Multitask Learning-Based Broadband Multiharmonic Signal 2D-DOA Estimation Using Sparse Arrays

Two-dimensional direction-of-arrival (2D-DOA) estimation plays a key role in array signal processing by providing accurate azimuth and elevation information. This full spatial awareness is critical for applications including 3D acoustic imaging, target localization, and autonomous sensing. With the growing demand for device miniaturization, in resource-constrained miniature devices, physical space limitations restrict the array aperture and the number of elements, thereby reducing angular resolution. Enlarging the inter-element spacing to improve resolution, however, may cause phase ambiguity in sparse arrays and consequently reduce estimation robustness. To address these challenges, this paper proposes a 2D-DOA estimation method for small-scale sparse arrays through co-optimization of waveform design, array configuration, and signal processing algorithms. First, inspired by the biosonar mechanism ofHipposideros pratti, a wideband transmit waveform with a unique harmonic structure is designed, which demonstrates superior robustness in reverberant environments compared to conventional signals. Second, a sparse triangular array structure with inter-element spacing exceeding the Rayleigh limit is constructed, significantly improving spatial resolution while meeting miniaturization requirements. Finally, an end-to-end network architecture based on multi-task learning is developed, where collaborative optimization of azimuth and elevation estimation branches effectively enhances estimation accuracy and efficiency. Based on these innovations, a triple-element sparse array DOA estimation system is implemented. Experimental results using measured data demonstrate that the proposed method achieves better estimation accuracy and robustness than existing wideband direction-finding approaches under the same array configuration.