Fiber transmission model with parameterized inputs based on generative pre-trained physics-informed neural networks.

Area of Science:

• Optical Communications
• Computational Physics
• Signal Processing

Background:

• Accurate modeling of fiber optic transmission is crucial for high-speed data communication.
• Existing models often require extensive retraining for varying transmission parameters.
• Developing computationally efficient and physically grounded models remains a challenge.

Purpose of the Study:

• To introduce a principle-driven fiber transmission model for short-distance communication.
• To enable universal solutions for diverse bit rates without model re-training.
• To enhance computational efficiency and physical interpretability in optical modeling.

Main Methods:

• Utilized a previously proposed principle-driven fiber model as the core solver.
• Applied the reduced basis expansion method for efficient computation.
• Transformed parameterized inputs into coefficients for the Nonlinear Schrödinger Equations.

Main Results:

• Achieved universal solutions for various bit rates by parameterizing model coefficients.
• Demonstrated prominent advantages in computational efficiency and physical background.
• Successfully trained the model without requiring pre-collected transmitted signals.
• Validated model fidelity using on-off keying and pulse amplitude modulated signals over 100 km.

Conclusions:

• The developed model provides an efficient and versatile approach for short-distance fiber transmission.
• It offers significant computational savings and retains strong physical grounding.
• The model's ability to train without prior signal data enhances its practical applicability.