Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A comparison between black-, grey- and white-box modeling for the bidirectional Raman amplifier optimization

Metodi P. Yankov, Mehran Soltani, Andrea Carena, Darko Zibar, Francesco Da Ros

TL;DR

The paper investigates how white-box, grey-box, and black-box modeling approaches perform in optimizing a bidirectional distributed Raman amplifier to flatten the frequency-distance power profile over the C-band. It compares four optimization strategies—CNN-based inverse mapping, gradient-free differential evolution, a CNN+DE hybrid, and a differentiable white-box Raman solver with gradient descent—on an 80 km link with 40×100 GHz channels, assessing them with loss criteria $J_0$, $J_1$, and $J_2$. Results show all methods achieving similar flatness (roughly 1–3.6 dB depending on the metric), with gradient-based and hybrid approaches typically delivering best $J_1$/$J_2$ when initialized by CNN predictions; CNN alone is fastest but data-limited and less generalizable. The work provides practical guidance on method selection, balancing speed, data needs, and adaptability to dynamic network conditions for offline amplifier optimization.

Abstract

Designing and optimizing optical amplifiers to maximize system performance is becoming increasingly important as optical communication systems strive to increase throughput. Offline optimization of optical amplifiers relies on models ranging from white-box models deeply rooted in physics to black-box data-driven and physics-agnostic models. Here, we compare the capabilities of white-, grey- and black-box models on the challenging test case of optimizing a bidirectional distributed Raman amplifier to achieve a target frequency-distance signal power profile. We show that any of the studied methods can achieve similar frequency and distance flatness of between 1 and 3.6 dB (depending on the definition of flatness) over the C-band in an 80-km span. Then, we discuss the models' applicability, advantages, and drawbacks based on the target application scenario, in particular in terms of flexibility, optimization speed, and access to training data.

A comparison between black-, grey- and white-box modeling for the bidirectional Raman amplifier optimization

TL;DR

The paper investigates how white-box, grey-box, and black-box modeling approaches perform in optimizing a bidirectional distributed Raman amplifier to flatten the frequency-distance power profile over the C-band. It compares four optimization strategies—CNN-based inverse mapping, gradient-free differential evolution, a CNN+DE hybrid, and a differentiable white-box Raman solver with gradient descent—on an 80 km link with 40×100 GHz channels, assessing them with loss criteria , , and . Results show all methods achieving similar flatness (roughly 1–3.6 dB depending on the metric), with gradient-based and hybrid approaches typically delivering best / when initialized by CNN predictions; CNN alone is fastest but data-limited and less generalizable. The work provides practical guidance on method selection, balancing speed, data needs, and adaptability to dynamic network conditions for offline amplifier optimization.

Abstract

Designing and optimizing optical amplifiers to maximize system performance is becoming increasingly important as optical communication systems strive to increase throughput. Offline optimization of optical amplifiers relies on models ranging from white-box models deeply rooted in physics to black-box data-driven and physics-agnostic models. Here, we compare the capabilities of white-, grey- and black-box models on the challenging test case of optimizing a bidirectional distributed Raman amplifier to achieve a target frequency-distance signal power profile. We show that any of the studied methods can achieve similar frequency and distance flatness of between 1 and 3.6 dB (depending on the definition of flatness) over the C-band in an 80-km span. Then, we discuss the models' applicability, advantages, and drawbacks based on the target application scenario, in particular in terms of flexibility, optimization speed, and access to training data.
Paper Structure (7 sections, 5 equations, 2 figures, 4 tables)

This paper contains 7 sections, 5 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Raman optimizers
  3. Setup
  4. Performance comparison
  5. Applicability and discussion
  6. Conclusions
  7. Acknowledgements

Figures (2)

  • Figure 1: The proposed distributed Raman amplifier setup with different pump power adjustment frameworks. The frameworks are discriminated with different colors (CNN: Blue, DE: Red, CNN+DE: Purple+Red, GD: Orange). The inset shows an example of a simulated 2D power profile for a given pump configuration and flat input power channel load.
  • Figure 2: Power evolution of the 40 signal channels over fiber distance showing the peak-peak estimation of the three loss criteria: maximum power excursion ($J_0$, grey), maximum power excursion over frequency ($J_1$, red), and deviation from loss-compensation ($J_2$, blue).