Angle-multiplexed holography based on optical diffractive neural networks

We propose an angle-multiplexed holography based on an optical diffractive neural network. The method expands the three-dimensional optical design space through a multi-level cascaded diffractive structure, offering enhanced optical field modulation capability and improved independence between angular channels. Using five layers of trainable phase masks, high-quality dynamic holographic storage and angle-selective reconstruction are achieved across 120 distinct channels (covering -10° to 10° with an interval of 1/6°), achieving an average peak signal-to-noise ratio (PSNR) exceeding 22 dB. Furthermore, a sample is fabricated using two-photon direct laser writing, experimentally validating the feasibility of the theoretical design. This research provides a new approach for achieving highly flexible, high-capacity, and low-crosstalk parallel light-field manipulation.