Transfer learning assisted convolutional neural networks for modulation format recognition in few-mode fibers

Area of Science:

• Optical Communications
• Signal Processing
• Machine Learning

Background:

• Few-mode fiber (FMF) is crucial for high-capacity next-generation elastic optical networks (EONs).
• Probabilistic shaping (PS) is widely adopted in EONs to optimize transmission near the Shannon limit.
• Modulation format recognition (MFR) is essential for managing complex signal formats in these networks.

Purpose of the Study:

• To develop a fast and accurate method for modulation format recognition (MFR) in PS-based FMF-EONs.
• To leverage transfer learning networks (TLNs) for analyzing signal constellations in FMF transmissions.
• To evaluate the performance of TLNs against conventional deep learning networks (DLNs) for MFR.

Main Methods:

• Investigated a transfer learning network (TLN) for MFR in PS-based FMF-EONs.
• Employed convolutional neural networks (CNNs) within the TLN for feature extraction from signal constellations.
• Conducted experiments involving six modulation formats (16QAM to PS-64QAM) and four propagating modes (LP01 to LP21).

Main Results:

• The proposed TLN achieved fast and accurate MFR in PS-based FMF transmissions.
• TLN demonstrated reduced computational requirements, with one-tenth the iterations of conventional DLNs.
• TLN effectively mitigated overfitting and required less training data compared to DLNs.

Conclusions:

• The developed TLN is an efficient and feasible solution for MFR in PS-based FMF communication systems.
• TLN offers a promising approach for enhancing signal management in advanced optical networks.
• The study validates the effectiveness of TLNs for complex signal recognition tasks in optical communications.