A high-resolution method for direction of arrival estimation based on an improved self-attention module.

Area of Science:

• Underwater acoustics
• Signal processing
• Machine learning

Background:

• High-resolution direction of arrival (DOA) estimation is crucial in underwater acoustics.
• Existing subspace and sparse representation methods have limitations in low SNR, limited snapshots, and high computational complexity.
• Neural network methods show promise but struggle with big data and conventional structures.

Purpose of the Study:

• To propose a novel neural network method for accurate and robust DOA estimation.
• To address limitations of existing DOA estimation techniques in underwater environments.
• To improve performance under challenging conditions like low SNR and limited data.

Main Methods:

• Developed a neural network incorporating an improved multi-head self-attention module.
• Utilized large-scale convolutional kernels and residual structures within the attention module.
• Introduced enhanced input features to handle non-uniform noise and unequal target intensities.

Main Results:

• The proposed method achieved superior angle resolution compared to sparse representation methods.
• Demonstrated exceptional accuracy and robustness in DOA estimation under low SNR and limited snapshots.
• Validated effectiveness through simulations and experimental results.

Conclusions:

• The improved self-attention neural network offers a superior approach for high-resolution DOA estimation.
• The method effectively overcomes limitations of traditional techniques in challenging underwater acoustic scenarios.
• The proposed approach provides a robust and accurate solution for DOA estimation.