By Maher Harb, Comcast; Wenlong Shen, Comcast; Matt Stehman, Comcast; Sanket Walavalkar, Comcast; Dan Rice, Comcast
Comcast’s network is undergoing an evolution towards the virtualized Cable Modem Termination System (vCMTS) and remote physical layer (PHY) architecture as part of the journey to 10G [1]. As of early 2023, millions of homes passed can enjoy the multi-Gbps downstream and a few 100 Mbps upstream speeds made possible by deploying a full-width orthogonal frequency division multiplexing (OFDM)channel at 810-1002 megahertz (MHz) and an orthogonal frequency division multiple access (OFDMA) channel in the mid-split (40-85 MHz) region of the spectrum across the vCMTS platform. The path to 10G involves incrementally deploying additional OFDM/OFDMA channels under the Data Over Cable Service Interface Specification (DOCSIS) 4.0 Full Duplex (FDX) technology [2]. However, reconfiguring the spectrum across markets and localities in support of the 10G roadmap is a daunting task because of the complex spectrum management involving the linear video and single carrier-quadrature amplitude modulation (SC-QAM) channels that need to be accommodated, moved, or phased out by converting linear QAM-based video to IP-based video. Adding to the challenge is the prevalence of local video insertions used by multi-dwelling units (MDUs), managed by the local regions, that may not be properly documented in a central location. In this technical paper, we introduce a methodology for discovering spectrum occupancy based on an object-oriented model of the cable modem’s full band capture (FBC). Alongside the configuration data collected from the vCMTS, the FBC model allows the discovery of a host of artifacts that, if not addressed, would be problematic for turning up energy in new regions of the spectrum. The introduced methodology served as the basis for a machine learning pipeline established for automating the required spectrum enablement pre-checks and for generating detailed reporting on all discovered issues. Furthermore, the deployed pipeline for artifact discovery was repurposed to support the effort for deploying a second OFDM channel on the vCMTS platform within the 618-810 MHz range. For this second phase of spectrum activation, the pipeline was also tasked with generating remote PHY device (RPD) configurations tailored for each node, with the goal of eliminating the vacant spectrum (gaps) between video and downstream SC-QAM and creating OFDM exclusion bands informed both by regional policies and by the discovery of insertions on the node. Lastly, we also present our early explorations of the third phase of spectrum activation, which involves repacking video channels to free up the spectrum for FDX. Traditionally, video programming templates have been constructed locally at a market level and were not directly subject to centralized planning and optimization. These early explorations reveal opportunities to free up the spectrum by ensuring that each video QAM is fully packed with multiplexed programs (TV channels). Developing an artificial intelligence (AI) driven pipeline to automatically calculate the capacity requirements and optimal spectrum allocations given a variety of variables such a customer premise equipment (CPE) and network technology penetrations, traffic load, infrastructure and technology constraints, dynamic efficiencies from PHY and media access control (MAC) layer optimization micro-services is our vision. Fully automating the spectrum management per service group not only optimizes capacity and speed, but removes the inevitable human errors, time to configure and inefficiencies required when operating networks at the scale of 10s of millions of broadband customers.