Fast self-learning modulation recognition method for smart underwater optical communication systems.

Area of Science:

• Optical Wireless Communications
• Machine Learning
• Signal Processing

Background:

• Automatic modulation recognition (AMR) is crucial for intelligent transceivers in underwater optical wireless communications (UOWC).
• Severe underwater conditions, including turbulence and noise, pose significant challenges for existing AMR techniques.
• Few-shot learning is essential for adapting to new tasks with limited data in dynamic environments.

Purpose of the Study:

• To propose and analyze a novel progressive growth meta-learning (PGML) based AMR scheme for UOWC.
• To enhance the self-learning capability of AMR systems in challenging underwater environments.
• To improve classification accuracy and robustness against noise and distortion.

Main Methods:

• An orthogonal frequency division multiplexing (OFDM) based PGML framework for few-shot AMR.
• Directly feeding constellations of noisy signals in low signal-to-noise ratio (SNR) scenarios to a few-shot classifier.
• Utilizing data augmentation (DA) to mitigate light-emitting diode (LED) distortion.

Main Results:

• The proposed PGML scheme demonstrates superior training efficiency compared to classical meta-learning (ML).
• The PGML approach exhibits enhanced robustness against Poisson noise.
• Significant improvements in generalization performance on new tasks were observed.

Conclusions:

• The PGML-based AMR scheme offers a promising solution for future UOWC systems.
• The framework effectively addresses the challenges of limited data and harsh underwater conditions.
• The method provides a robust and efficient approach to modulation recognition in UOWC.