Due to the size and complexity of the coaxial cable network, the network is prone to many anomalies that cause network disruption. This means that TDC NET has a large number of technicians driving around trying to locate and fix a given error. It is also difficult to isolate which anomalies are severe enough to cause the network to breakdown as noise is always present in the network.
The coaxial cable technology, hereafter known as the coax-network, is used for both private internet access and for broadcasting broadband TV. It is the most common technology for cabled data transmission used in Denmark. TDC NET is the biggest supplier of digital infrastructure in Denmark with more than 600,000 customers on their coax-network. The network consists of modems and amplifiers connected under a single Coaxial Media Converter or CMC. The tree-like structure can be observed in Figure 1. It is important for TDC that there is as little disruption to their network as possible. Unfortunately, due to the long term deployment of the coax-network infrastructure, there has been a gradual depletion of the condition of the network. This leads to occasional network failure and breakdown. Problems arise such as cable corrosion and loose connectors which can lead to signal distortion in the network. As well as this, coaxial cables are vulnerable to RF interference. These issues result in poor application-layer performance such as slow internet responses or low-quality streaming [1].
Although data on the status of the signal is collected at CMC level and at modem level, it is not collected at amplifier level, meaning that the quality of the signal at these points are unknown. This is problematic for a number of reasons. Firstly, as the signal of the network is distorted at the amplifier, an error is more likely to occur in its vicinity. Secondly, if the location of the network error is unknown, it is difficult for a technician to efficiently repair it.
The solution is to create an efficient anomaly detection algorithm. This would examine the possibility that an anomaly could trigger significant network error, and provide an approximate location of the anomaly in the network. According to Zhang et. al determining exactly which sensors or system components are causing this anomaly can result in a more expedient diagnostic repair [2].
Recent developments in the field of anomaly detection have incorporated deep learning methods in order to successfully isolate unusual behaviours. Deep learning methods which incorporate encoder/decoder models are considered a suitable option for modelling sequence data such as time series as they are a straightforward way of using recurrent neural networks to process this data [3].
Proactive Network Maintenance (PNM) is dynamic detection of abnormalities in observations before network users experience issues which improves the quality of the network overall [1]. Discordant observations may be present in the cable network, however, it is important the detection system created for the severity of the anomaly before assigning technicians for repair work. A statistical analysis of the most commonly occurring network errors and causes of breakdown will be undertaken to determine how errors such as ingress, egress and common path distortion manifest themselves in the data, if at all. It would also be useful to investigate the risk factors which cause these errors, to mitigate for them in future. It is important to undertake these techniques rather than a naive approach to anomaly detection so that the false positive rate of network errors can be significantly reduced [4].
[1] Jiyao Hu, Zhenyu Zhou, Xiaowei Yang and Jonathan W Williams: CableMon: Improving the Reliability of Cable Broadband Networks via Proactive Network Maintenance, Proceedings of the 17th USENIX Symposium on Networked Systems Design and Implementation, 2020
[2] Domjan Bari´c , Petar Fumi´c , Davor Horvati´c, and Tomislav Lipic: Benchmarking Attention Based Interpretability of Deep Learning in Multivariate Time Series Predictions, Human-Centric AI: The Symbiosis of Human and Artificial Intelligence, 2021
[3] Ralph Foorthuis: Algorithmic Frameworks for the Detection of High-Density Anomalies, (Symposium on Computational Intelligence in Data Mining, 2020
[4] Yu Tian, Haihua Liao, Jing Xu, Ya Wang, Shuai Yuan, Naijin Liu, Unsupervised Spectrum Anomaly
Detection Method for Unauthorized Bands Space: Science & Technology, 2022