Hybrid-driven 2D DOA Estimation Based on Uniform Circular Arrays Algorithm

This paper proposes a hybrid driven MUSIC two-dimensional direction finding algorithm (DM-MUSIC) applied to a uniform circular array (UCA) to address the problem of traditional multi signal classification direction finding algorithms overly relying on basic model assumptions, signal distortion and reduced algorithm accuracy due to multipath effects and signal attenuation in low signal-to-noise ratio, low snapshot count, and other situations. The MUSIC model structure is enhanced using neural networks to replace the sensitive parts of the traditional algorithm such as subspace construction and peak search. The method proposed in this article first performs pattern space transformation and dimension transformation on the received signal through data preprocessing, and then uses two parallel DOA networks to predict the azimuth and elevation angles respectively. Comprehensive experiments have shown that the proposed DM-MUSIC network with a low dimensional network structure has lower complexity and better direction finding performance compared to existing direction finding models, especially under extreme signal-to-noise ratio conditions where the direction finding performance is significantly improved.