Diffractive networks enable optical information transfer through random and unknown diffusers

LOS ANGELES - July 1, 2026 - PRLog -- The transmission of optical information through random scattering media is a major challenge in optics, biomedical imaging, telecommunications, and remote sensing. When light passes through a turbid or diffusive medium, such as biological tissue or a randomly structured optical material, the original image information can be severely distorted, making reliable recovery of information difficult. Researchers at the University of California, Los Angeles (UCLA) have introduced interleaved diffractive networks to address this challenge by enabling optical information transfer through random and unknown diffusers.

This new framework uses a cascaded diffractive optical network composed of passive, spatially structured layers that are physically interleaved within a scattering medium. Instead of placing all optical processing elements before or after the diffuser, the system embeds optimized diffractive layers throughout the scattering volume. These interleaved layers are trained to reshape the propagating optical field and mitigate the effects of scattering, allowing the system to recover object information after propagation through random and unknown diffusers.

UCLA researchers demonstrated that the performance of this all-optical architecture depends on key physical parameters, including the depth of the diffractive processor, and the spatial arrangement of the diffractive layers. Their results showed that interleaving trained diffractive layers within the scattering volume provides a powerful physical strategy for improving information transfer and image recovery compared with conventional arrangements in which optical processors are placed outside the diffusive medium.

To further enhance image reconstruction fidelity and robustness, the research team also developed a hybrid optical–digital system. In this configuration, the passive diffractive processor is coupled with a jointly trained digital neural network. This hybrid design achieved improved information recovery even when the input objects were subjected to unknown random rotations, shifts, and scaling, highlighting the adaptability of the method under realistic and variable imaging conditions.

The UCLA team also experimentally validated the approach using a fabricated multi-layer diffractive system operating in the visible spectrum. These experiments demonstrated reliable recovery of optical information through random diffuser layers, supporting the feasibility of implementing interleaved diffractive processors as physical information recovery systems.

This work points to a new direction for optical computing and imaging through random scattering media. By distributing passive diffractive computation inside a random medium, interleaved diffractive networks can transform scattering from a barrier into part of the optical information-processing architecture. Such systems may provide new opportunities for biomedical imaging, remote sensing, optical communications, and other applications where light must propagate through complex or turbid environments.

The study was supervised by Prof. Aydogan Ozcan of UCLA and funded by the US NSF (National Science Foundation). The other authors of this work include Yuhang Li, Yiyang Wu, Shiqi Chen, and Xilin Yang of UCLA Samueli School of Engineering and California NanoSystems Institute (CNSI).

Article: https://doi.org/10.1002/lpor.71471