NBN looks to 'proactive' network health with machine learning

Pushing machine-learning technology and big data applications is enabling Australia's National Broadband Network (NBN) to be more "proactive" in analysing and repairing network health, the company has said.

Speaking during the annual CeBIT conference in Sydney on Tuesday morning, NBN executive general manager of IT Strategy and Architecture Arun Kohli said NBN is collecting real-time data from "every part of the network which is connected or can be connected".

"On top of the typical applications of insights, metrics, and predictive analytics, we've gone to the next level of machine learning to be more proactive for the customer experience," Kohli said.

"Also, this helps us optimise our program -- what kind of technology we have to do, what kind of issues can come up in future when we decide the upgrade."

Kohli said network data already collected by NBN includes modulation stats, sync and error rates, frame loss and delay, and alarms and faults. It is next looking to collect data on micro-reflections, spectral response curves, and throughput over time.

NBN has also successfully trained machine learning models to recognise various impairment types from spectral data, he said. As a result, the company is able to discover in-house issues such as bridged taps -- unterminated branches of copper pairs in the premises that cause signal reflections and result in speed loss -- and outside plant issues such as copper pair corrosion caused by water penetration, which results in speed loss and service instability.

"When we look into the in-house copper data of the connectivity data [combined] with the machine learning ... we can predict two things: That services may start degrading over time -- the service is still there and probably the end user doesn't know, but over time it may become more degraded," he explained.

"And the other important thing is using that information, we can now calculate exactly 6.2 metres from the pit to the home there's an issue."

NBN is also collecting data from hybrid fibre-coaxial (HFC) modem resyncs, or "flaps", which he described as being a temporary loss of connection due to physical line impairments or capacity issues.

NBN has also successfully trained machine learning models to recognise HFC modems that are likely to flap, he said, which the company uses to "predict modem flaps at scale" before the end user even notices.

"We see ... a degradation before the service goes out," Kohli said.

"Using the big data ... what you see is some kind of interruption in the service, and it happens probably for a couple of minutes, some people notice ... [but] even if the end user is not able to notice, there's enough data from the network which we can use to predict and find out where the issues are."

By using a machine learning application on top of the collection of real-time data, NBN is therefore now able to go from reactive to proactive management of the network, Kohli explained.

"With so much of our data about the health of the network, and with the environment of the software technologies on the big data, we have just started to move the data from being reactive to being proactive," he said.

"And that's where the application of machine learning comes into play ... if you are proactive, that drives the customer experience."

Kohli's speech followed NBN last year announcing trials of a new diagnostic tool aimed at remotely finding out whether a premises has faults along the copper cabling running to and from telephone sockets inside homes.

"We will begin a trial of a new diagnostic tool that we hope will quickly and accurately detect premises that may be suffering from speed issues related to in-home wiring faults," NBN acting CTO Carolyn Phiddian said in September.

"Of those studied [in the internal tests], speed performance issues identified in one in two premises on fibre-to-the-node networks were caused by in-home wiring. In many of these cases, poor wiring caused download speeds to degrade by more than 50 percent."

The tool would then be made available to retail service providers (RSPs) to help diagnose broadband service issues.

NBN is also working on data analytics and visualisation with the University of Technology Sydney (UTS) and the University of Melbourne under three-year R&D partnerships, along with programmable networks, artificial intelligence (AI), robotics, and wireless technology.

Kohli also used his CeBIT speech on Tuesday to speak about how NBN is enabling other businesses to use big data by providing a high-speed ubiquitous network.

"The big data will not scale if it is only within the datacentre. The application of big data is the velocity at which you can access the data, and to do that, having a distributed network which connects every location, that's where NBN comes into play," he argued.

"NBN's network -- which is available, which is high speed, low latency -- helps the distribution of the data and influences the economics of what's possible for a data-driven business ... with the decentralisation and distribution of data across onsite datacentres and multiple cloud infrastructure."

NBN Coverage

• NBN's FttC network reaches 18K premises
• NBN aiming to close temporary POIs by year end
• NBN announces AI, IoT R&D with Sydney and Melbourne universities
• NBN complaints to the TIO triple
• Queensland examining fibre capacity while awaiting NBN
• NBN attempts to push speeds higher with new pricing
• Software automation policy guidelines (Tech Pro Research)