Detection of Modal Numbers From Field Configurations in Rectangular Waveguides via Machine Learning Models of Noisy Datasets

Abstract

We propose a machine learning (ML) modeling methodology to predict the propagation mode number of electromagnetic (EM) fields inside a metallic rectangular waveguide based on the field configuration in the waveguide cross-section, in the presence of noise. We consider the Transverse Electric ( $TE_{mn}$ ) modes and assume m and n in the range of 0 to 2 inside the waveguides, where the magnitude and phase of the noiseless field configurations are obtained from the analytical solution of the electric vector field $\vec {E}$ . We generate training/testing datasets that includes 64,000 plots of the magnitude and phase of $\vec {E}$ over the waveguide cross-section, spanning various TE modes in the frequency range of 13–17 GHz. Our methodology for training and evaluation is based on the classification model, and relies primarily on Stochastic Gradient Descent (SGD) and k-Nearest Neighbors. For real-world scenarios which include noise, we introduce two random distributions in the datasets; specifically, the exponential and the Gaussian distributions are added onto the computed E-fields to further challenge the ML model. We discuss the limitations of the proposed ML modeling approach and the challenges in finding the optimal ML model for these types of problems. The proposed methodology may be generalized to predict both the TE and Transverse Magnetic ( $TM_{mn}$ ) mode numbers with a wide ranges of m and n, as well as for other types of waveguides; e.g., circular, elliptical, etc.

Keywords

• Classification
• electromagnetic (EM) field
• exponential distribution
• Gaussian distribution
• k-Nearest neighbors
• machine learning