Training Machines to Learn From Signal Meter Readings (2020)

By Gary Ventriglia, Jack Birnbaum, Robert Gonsalves, Anastasia Vishnyakova, Michael Kreisel & Larry Wolcott, Comcast Corporation

When informed by vast amounts of network performance information, identifying radio frequency (RF) problems with the Data-Over-Cable Service Interface Specifications (DOCSIS) isn’t that hard. However, determining if the problems are inside or outside the home can be difficult. This is a decades-old problem, with hopes often pinned on the elusive promise of artificial intelligence (AI) or machine learning (ML) to help. The challenge that many data scientists will tell you is that having good training data is critical. The lack of a reliable feedback loop to establish cause-and-effect often results in poorly trained machines.

A significant amount of time and resources has been poured into remote diagnostic tools to identify plant problems. Those tools historically have been segmented, specialized, and tuned to evaluate singular aspects of the RF health – for example, receive modulation error ratio (RxMER) and forward error correction (FEC). Once a problem is identified, determining if its source is in a customer’s home, drop or tap has historically been left to technicians, to provide feedback about what they found. The feedback mechanisms typically involve selecting a code or result and updating the work order when it’s complete.

With the COVID-19 pandemic starting mid-March 2020, the rapid development of an “outside network check” provided an opportunity to gather better features and labels. With a renewed desire to keep technicians and customers isolated, the team is exploring new ML/AI models. These new models are trained to use cloud-based RF measurements. These measurements include remote telemetry from DOCSIS devices, and other equipment logged by collection systems. Another set of measurements is taken at the tap and ground block, finally offering a way to segment the network and train the machines differently. The authors review the outcome of this fascinating exercise currently under way, as this paper is being written, in the summer of 2020.

By clicking the "Download Paper" button, you are agreeing to our terms and conditions.

Similar Papers

Right Technician at the Right Time: Using Machine Learning to Predict Network Maintenance Issues
By Anastasia Vishnyakova, Rama Mahajanam, Mike O’Dell, May Merkle-Tan, Catherine Hay & Lisa Pham, Comcast Cable
2021
A PNM System Using Artificial Intelligence, HFC Network Impairment, Atmospheric and Weather Data to Predict HFC Network Degradation and Avert Customer Impact
By Larry Wolcott, Michael O'Dell, Peter Kuykendall, Vishnu Gopal, Jason Woodrich & Nick Pinckernell, Comcast
2018
Execute The Upstream Makeover Without Leaving Scars
By Dr. Robert Howald & Larry Wolcott, Comcast; Leslie Ellis, EllisEdits
2021
Signal Level Meter Calibration Techniques
By Fred J. Schulz, Sterling Communications Inc.
1970
Software Revolution of Field Meters Using a Field-Capable Measurement Device
By Anthony Curran & Andy Martushev, Comcast
2020
Exploring Programmatically Generated HFC Plant Topology
By Lin Cheng, CableLabs; L. Alberto Campos, CableLabs; Justin Riggert, Comcast Corporation; Larry Wolcott, Comcast Corporation
2023
Proactive Network Maintenance (PNM) Paves the Way for More Upstream Bandwidth
By Takashi Hayakawa, Mike O’Dell, Paul Schauer, Larry Wolcott; Comcast
2022
PNM Upstream Data Analysis
By Larry Wolcott, Comcast Corporation; Rob Gonsalves, Comcast Corporation; Jonathan Leech, Comcast Corporation
2023
Improving Operational Intelligence for Maintaining Cable Networks
By Mike Spaulding, Comcast Corporation; Larry Wolcott, Comcast Corporation; Jason Rupe, CableLabs
2022
Detection and Classification of OFDMA Spectrum Impairments by Machine Learning
By Jude Ferreira, Comcast; Kevin Dugan, Comcast; Maher Harb, Comcast; Mike O’Dell, Comcast; Larry Wolcott, Comcast
2023
More Results >>