths up to 2 MHz. This allows for the use of high-level modulation formats and large-scale Fast Fourier Transform (FFT) processing, maximizing spectral efficiency in CO-OFDM systems. By eliminating the need for complex traditional techniques, this approach offers a potentially more efficient and streamlined solution for CO-OFDM systems. The most significant achievement of this study is the demonstration that the proposed AE-based model can enhance system performance by reducing the bit error rate (BER) to below the threshold of forward error correction (FEC), even under severe phase noise conditions, thus proving its effectiveness and efficiency in practical deployment scenarios.