Transfer Learning Strategy in Neural Network Application for Underwater Visible Light Communication System

Area of Science:

• Optical Communications
• Machine Learning
• Signal Processing

Background:

• Visible Light Communication (VLC) is a key technology for the 6G era, but suffers from nonlinear distortion.
• Neural Network (NN) post-equalization effectively models these distortions without prior physical knowledge.
• Standard NN training faces challenges with generalization across different channel conditions, requiring extensive retraining.

Purpose of the Study:

• To investigate the efficacy of transfer learning for NN post-equalizers in VLC systems.
• To develop a more efficient training strategy that improves model generalization and reduces computational load.

Main Methods:

• Proposed a transfer learning strategy using a pre-trained 'stem model' for NN post-equalizers.
• The stem model captures essential channel features, balancing signal-to-noise ratio and nonlinearity.
• Subsequent training focuses on adapting to specific channel variations, rather than starting from scratch.

Main Results:

• Stem-originated DNN models achieved up to 95% of the working range with 150% higher training efficiency compared to exhaustive training.
• In some cases, models achieved 64% of the working range with five times the training efficiency.
• Demonstrated significant improvements in training efficiency and model generalization.

Conclusions:

• Transfer learning via stem models offers a more efficient and generalizable approach to NN post-equalization in VLC.
• This method significantly reduces the computational cost associated with retraining models for diverse channel conditions.
• The findings enhance the practical feasibility of deep neural network (DNN) applications in real-world VLC systems.