Enhancing Direction-of-Arrival Estimation with Multi-Task Learning.

Area of Science:

• Signal Processing
• Machine Learning
• Array Signal Processing

Background:

• Traditional Direction-of-Arrival (DOA) and Number of Sources (NOS) estimation methods often operate independently.
• Existing joint estimation techniques may not fully exploit the synergistic information between NOS and DOA estimation tasks.

Purpose of the Study:

• To introduce a novel multi-task Convolutional Neural Network (CNN) for the simultaneous estimation of NOS and DOA.
• To investigate the performance benefits of jointly learning NOS and DOA estimation using a unified deep learning framework.

Main Methods:

• Development of a multi-task CNN architecture designed to process signal data.
• Training and evaluation of the CNN model using simulated datasets with varying noise levels and environmental dynamics.

Main Results:

• The proposed multi-task CNN model demonstrated superior performance compared to existing state-of-the-art methods.
• Significant performance gains were observed particularly in challenging scenarios with high noise and dynamic conditions.

Conclusions:

• Jointly estimating NOS and DOA with a multi-task CNN offers a significant advantage over independent estimation methods.
• The developed CNN provides a robust and effective solution for DOA and NOS estimation in complex signal environments.