Machine learning aided inverse design for few-mode fiber weak-coupling optimization.

Area of Science:

• Optical Engineering
• Materials Science
• Computational Physics

Background:

• Few-mode fibers (FMFs) are crucial for advancing mode division multiplexing (MDM) systems by enabling higher data capacities.
• Designing FMFs with specific mode properties, particularly weak-coupling for many modes, is challenging due to complexity and time constraints.

Purpose of the Study:

• To develop an efficient and accurate inverse design method for few-mode fibers (FMFs).
• To utilize machine learning, specifically neural networks, for optimizing FMF structures for weak-coupling and multi-mode operation.

Main Methods:

• Employed a neural network-based machine learning approach for the inverse design of FMFs.
• Focused on optimizing the minimum index difference between adjacent modes to achieve weak-coupling.
• Designed multi-ring step-index FMF structures.

Main Results:

• Successfully designed 4-ring step-index FMFs supporting 4, 6, and 10 modes with weak-coupling.
• Designed a 6-ring step-index FMF capable of supporting 20 modes.
• Demonstrated high accuracy, efficiency, and low complexity in the inverse design process.

Conclusions:

• The machine learning method offers a fast, accurate, and reusable solution for designing FMFs, especially those requiring weak-coupling.
• This approach has broad applicability for various optical fiber designs and significant potential for industrial applications.