A Two Sub-problem Decomposition for the Optimal Design of Filterless Optical Networks

Brigitte Jaumard, Yan WangarXiv 2021 · arXiv

Introduction

Filterless optical networks (FONs) represent a paradigm shift in network design, prioritizing cost-efficiency and energy sustainability. By replacing expensive active switching components like Reconfigurable Optical Add-Drop Multiplexers (ROADMs) with passive splitters and couplers, FONs offer a robust and "green" alternative for modern telecommunications. However, this simplicity introduces a significant design challenge known as the broadcast effect.

In a filterless architecture, an optical signal is not terminated at its destination; instead, it propagates throughout the entire fiber tree (or f-subnet). This unfiltered propagation consumes spectrum across multiple links, potentially leading to low spectral efficiency if not managed correctly. Designing these networks requires a complex optimization of fiber tree configurations and wavelength assignments to prevent signal interference.

Physical network (a) vs. logical filterless sub-networks (b) illustrating the broadcast effect.

The DFOP Optimization Framework

Traditional approaches to filterless network design often rely on multi-step heuristics that separate tree selection from wavelength assignment, frequently resulting in suboptimal solutions. This paper introduces the Decomposition for Filterless Optical Provisioning (DFOP) model, the first "one-step" exact optimization framework for this problem.

At its core, DFOP utilizes Column Generation (CG), a large-scale mathematical programming technique. The problem is decomposed into a Master Problem and two specialized Pricing Problems:

  1. Master Problem: Coordinates the selection of the most efficient filterless sub-networks (FSNs) and wavelength configurations to minimize the total wavelength count.
  2. FSN Pricing Problem: Mathematically "discovers" new, optimal fiber tree structures based on network demands.
  3. Coloring Pricing Problem: Solves a variation of the graph coloring problem to ensure conflict-free wavelength assignment across the broadcast domains.
The DFOP optimization workflow utilizing column generation and sub-problem decomposition.

By integrating these components, the model can identify routing trade-offs—such as choosing a slightly longer path—that enable significantly better wavelength reuse across the entire network.

Experimental Evaluation and Results

The DFOP model was benchmarked against five standard optical topologies, including the Italy, California, Germany17, Cost239, and USA networks. Using unit uniform demands, the researchers compared their results against previous state-of-the-art heuristics.

Key findings include:

  • Germany17 Network: A substantial improvement was observed, reducing the required wavelength count from 88 to 73.
  • Italy Network: The model successfully reduced the wavelength requirement from 25 to 23 using only two optimized trees.
  • California Network: The framework matched existing benchmarks (120 wavelengths) while providing a more rigorous mathematical foundation and identifying a theoretical lower bound ($z^*_{LP}$) that suggests further optimization potential.

Conclusion

The DFOP framework demonstrates that integrated optimization can significantly enhance the spectral efficiency of filterless optical networks. While the computational complexity remains high for very large-scale backbones—sometimes requiring several hours for complex topologies—the model provides a crucial benchmark for optimal design. Future work will focus on implementing full "Branch-and-Price" algorithms to guarantee integer optimality and exploring the model's performance under dynamic, non-uniform traffic loads.

Summary updated

BibTeX
@misc{jaumard2021two,
  author = {Brigitte Jaumard and Yan Wang},
  title = {{A Two Sub-problem Decomposition for the Optimal Design of Filterless Optical Networks}},
  year = {2021},
  eprint = {2101.01022},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2101.01022},
}
Figure