Optical Fiber ML Equalization - Baseline Version

Deep Learning for Optical Fiber Channel Equalization using CNN and Transformer architectures.

Features

• Interactive Streamlit Demo with real-time visualizations
• Multiple Modulation Schemes: Binary NRZ, 4-PAM, QPSK, 16-QAM, 64-QAM
• CNN Baseline Model: Convolutional neural network equalizer
• Transformer Model: Attention-based equalizer with superior performance
• Live Performance Metrics: SNR, EVM, BER comparison
• 3-Way Comparison: Without ML vs CNN vs Transformer

Performance Results

• SNR Improvement: Up to 9.1 dB gain with Transformer
• EVM Reduction: 87% improvement over unequalized signals
• BER: 3-4 orders of magnitude improvement

Quick Start

# Install dependencies
pip install -r requirements.txt

# Launch demo
LAUNCH_DEMO.bat

Project Structure

├── demo_app.py              # Main Streamlit application
├── models.py                # CNN and Transformer architectures
├── channel.py               # Optical fiber channel simulation
├── modulation.py            # Modulation schemes
├── visualizations.py        # Eye diagram, constellation plots
├── data/                    # Training datasets
├── trained_models/          # Pre-trained model weights
└── LAUNCH_DEMO.bat         # Demo launcher

Customizable Parameters

• Fiber distance (0-250 km)
• SNR (5-30 dB)
• Dispersion coefficient (5-30 ps²/km)
• Noise figure (2-12 dB)
• Fiber loss (0.1-0.5 dB/km)
• Nonlinearity factor (0-5)

Models

CNN Baseline

• Convolutional layers for feature extraction
• ~150K parameters
• Moderate performance improvement

Transformer (Novel)

• Multi-head self-attention mechanism
• ~100K parameters (fewer than CNN!)
• Superior long-range ISI compensation
• Best performance across all metrics

Version

Baseline v1.0 - Initial working version with 3-way comparison and performance metrics

Future Improvements

• Additional architectures (LSTM, Hybrid models)
• Enhanced training techniques
• Real-time model inference
• Extended modulation support

License

MIT License