Meta‐Training of Diffractive Meta‐Neural Networks for Super‐Resolution Direction of Arrival Estimation

ABSTRACT Diffractive neural networks leverage the high‐dimensional characteristics of electromagnetic (EM) fields for high‐throughput computing. However, existing architectures face challenges in integrating large‐scale multidimensional metasurfaces into precise network training and haven't utilized multidimensional EM field coding schemes for super‐resolution sensing. Here, diffractive meta‐neural networks (DMNNs) are proposed for accurate EM field modulation through metasurfaces, which enable multidimensional multiplexing and coding for multi‐task learning and high‐throughput super‐resolution direction of arrival estimation. DMNN integrates pre‐trained mini‐metanets to characterize meta‐atom optical modulations across polarizations and frequencies, with structure parameters designed using meta‐training. DMNN simultaneously resolves azimuthal and elevational angles through x and y‐polarizations, while the interleaving of frequency‐multiplexed angular intervals generate spectral‐encoded super‐oscillatory angular responses to achieve wide‐field super‐resolution estimation. Post‐processing lightweight fully‐connected neural networks further enhance the performance. Experimental results validate that a three‐layer DMNN operating at 27, 29, and 31 GHz achieves 0.5 angular resolution, i.e., the Rayleigh diffraction limit, a mean absolute error of 0.048 for incoherent targets within , and an angular estimation capacity of 1917, an order of magnitude higher than that of existing methods. Our work advances super‐resolution photonic computing systems by utilizing inherent high‐parallelism and all‐optical coding methods for next‐generation sensing applications.