Machine Learning-Based Photonic Integrated Spectrometer

Author's Department

Physics Department

Second Author's Department

Physics Department

Third Author's Department

Physics Department

Fourth Author's Department

Physics Department

Find in your Library

https://doi.org/10.1117/12.3043612

All Authors

Passant Hesham Abdelrahman Nasser Thomas J. Mikhail Mohamed A. Swillam

Document Type

Research Article

Publication Title

Proceedings of SPIE the International Society for Optical Engineering

Publication Date

1-1-2025

doi

10.1117/12.3043612

Abstract

This paper proposes a spectrometer design based on the speckle pattern reconstruction of a multi-mode interferometer using the K- Nearest Neighbor based-regression Machine Learning Algorithm. A disorder medium has been used to obtain the speckle pattern controlled by varying incident light wavelength. Eigenmode expansion solver is used to design and simulate the multi-mode interferometer. The dependency of the wavelength speckle pattern has been tested over a range from 500nm to 1000nm. Nearest Neighbor based-regression algorithm is a supervised machine learning model used to analyze the variation in the wavelength using speckle pattern. The high accuracy of the prediction of the speckle pattern is based on two primary methodologies. Firstly, a pre-processing technique for feature extraction. A nonlinear method t- SNE is used for the feature extraction by doing a dimensionality reduction to a 1D vector. Secondly, utilize this 1D vector as an input for the KNN model.

Link to Full Text
Find this item online

Share

COinS