Proactive Network Maintenance (PNM) aims to proactively determine issues in a network so higher quality service can be provided and service impairments can be fixed before subscriber’s experience issues. PNM can leverage updates in Data Over Cable Service Interface Specification (DOCSIS), an international telecommunications standard that enables high-bandwidth data transfer through existing cable television systems. The introduction of many PNM related test metrics has made it possible to pinpoint the root causes of issues in an HFC network. Full band capture (FBC) data allows operators to have visibility into all downstream RF signals anywhere DOCSIS 3.0 or 3.1 modems are deployed. This eliminates the need to bring spectrum analyzers to customer homes and perform inspection. Through PNM, downstream RF signals can be monitored 24x7x365 just using the subscriber’s cable modem, which leads to better performance and impairment resolution. Issues can be identified and located faster, leading to greater cost savings and improved subscriber experience.
A challenge for operators is manually analyzing the FBC data from thousands or millions of modems. Further, the cable operator must be able to determine if RF impairments in FBC data are associated with a single home or multiple homes. When an impairment impacts a single home one can usually assumes ending a technician to the individual home is the correct action. However, when multiple homes see the same impairment, sending a technician to a single home is almost always the wrong answer as the impairment is in the outside plant. In this scenario, rolling a truck to a single home for an outside plant impairment wastes time, money, extends MTR and annoys the subscriber.
This is where the power of machine learning and PNM shine. Machine learning can quickly analyze the data of thousands or millions of modems in just minutes. Then it will lead the end user to determine if there are impairments and if so, where the impairments are located.
This paper will discuss the type of RF impairments observable by PNM. Next it will discuss how machine learning is used to analyze impairments using an unsupervised model. Then it will look at how machine learning is combined with CableLab’s spectral impairment detector (SID) to substantially improve on SID’s impairment classifiers. Finally, the paper will look at how the author is using gamification to use feedback from end users to migrate to a supervised learning model.