Fast mode measurement of optical fiber using a dimension-reduced radial physical network

Mode measurement (MM) enables the quantitative characterization of modal weights and relative phase information at the few-mode fiber (FMF) output, providing essential insights for optical fiber communication system performance optimization. The current approaches underlie the two-dimensional (2D) image processing with the amount of data B × A pixels per image. As the number of modes increases, conventional approaches necessitate extensive datasets and substantial computational time. In this paper, an untrained radial physical neural network (URPNN) is proposed, utilizing only one column of image pixels, based on the inherent principle of the radial modal profile. The URPNN integrates the neural network (NN) with a radial eigenmode superposition mechanism (RESM), extracting mode information from a single column of radial data through dimensionality reduction. The simulation results show that the average modal coefficient error remains on the order of 10⁻³. Experimental results indicate that the correlation between the reconstructed and original intensity patterns exceeds 98%. This method eliminates the need for hours of training and reduces the data requirements by several orders of magnitude.