End-to-end DOA estimation via self-supervised cascaded DNNs with array errors mitigation

Area of Science:

• Signal Processing
• Machine Learning
• Array Signal Processing

Background:

• Direction-of-arrival (DOA) estimation is crucial for applications like radar and sonar.
• Existing methods struggle with multiple array errors and nonlinear distortions.
• Accurate DOA estimation is challenging under low signal-to-noise ratio (SNR) and limited snapshot conditions.

Purpose of the Study:

• To propose a novel cascaded deep neural network for high-precision DOA estimation.
• To address complex nonlinear distortions and multiple array errors without explicit error modeling.
• To enhance robustness and generalization capabilities in DOA estimation.

Main Methods:

• A two-stage cascaded deep neural network architecture is developed.
• The first stage uses four-layer adaptive spatial filters trained with spatial-domain supervision.
• The second stage employs an eight-layer MUSIC-NET for mapping covariance matrices to a MUSIC-like spectrum via self-supervised learning.

Main Results:

• The proposed architecture effectively mitigates array errors through an end-to-end feedforward strategy.
• Simulation results show superior robustness and generalization compared to state-of-the-art methods.
• The system achieves high precision in DOA estimation even with low SNR and limited snapshots.

Conclusions:

• The novel cascaded deep neural network offers a robust solution for DOA estimation in the presence of array errors.
• The end-to-end approach simplifies the process by eliminating the need for explicit error modeling.
• This framework demonstrates significant potential for improving performance in challenging signal environments.