Antenna Selection for Direction of Arrival Estimation with Deep Learning

Sparse linear arrays have gained attention in array processing due to their advantages over uniform linear arrays (ULA) in reduced energy consumption and mitigated mutual coupling. A key challenge in sparse array design remains in finding an optimal sparse array, specifically as the aperture size and the number of elements increase. Moreover, in minimum redundancy arrays, computing the difference coarray adds to the computational burden along with subspace separation, making angle estimation impractical in real-time scenarios. This study introduces a novel approach that employs a deep learning network to jointly learn a sparse array and a direction of arrival (DoA) estimation model from synthetic data. The shared benchmark includes the DoA estimation performance of the learned sparse array against existing sparse array configurations, including minimum redundancy arrays, coprime arrays, and nested arrays, and the analysis of the effect of the number of antenna elements on the estimation performance. The quantitative analysis shows that the proposed method with the learned sparse array results in a lower estimated angle RMSE than the alternative sparse arrays.