DOA Estimation Method Based on Improved Deep Convolutional Neural Network.

Area of Science:

• Signal Processing
• Array Signal Processing
• Machine Learning Applications

Background:

• Multi-target Direction-of-Arrival (DOA) estimation is crucial for applications like radar and sonar.
• Traditional methods struggle with low signal-to-noise ratio (SNR) and limited snapshots.
• Deep learning offers potential for improved DOA estimation performance.

Purpose of the Study:

• To propose a novel DOA estimation method using deep convolutional neural networks (CNNs) for uniform linear arrays.
• To transform the DOA estimation problem into an inverse mapping problem of the covariance matrix to a binary sequence.
• To evaluate the performance of the proposed CNN-based method against traditional algorithms.

Main Methods:

• Utilizing a deep convolutional neural network (CNN) architecture.
• Mapping the array output covariance matrix to a binary sequence indicating target presence.
• Employing the upper triangular part of the discrete covariance matrix as input data.
• Comparing performance against MUSIC, ESPRIT, ML, and deep fully connected neural networks.

Main Results:

• The proposed CNN algorithm significantly outperforms typical super-resolution algorithms in low SNR and small snapshot scenarios.
• Performance is comparable to MUSIC, ESPRIT, and ML algorithms under high SNR and large snapshot conditions.
• The CNN method demonstrates better accuracy than deep fully connected neural networks.
• Using the upper triangular covariance matrix reduces computational complexity.

Conclusions:

• The proposed deep convolutional neural network method is effective for multi-target DOA estimation.
• The approach offers improved accuracy, particularly in challenging low SNR environments.
• The method provides a competitive alternative to existing DOA estimation techniques with potential for complexity reduction.