Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Novel Machine Learning-based Equalizer for a Downstream 100G PAM-4 PON

Chen Shao, Elias Giacoumidis, Shi Li, Jialei Li, Michael Faerber, Tobias Kaefer, Andre Richter

TL;DR

This work tackles the challenge of high-rate downstream 100G PAM-4 PON where chromatic dispersion and related impairments degrade BER. It introduces FC-SCINet, a frequency-calibrated SCINet that couples spectral calibration with temporal modeling to compensate both spectral and temporal distortions. Empirical results show substantial BER improvements over conventional FFE and a 3-layer DNN (up to $87.5\%$ at 11 km in CD and $88.87\%$ at 5 km in a realistic scenario) and a $10.577\%$ reduction in complexity, demonstrating robust performance against EAM chirp, jitter, and Kerr nonlinearity. The approach offers a practical path toward deploying high-rate PON links with manageable complexity.

Abstract

A frequency-calibrated SCINet (FC-SCINet) equalizer is proposed for down-stream 100G PON with 28.7 dB path loss. At 5 km, FC-SCINet improves the BER by 88.87% compared to FFE and a 3-layer DNN with 10.57% lower complexity.

A Novel Machine Learning-based Equalizer for a Downstream 100G PAM-4 PON

TL;DR

This work tackles the challenge of high-rate downstream 100G PAM-4 PON where chromatic dispersion and related impairments degrade BER. It introduces FC-SCINet, a frequency-calibrated SCINet that couples spectral calibration with temporal modeling to compensate both spectral and temporal distortions. Empirical results show substantial BER improvements over conventional FFE and a 3-layer DNN (up to at 11 km in CD and at 5 km in a realistic scenario) and a reduction in complexity, demonstrating robust performance against EAM chirp, jitter, and Kerr nonlinearity. The approach offers a practical path toward deploying high-rate PON links with manageable complexity.

Abstract

A frequency-calibrated SCINet (FC-SCINet) equalizer is proposed for down-stream 100G PON with 28.7 dB path loss. At 5 km, FC-SCINet improves the BER by 88.87% compared to FFE and a 3-layer DNN with 10.57% lower complexity.
Paper Structure (6 sections, 3 equations, 2 figures, 1 table)

This paper contains 6 sections, 3 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Evaluation Setup
  3. FC-SCINet Model
  4. Results and Discussion
  5. Conclusion
  6. Acknowledgement

Figures (2)

  • Figure 1: (a) Block diagram illustrating the innovative FC-SCINet equalizer. (b) Time/Frequency domain representation of 50 transmitted (target)/received consecutive samples, showing separately the impact of FC and SCInet.
  • Figure 2: (a),(b) BER vs. distance for FC-SCInet, DNN, FFE and w/o equalization for Cases 1-2 (CD-Realistic). (c) Received constellation diagram for Case 1 (CD) at 9 km. (d) BER color map for FC-SCInet's window size and levels of Interactors for Case 1 (CD) at 9 km.