A Machine Learning Pipeline for D3.1 Profile Management (2019)

By Maher Harb, Jude Ferreira, Dan Rice, Bryan Santangelo & Rick Spanbauer, Comcast

We have entered an era where leveraging Machine Learning to optimize the performance of cable access networks is possible and, perhaps, even a must. The fast arising opportunities in this realm are due to advances in cable technology and increasing investment in data science functions across organizations.

Specifically, DOCSIS 3.1 (D3.1) includes Orthogonal Frequency Division Multiplexing (OFDM), enabling the possibility of tailoring the modulation of OFDM channels to realize much improved spectral efficiency and impairment resiliency. Additionally, due to the nature of the wider OFDM channels, Comcast identified several opportunities and key deployment challenges affecting network stability & performance. As part of operationally hardening D3.1, it became clear that an effective modulation Profile Management Application (PMA) is essential for operating D3.1 to its full potential. The initial perspective -- that PMA was an optimization technique to maximize capacity in the future -- changed to a conclusion that PMA is really a table-stakes feature required to ensure network stability, manage operational metrics, and ensure a great customer experience. This document describes the profile management solution developed to address these challenges. As a primer to the ensuing discussion, consider the distribution (shown in Figure 1) of Modulation Error Ratio (MER) collected from Comcast’s entire population of D3.1 devices. The distribution, while encapsulating information aggregated across billions of subcarriers, hints at the core idea of PMA: since the quality of the spectrum varies across the network, customizing modulation across subcarriers and devices holds the opportunity to enhance network performance in terms of increasing both capacity and resiliency. The goal of PMA is to pursue this ideal.

The paper is organized as follows: DOCSIS 3.1 Hardening describes our efforts in addressing D3.1 deployment challenges; these efforts are an important precursor to PMA. Problem Statement introduces formulation of the PMA problem. Overview of Solution presents the high-level PMA solution architecture. MER & Time describes our solution for addressing MER variation in time. Core Algorithm introduces the algorithm developed for constructing D3.1 Profiles. Lab Environment describes establishing a lab for testing the PMA solution. Pattern Detection describes a host of algorithms aimed at detecting impairments; these are complimentary to the PMA effort. Future Work comments on the future evolution of the PMA algorithm.

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

Similar Papers

Convolutional Neural Networks for Proactive Network Management
By Jude Ferreira, Maher Harb, Karthik Subramanya, Bryan Santangelo & Dan Rice, Comcast
Full Scale Deployment of PMA
By Maher Harb, Bryan Santangelo, Dan Rice & Jude Ferreira, Comcast
Practical Implementation of Profile Management Application
By Brady Volpe, NimbleThis and The VolpeFirm
How Network Topology Impacts Rf Performance: A Study Powered By Graph Representation Of The Access Network
By Maher Harb, Karthik Subramanya, Ramya Narayanaswamy, Sanket Walavalkar & Dan Rice, Comcast
Optimizing DOCSIS 3.0 Configuration in the Upstream through Applied Reinforcement Learning
By Kevin Dugan, Maher Harb, Dan Rice & Robert Lund, Comcast
Dynamic Data Collection & Configuration Management
By Rohini Vugumudi, Hany Fame, Pardeep Singh & Zhen Lu, Comcast
Practical Lessons from D3.1 Deployments and a Profile Management Application
By Karthik Sundaresan, Jay Zhu & Mayank Mishra, CableLabs; James Lin, Kyrio/CableLabs
It’s 10 PM: Do You Know Where Your Wavelengths Are?
By Venk Mutalik, Dan Rice, Rick Spanbauer, Simone Capuano, Rob Gonsalves & Bob Gaydos, Comcast
Roaring into the ‘20s with 10G
By Dr. Robert Howald, Robert Thompson, Sebnem Ozer, Daniel Rice & Larry Wolcott, Comcast; Dr. Tom Cloonan, Dr. Ruth Cloonan & John Ulm, CommScope; Jan Ariesen, Technetix
Simplifying Field Operations Using Machine Learning
By Sanjay Dorairaj, Bernard Burg & Nicholas Pinckernell, Comcast Corporation; Chris Bastian, SCTE
More Results >>