# DACL-Net: A Dual-Branch Attention-Based CNN-LSTM Network for DOA Estimation

## TL;DR

This paper introduces DACL-Net, a new deep learning model for estimating direction-of-arrival (DOA) that improves accuracy by using attention mechanisms on transformed covariance matrix data.

## Contribution

The novel DACL-Net model uses a dual-branch architecture with a 2D-FT and attention mechanisms to enhance DOA estimation accuracy.

## Key findings

- DACL-Net achieves an RMSE of 0.04° at 0 dB SNR, outperforming existing DOA estimation algorithms.
- Transforming the covariance matrix into a dark image with bright spots improves CNN-based attention focusing.
- The model's spatio-temporal fusion approach enhances both spatial and temporal feature extraction.

## Abstract

While deep learning methods are increasingly applied in the field of DOA estimation, existing approaches generally feed the real and imaginary parts of the covariance matrix directly into neural networks without optimizing the input features, which prevents classical attention mechanisms from improving accuracy. This paper proposes a spatio-temporal fusion model named DACL-Net for DOA estimation. The spatial branch applies a two-dimensional Fourier transform (2D-FT) to the covariance matrix, causing angles to appear as peaks in the magnitude spectrum. This operation transforms the original covariance matrix into a dark image with bright spots, enabling the convolutional neural network (CNN) to focus on the bright-spot components via an attention module. Additionally, a spectrum attention mechanism (SAM) is introduced to enhance the extraction of temporal features in the time branch. The model learns simultaneously from two data branches and finally outputs DOA results through a linear layer. Simulation results demonstrate that DACL-Net outperforms existing algorithms in terms of accuracy, achieving an RMSE of 0.04° at an SNR of 0 dB.