A Passive-Measurement Method for Physical Security and Cable Diagnosis

Abstract

This article introduces a passive measurement method for diagnosing anomalies in two-wire communication channels using machine learning (ML). The proposed method involves the acquisition of the signal received by a transceiver and the decoded sequence provided by the receiver. In particular, it does not require the acquisition of a particular injected signal and any synchronization of the acquisition with the data transmission, making it suitable for the diagnosis of existing two-wire communication channels without interrupting their operability. An experimental setup has been implemented to generate a dataset of acquired signals through a channel having the following anomalies: air-exposed conductors, water-exposed conductors, and tapping of various lengths. The performance of an ML-based decision tree classifier has been assessed according to features extracted in the time and frequency domains from the acquired signal and an estimated impulse response of the cable obtained from the decoded sequence. The most sensitive features to the anomalies have been analyzed, and the decision tree classifier has been trained according to them by considering several sampling frequencies of the signal acquisition, ranging from 62.5 MHz to 6.25 GHz. The classification accuracy obtained in a set of laboratory experiments carried out on actual anomalies is 99.04% at the sampling frequency of 312.5 MHz. Moreover, an analysis is carried out to assess the sensitivity of the diagnostic tool to the anomaly lengths, thus demonstrating its capability to estimate them.