RC-DOA: DOA Estimation Based on Residual-ConvNeXt

Direction-of-arrival (DOA) estimation is a crucial task in array signal processing for applications spanning radar, wireless communications, and acoustic sensing. Despite its significance, current methodologies face persistent challenges. Traditional model-based approaches exhibit severe performance degradation under low signal-to-noise ratios (SNRs), limited snapshots, and challenging array conditions, while existing deep learning methods struggle with an inherent dilemma between model complexity and estimation accuracy. To overcome these limitations, we propose Residual-ConvNeXt DOA (RC-DOA) estimation, a novel and parameter-efficient approach designed for high-precision DOA estimation. RC-DOA innovatively leverages the hierarchical feature extraction capabilities of ConvNeXt, critically enhanced by designing cross-stage residual pathways to fuse features and stabilize gradients, enabling a lightweight yet powerful design. This architecture efficiently processes dual-channel In-phase/Quadrature (I/Q) signals without complex conventional pre-processing. Our comprehensive simulations demonstrate that RC-DOA not only significantly outperforms existing state-of-the-art deep learning methods across diverse challenging scenarios, including those with low SNRs, limited snapshots, and constrained array sizes, but also achieves this while maintaining a substantially lower parameter count. This work establishes RC-DOA as a robust, parameter efficient, and highly accurate solution for practical DOA estimation, particularly in challenging operational conditions.