Research on Underdetermined DOA Estimation Method with Unknown Number of Sources Based on Improved CNN

Area of Science:

• Signal Processing
• Machine Learning
• Array Signal Processing

Background:

• Accurate estimation of source number and direction of arrival (DOA) is crucial in array signal processing.
• Traditional methods often struggle with unknown source numbers and underdetermined scenarios.

Purpose of the Study:

• To propose a joint estimation method for source number and DOA using an improved convolutional neural network (CNN).
• To address limitations of existing algorithms in unknown source number and underdetermined DOA estimation.

Main Methods:

• A CNN model is designed, leveraging the mapping between the signal covariance matrix and source number/DOA estimation.
• The CNN discards pooling layers to prevent data loss and uses dropout for improved generalization.
• The model takes the signal covariance matrix as input and outputs both source number and DOA estimates.

Main Results:

• The proposed algorithm effectively achieves joint estimation of source number and DOA.
• It demonstrates high accuracy under high signal-to-noise ratio (SNR) and large snapshot conditions.
• Outperforms traditional algorithms in low SNR, small snapshot, and underdetermined conditions.

Conclusions:

• The improved CNN-based method offers robust performance for joint source number and DOA estimation.
• It provides a viable solution for underdetermined DOA estimation where traditional methods fail.