Deep Learning-Enhanced MUSIC: An FCDNN-MUSIC Fusion Approach for DOA Estimation in the Presence of Array Phase Errors

The estimation of direction of Arrival (DOA) plays an important role in radar detection, wireless communication, and integrated sensing and communication. However, conventional methods often suffer from performance degradation in DOA estimation due to array phase errors. This paper proposes a fusion solution, which combines a fully connected deep neural network with the Multiple Signal Classification algorithm (shortened as FCDNN-MUSIC) for accurate DOA estimation in the presence of array phase errors. The proposed solution consists of two key components. The FCDNN establishes a nonlinear mapping relationship between the array received signal and the array phase errors to achieve precise estimation and compensation of array phase errors. Subsequently, the classical MUSIC algorithm is used to enable high-precision DOA estimation in multi-source scenarios. Simulation results in various scenarios demonstrate that the FCDNN-MUSIC shows superior generalization capability and strong robustness against array phase errors and different numbers of sources compared to conventional methods. In addition, FCDNN-MUSIC offers a 69% reduction in runtime compared to the conventional WF method.