A new type of all-optical intelligent spectrometer

Recently, Professor Xu Tingfa's research team from the School of Optoelectronics at Beijing Institute of Technology and Assistant Professor Lin Xing's team from Tsinghua University jointly developed a new type of Opto Intelligence Spectrometer (OIS). The device is based on diffractive neural network technology and achieves precise spectral reconstruction under spatially coherent or spatially incoherent light sources, with significant advantages of low energy consumption and light speed processing. The relevant achievements have been published under the title "Opto intelligence Spectrometer Using Diffractive Neural Networks" in the top international optical journal "Nanophotonics" (China University of Science and Technology, Class 1, Top Journal).

Nanophotonics, published by Walter de Gruyter in Germany, focuses on exploring cutting-edge advances in the interaction between light and matter, as well as their fundamental principles and applications. The first authors of this paper are Wang Ze, a master's student at Beijing Institute of Technology, and Chen Hang, a postdoctoral fellow at Tsinghua University. The corresponding authors are Associate Researcher Li Jianan, Professor Xu Tingfa, and Assistant Professor Lin Xing.

The newly developed all optical intelligent spectrometer (OIS) converts the spectral amplitude of the input light source into the detection intensity on the output plane, and uses multiple detectors to accurately perceive the intensity of different spectral bands. By establishing a mapping relationship between input and output and optimizing the phase distribution of the modulation layer using two mean square error (MSE) loss functions, high contrast output intensity distribution and accurate reconstruction of the input light source spectrum were achieved. The principle is shown in Figure 1.

Figure 1. Architecture of all-optical intelligent spectrometer based on diffractive neural network.

The experimental results show that OIS exhibits excellent spectral reconstruction capability under both spatially coherent and spatially incoherent light sources (see Figure 2). In addition, the application testing of the device on the real-world dataset CAVE shows that it has good generalization ability and practical application potential (see Figure 3).

Figure 2. OIS spectral reconstruction results with a spectral resolution of 10nm. Left image: Randomly generated spectral amplitude distribution and spectral reconstruction results. Right figure: Intensity distribution of the output plane.

Figure 3. Spectral reconstruction results of OIS on the real-world dataset CAVE, with a spectral resolution of 10nm.
This study has overcome the long-standing challenges of traditional spectral reconstruction architectures, such as bulky optical components, complex electronic reconstruction algorithms, and limited flexibility. It can serve as a basic unit for array layout, laying the foundation for full light speed and high-quality spectral imaging.

Source: Beijing Institute of Technology