High-speed Internet access plays an important role in modern life. It helps narrow the digital divide, enables e-commerce, and provides opportunities for remote work, study, and entertainment. In the US, cable Internet Service Providers (ISPs) are one of the few available infrastructures that can provide high-speed Internet access to US homes, and in many rural areas, they are often the only broadband choice. However, much measurement study has shown that broadband access networks have poor reliability. (Many parts of the cable networks are now decades old.) This work aims to address this problem. If successful, it can significantly improve the reliability of cable broadband networks, thereby improving Internet availability and quality of experience of millions of cable broadband users.
It is challenging to improve the reliability of cable networks because the "last-mile" links that connect a subscriber's home to the Internet are often made of coaxial cables. They are vulnerable to radio frequency interference and suffer from aging-related issues. These problems can manifest themselves as unrecoverable noisy signals, disrupting a subscriber's Internet connectivity. The cable industry currently collects performance data from users' cable modems. However, there is a lack of systematic study on how to use this data to detect and localize network problems and to rank the severity of problems. Existing techniques often lead to an unacceptably high number of false positives in practice. This work will develop algorithms and tools that can help detect and localize faults in cable networks and estimate the performance impact of network faults. Specifically, the work will include 1) algorithms that use collected data to accurately detect faults inside a cable network; 2) algorithms that localize a fault to a geographical location by clustering anomalous data patterns; and 3) algorithms that estimate how many calls an ISP may receive given a network problem as a mechanism to prioritize repairs. In this project, these algorithms will be implemented in software. All algorithms and code resulting from this work will be made publicly available. The researchers anticipate that the algorithms developed in this project are also applicable to WiFi and cellular networks.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.