Direction-of-Arrival Estimation for a Random Sparse Linear Array Based on a Graph Neural Network

Area of Science:

• Signal Processing
• Machine Learning
• Array Signal Processing

Background:

• Traditional direction-of-arrival (DOA) estimation algorithms struggle with non-uniform and sparse linear arrays.
• Existing deep learning models often fail to effectively capture spatial information in complex array configurations.

Purpose of the Study:

• To propose a novel graph neural network (GNN) based algorithm for DOA estimation.
• To address the limitations of conventional methods in handling random sparse linear arrays.
• To achieve robust DOA estimation under challenging environmental conditions.

Main Methods:

• Development of a GNN model that utilizes neighbor node aggregation and update operations.
• End-to-end training of the GNN for reduced network complexity.
• Comparative experiments on uniform and sparse linear arrays across varying signal-to-noise ratios (SNR) and snapshot counts.

Main Results:

• The GNN model demonstrates superior angle estimation performance on highly sparse arrays.
• Outperforms traditional algorithms and existing deep learning models (CNN, FC) in accuracy.
• Achieves excellent DOA estimation with limited snapshots, low SNR, and large array sparsity.

Conclusions:

• The proposed GNN algorithm offers a significant advancement in DOA estimation for sparse arrays.
• It provides a low-latency, computationally efficient solution suitable for real-time applications.
• The method is effective even under complex conditions where traditional algorithms fail.