Wide Angular Range Direction‐of‐Arrival Estimation Based on Diffractive Neural Network Integrated Stacked Intelligent Metasurfaces

ABSTRACT Stacked intelligent metasurfaces (SIMs), integrated within diffractive neural network (DNN), have emerged as a promising platform for electromagnetic (EM) sensing. In particular, SIMs enable compact, near‐speed‐of‐light direction‐of‐arrival (DOA) estimation through cascaded wavefront transformations and parallel physical computation. However, existing DNN‐based DOA approaches rely on a spatial‐channel exclusive mapping, coupling angular resolution to detector array scalability and limiting the extension of DOA estimation to wide angular range. To address this issue, we introduce a distributed combinatorial encoding scheme based on multi‐detector joint energy decision logic, overcoming the scalability limitations inherent in the conventional mapping paradigm. The DNN‐designed phase profiles are implemented on two polarization‐conversion metasurfaces to construct an SIM. After modulation by the SIM, incident waves arriving within angular range from −50° to 50° generate angle‐unique intensity distribution patterns on the detection plane, from which the DOA information can be inferred. Full‐wave simulations and microwave experiments validate the proposed approach, demonstrating reliable DOA estimation, including inferring unlabeled incident angles. These results highlight the potential of the proposed simple passive framework for wide angular range DOA estimation and its broader implications for EM sensing in future wireless communication and internet‐of‐things (IoT) ecosystems.