Reinforcement Learning-Based Automation of Time-Frequency Domain Reflectometry
Su Sik Bang, Gu-Young Kwon
Abstract
Advanced reflectometry, which can be used to estimate the location of defects in electrical cables with high accuracy, has undergone continuous development based on advanced signal processing techniques. However, in actual fields, nonexperts find it difficult to use advanced reflectometry owing to the difficulty of the application process. Therefore, in this article, a novel technique that can automate the signal-design process in reflectometry, enabling anyone to use advanced reflectometry in the actual field, is proposed. In this work, the signal-design process is automated by utilizing reinforcement learning (RL), and the application methodology is explained using time-frequency domain reflectometry. The agent of the proposed RL model is trained in a simulation-based environment. Finally, the trained agent is applied to a real-world cable to demonstrate its performance.