DeepFaultNet: Predicting Fault Severity of Communication Cable With Hybrid-ResCNN

Detecting faults in communication wiring systems is always an intriguing research topic. Most current fault finding approaches are based on reflectometry. However, traditional reflectometry requires prior knowledge, such as the cable propagation wave velocity, to analyze the cable state and cannot determine the severity of cable faults. This paper introduces a fault severity prediction method based on a one-dimensional convolution residual network without cable prior knowledge. By correlating the channel transfer function (CTF) with the fault information, a pilot-based deep learning network that can extract fault features more accurately emerges. The method was applied to a communication cable fault detection experiment on a software-defined radio (SDR) platform to verify its effectiveness. Compared with other traditional detection methods, the proposed method is simpler and has better performance.