Data-Driven Joint Optimization of Array Design and Imaging

Sparse arrays provide an effective approach for high-resolution imaging by synthesizing a larger effective aperture with fewer antenna elements. However, the large inter-element spacing often introduces severe sidelobes and grating-lobe artifacts, thereby degrading imaging quality. In conventional methods, sparse array design is usually separated from the imaging process, making it difficult to fully exploit the synergy between two-dimensional (2D) array topology and nonlinear reconstruction. Moreover, existing studies on joint optimization are mostly limited to one-dimensional (1D) arrays. To address these issues, this paper proposes a data-driven joint optimization framework for 2D sparse array design and imaging, in which the array design problem is formulated as a differentiable probabilistic sampling task, and a learnable antenna selection module, a deterministic linear imaging layer, and a residual-enhanced U-Net imaging network are integrated into a unified end-to-end framework. Numerical simulation results demonstrate that the proposed method can effectively suppress sidelobe interference and grating-lobe artifacts under highly sparse sampling conditions. Compared with random arrays, uniform arrays, and genetic algorithm optimized arrays, the proposed method achieves superior imaging performance with the same number of array elements.