Automatic Modulation Classification for Underwater Acoustic Communication Signals Based on Deep Complex Networks.

Area of Science:

• Signal Processing
• Machine Learning
• Underwater Acoustics

Background:

• Automatic modulation classification (AMC) is crucial for underwater communication but is hindered by multi-path fading and ocean ambient noise (OAN).
• Traditional methods struggle with the complex environmental influences in underwater acoustic channels.

Purpose of the Study:

• To enhance automatic modulation classification (AMC) for underwater acoustic communication signals.
• To address the challenges posed by multi-path fading and OAN in underwater environments.

Main Methods:

• Proposed two novel complex physical signal processing layers based on Deep Complex Networks (DCN): a deep complex matched filter (DCMF) and a deep complex channel equalizer (DCCE).
• Constructed a hierarchical DCN integrating these layers to process complex underwater acoustic signals.
• Validated the method using real-world ocean observation datasets, white Gaussian noise, and real-world OAN.

Main Results:

• The DCN-based AMC achieved 5.3% higher average accuracy compared to traditional real-valued deep neural networks.
• The proposed method effectively reduced the impact of underwater acoustic channels, including multi-path fading and noise.
• Demonstrated superior performance over existing advanced AMC methods in real-world underwater acoustic channel conditions.

Conclusions:

• Deep Complex Networks (DCN) offer a robust solution for automatic modulation classification (AMC) in challenging underwater acoustic environments.
• The integration of complex signal processing layers within DCN enhances resilience to environmental interference.
• The developed method shows significant potential for reliable underwater communication monitoring and interference identification.