AI 中文总结
本文提出一种适用于大量层数的复杂多层光学元件的简单设计制备方法，基于2×2模块与梯形架构，可支持多种矩阵函数，应用于相位梯度测量相机、小波核卷积等场景。
AI 中文摘要
本文提出了一种简单且渐进的复杂多层光学元件设计与制备方法，即使在层数扩展至大量的情况下也适用。该方法基于由2×2模块构成的连续层，其中光仅沿正向传输；它利用所提出的梯形架构，该架构还可支持可编程结构的自配置，通过对应矩阵行的输入向量设置所需的线性函数或矩阵，仅需连续的单参数功率最大化操作，无需其他迭代。关键所需组件是一层2×2分束器。该方法可支持对应广泛带状对角矩阵的函数，示例应用包括可同时测量相位梯度的相机，以及输入序列与基于小波的核的直接卷积。
英文摘要
I propose a simple and progressive way of designing and fabricating complex multilayered optics, even as we scale to large numbers of layers. The approach relies on successive layers of two-by-two blocks in which light only flows in the forward direction. It exploits a proposed trapezoidal architecture that can also support self-configuration in programmable structures, with the desired linear function or matrix set up using input vectors that correspond to the matrix rows, without any iteration other than successive single-parameter power maximizations. A key required component is a layer of 2 by 2 beamsplitters. The approach can support functions corresponding to a broad range of banded diagonal matrices. Example applications include a camera that also measures phase gradients and the direct convolution of a line of inputs with a kernel based on wavelets.
