Predicting the Diffraction Behavior of LCoS-SLMs With High Accuracy Using Temporal Convolutional Network

As a critical constraint in liquid crystal on silicon spatial light modulators (LCoS-SLMs), pixel crosstalk degrades their performance in various applications. Accurately predicting the diffraction behavior of LCoS-SLMs is a prerequisite for effectively compensating for pixel crosstalk. However, current approaches, whether physical models or phenomenological approximations, cannot provide sufficient prediction accuracy and broad applicability for practical deployment. To address these challenges, this work proposes a deep learning-based approach that combines a modified Temporal Convolutional Network with a comprehensive physical diffraction model. Experimental results demonstrate that the prediction accuracy of our approach is 3.13, 9.40, and 3.89 times that of a widely used convolution-based model for the +1st order, 0th order, and other crosstalk orders (> -30 dB), respectively, while operating with a millisecond-scale response time. With its high accuracy, real-time capability, and broad applicability, our approach paves the way for applications such as optical communications, laser processing, LiDAR, and holography.