Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fully-blind Neural Network Based Equalization for Severe Nonlinear Distortions in 112 Gbit/s Passive Optical Networks

Vincent Lauinger, Patrick Matalla, Jonas Ney, Norbert Wehn, Sebastian Randel, Laurent Schmalen

TL;DR

This study tackles reliable ultra-high-rate PON upstream transmission by developing a fully-blind, adaptive neural-network equalizer using a VQVAE-inspired loss to mitigate severe nonlinear distortions without pilots. It systematically compares small, hardware-friendly CNN and GRU topologies against a baseline FIR, aided by a channel-estimator network to support blind adaptation. Experiments at 56 Gbaud PAM4 over 2.2 km show NN-based equalizers achieve lower BER than FIR, with topology-dependent performance and blind learning approaching the non-blind MSE benchmark. The results indicate practical, FPGA-implementable solutions for cost-effective 100G-PON uplinks with reduced pilot overhead and robust adaptation to impairments.

Abstract

We demonstrate and evaluate a fully-blind digital signal processing (DSP) chain for 100G passive optical networks (PONs), and analyze different equalizer topologies based on neural networks with low hardware complexity.

Fully-blind Neural Network Based Equalization for Severe Nonlinear Distortions in 112 Gbit/s Passive Optical Networks

TL;DR

This study tackles reliable ultra-high-rate PON upstream transmission by developing a fully-blind, adaptive neural-network equalizer using a VQVAE-inspired loss to mitigate severe nonlinear distortions without pilots. It systematically compares small, hardware-friendly CNN and GRU topologies against a baseline FIR, aided by a channel-estimator network to support blind adaptation. Experiments at 56 Gbaud PAM4 over 2.2 km show NN-based equalizers achieve lower BER than FIR, with topology-dependent performance and blind learning approaching the non-blind MSE benchmark. The results indicate practical, FPGA-implementable solutions for cost-effective 100G-PON uplinks with reduced pilot overhead and robust adaptation to impairments.

Abstract

We demonstrate and evaluate a fully-blind digital signal processing (DSP) chain for 100G passive optical networks (PONs), and analyze different equalizer topologies based on neural networks with low hardware complexity.
Paper Structure (4 sections, 4 figures)

This paper contains 4 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Blind Adaptive NN-based Equalizers
  3. Experimental Setup and Results
  4. Conclusion

Figures (4)

  • Figure 1: Experimental setup for the 112Gbit/s (56GBd) PON upstream through a *SSMF in the C-band.
  • Figure 2: Median $\overline{\text{BER}}$ for a fiber length of 2.2km (C-band) with NN of 51 *rvm (a) and 181 rvms (b) .
  • Figure 3: Median $\overline{\text{BER}}$ for optical B2B.
  • Figure 4: Median $\overline{\text{BER}}$ (markers) and the corresponding error bars (to the best and worst estimates per 15 captured sequences) at 2.2km, -2dBm for the proposed equalizers, trained by the blind VQVAE with different channel estimator nets.