There are more than 300,000 HFC nodes in the US currently, and several million more worldwide. The additional network traffic triggered by Covid-19 in the spring of 2020 increased the level of HFC capacity augments by as much as 300% compared to 2019 volume. Network augmentation techniques such as node splits, adding HSI EIA’s (6 MHz channels), service group de-combines, bandwidth expansion, Node+0/RPD’s and mid-/high split reverse path expansion require engineering and operations resources in both ISP and OSP.
This paper will examine techniques in which much of this activity can be automated and iteratively optimized through Machine Learning (ML). With inputs provided from network mapping systems, capacity monitoring platforms, spectrum management applications, and business rules such as preferred augmentation hierarchy, expected duration before next augment, balancing of house-counts, municipal permitting difficulty, and cost efficiency, the ML environment would rapidly analyze entire geographic segments of the network, and provide augmentation planning data including network design changes and BOM’s for areas requiring immediate physical layer upgrades.
As the ML platform iteratively processes network geographies, it will learn from how past predictions tracked to current status and continuously adjust to optimize capacity augmentation methods and designs.
As the ML environment will be analyzing the entire network geography, data to drive Capex planning for future years will be derived enabling the operator to more efficiently allocate capital.
The ML environment would also support “what/if” network topology planning for approaches, such as bandwidth expansion to 1.8 GHz vs Node+0 at current bandwidth, and provide data on cost, duration before next augment needed, and percentage of network elements requiring replacement or repositioning.