Classification and Prediction of Communication Cables Length Based on S-Parameters Using a Machine-Learning Method

Abstract

The length of the communication cables is a significant indicator of signal integrity. The associated scattering parameters characteristics of a communication cable effectively enable its length estimation. This paper proposes a novel machine learning-based algorithm that utilizes Support Vector Machines (SVM) to predict and classify cable lengths. Specifically, the algorithm employs an SVM Regression Model (SVR) and an SVM Classification Model (SVC) to predict and classify cable lengths based on their S-parameters ( $S_{21}$ measurements). As the data under investigation are inseparable, linear, and high-dimensional, SVM has been implemented. The current approach was implemented to verify the length of two datasets, underground and overground cables, with different environmental conditions. The present research introduces an innovative machine-learning algorithm that employs an S-parameter-centric methodology to predict variations in communication cable lengths. Specifically, the SVR model achieved $R^{2}$ values of approximately 0.987 for underground cables and 0.991 for overground cables. Meanwhile, the SVC model demonstrated varying levels of accuracy, with optimal performance seen in five classes for underground cables and four classes for overground cables. The SVM model efficiently extracts and weighs features for high-accuracy predictions in nonlinear, multiclass scenarios, making it the optimal model for this work.

Keywords

• Classification
• prediction
• cable length
• S-parameters
• machine-learning method