Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Recent Advances on Machine Learning-aided DSP for Short-reach and Long-haul Optical Communications

Laurent Schmalen, Vincent Lauinger, Jonas Ney, Norbert Wehn, Patrick Matalla, Sebastian Randel, Alexander von Bank, Eike-Manuel Edelmann

TL;DR

This paper highlights both algorithmic advances as well as implementation aspects using conventional and neuromorphic hardware in the use of machine learning for implementing equalizers for optical communications.

Abstract

In this paper, we highlight recent advances in the use of machine learning for implementing equalizers for optical communications. We highlight both algorithmic advances as well as implementation aspects using conventional and neuromorphic hardware.

Recent Advances on Machine Learning-aided DSP for Short-reach and Long-haul Optical Communications

TL;DR

This paper highlights both algorithmic advances as well as implementation aspects using conventional and neuromorphic hardware in the use of machine learning for implementing equalizers for optical communications.

Abstract

In this paper, we highlight recent advances in the use of machine learning for implementing equalizers for optical communications. We highlight both algorithmic advances as well as implementation aspects using conventional and neuromorphic hardware.

Paper Structure

This paper contains 5 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Novel ML-inspired Cost Functions for Conventional Equalizers
  3. Novel ML-based Equalizers and Their Hardware Implementation
  4. Energy-efficient Equalizers with Spiking Neural networks
  5. Conclusion

Figures (3)

  • Figure 1: Left: simulation results of the VAE-LE and the CMA for the transmission of 64-QAM with probabilistic constellation shaping (PCS) (entropy $\mathcal{H}=4.6$) over a linear dispersive optical dual-polarization channel at a symbol rate of $40 \ \mathrm{GBd}$ (dotted) and $90 \ \mathrm{GBd}$ (dashed) lauinger2022blind. Right: block diagram of the VAE based equalizer.
  • Figure 2: Left: exploration of the performance complexity trade-off of CNN, Volterra, and linear equalizers for IM/DD transmission with pulse-amplitude modulation (PAM)-2 at $40 \ \mathrm{GBd}$. The red dashed vertical line approximates the complexity limit on the AMD XCVU13P FPGA neySAMOSJournal. The lines connecting the points give the Pareto fronts and the circle marks the Pareto optimal model for this system. Right: Transmission setup of an FPGA implementation demonstrating real-time ML-based equalization Ney24ICMLCN.
  • Figure 3: Feed-forward SNN-based equalizer for short-reach optical communications according to Bank24OFC.