Cable Fault Detection in DSL Communication System-based on Machine Learning

Abstract

Digital Subscriber Line (DSL) is a technology that commonly used copper cable as the transmission medium in telephone systems, and it is used widely around the world. In this modern era, the demand for high-speed internet keeps increasing to fulfill customer needs. The very-high-bit-rate Digital Subscriber Line (VDSL) is the latest DSL technology emulated in the copper network, providing internet speed up to 100Mbps. However, the copper network is still vulnerable to electromagnetic interference, which can cause degradation in the performance system to achieve a high-speed data rate. Cable faults are also common problems in copper networks like open, partially open, bridge tap, short and partial short circuits. Currently, there is no online monitoring system that able to detect the cable fault conditions accurately. Hence, this project has the best machine learning algorithm that can provide the best accuracy to identify or classify the cable fault condition compared to the ideal condition. Initially, experimentally, the emulation was conducted before deploying machine learning to evaluate the cable fault classifying accuracy. The prediction of cable condition data was simulated using WEKA Software based on few machine learning algorithms such as decision tree (J48), k-Nearest Neighbour (k-NN), Multilayer Perceptron, Naïve Bayes and Random Forest. This algorithm was tested for a cable length between 100 m, 200 m and 300 m on the line operation parameters (LOP) and loop line test (LLT) parameters. These DSL parameters were identified to indicate the overall performance of the DSL technology. The best algorithm for classifying cable fault conditions by LOP and LLT parameters is selected based on the most accurate percentage. The test results showed that the Random Forest algorithm could give consistency a higher accuracy rate with 99% and above for all cable length distances than the other algorithms.