Direction-of-Arrival Estimation Method Based on Neural Network with Temporal Structure for Underwater Acoustic Vector...

Area of Science:

• Underwater acoustics
• Signal processing
• Machine learning for sensor arrays

Background:

• Acoustic vector sensors (AVS) are crucial for underwater detection.
• Traditional direction-of-arrival (DOA) estimation methods using covariance matrices suffer from signal timing loss and poor noise immunity.
• Existing methods struggle with accuracy in low signal-to-noise ratio (SNR) conditions.

Purpose of the Study:

• To propose novel deep learning-based DOA estimation methods for underwater AVS arrays.
• To address the limitations of traditional covariance-based DOA estimation techniques.
• To improve DOA estimation accuracy and robustness in challenging underwater acoustic environments.

Main Methods:

• Development of a DOA estimation method utilizing a long short-term memory network with an attention mechanism (LSTM-ATT).
• Development of a DOA estimation method based on the Transformer architecture.
• Comparative analysis against the traditional Multiple Signal Classification (MUSIC) method.

Main Results:

• Both LSTM-ATT and Transformer methods demonstrate superior performance compared to MUSIC, particularly in low SNR scenarios.
• The Transformer-based method achieves DOA estimation accuracy comparable to the LSTM-ATT method.
• The Transformer method exhibits significantly better computational efficiency than the LSTM-ATT method.

Conclusions:

• Deep learning approaches, specifically LSTM-ATT and Transformer, offer substantial improvements in underwater AVS DOA estimation.
• The Transformer-based method provides a promising solution for fast and effective DOA estimation under low SNR conditions.
• These advanced methods enhance the reliability of underwater detection systems.