The ‘self-driving telco’ will take years to achieve, says ETSI’s AI initiative

The fully automated, ‘self-driving telco network’ is a one of the cornerstones of the next generation mobile vision for many operators, since it would dramatically improve operating costs and quality of service. Applying artificial intelligence (AI) to network management and orchestration (MANO) software could help achieve automation with a high level of intelligence, and a new ETSI ISG (Industry Specification Group) is focusing on this challenge – but warns it will take years to become commercial reality.

Densification and the rise of virtualized networks both introduce a huge number of new components into the picture, and this means modern networks are becoming hard for humans to understand quickly. This is the view of Ray Forbes of Huawei, who is the Convener of ETSI’s recently formed ISG, whose title is Experiential Network Intelligence (ENI). This kicked off in April with founder members including Verizon, China Telecom and, along with Huawei itself, the China Academy of Telecommunications Research, Hong Kong University of Science and Technology, Samsung, Xilinx and Samsung R&D Institute UK. Forbes says ETSI is talking to other operators like NTT Docomo, Orange and Deutsche Telekom about joining.

Forbes said, in an interview with LightReading, that many operators are incurring unnecessary cost by over-engineering their networks (adding surplus capacity to cope with peak loads) and running services at the least efficient times. He estimates that networks are typically 50% over-engineered, but that AI could reduce that to 10% by allocating resources more dynamically and efficiently according to workload, time of day and geographic patterns.

This would further enhance the resource efficiency impact which operators are already envisaging from virtualization and software-defined networking, which make it possible to dial network resources up and down on-demand (by calling up new instances of virtual network functions), while an SDN controller orchestrates those VNFs across the whole network.

AI would provide a new level of intelligence and automation for those SDN decisions, while also using machine learning to understand usage patterns in more detail, relating these to network resources, and predicting any faults and failures before they affect user experience. As China Telecom’s Haining Wang, vice chair of the ENI ISG, put it, while SDN and virtualization are helping to make networks more flexible, management complexity remains as high as ever – it is just transferred from hardware to software.

“It’s making resource management less human dependent in terms of knowledge,” Forbes said. “You want to have a business analyst who just performs data models; you don’t want to have a lot of pre-thinking and management analysts that analyze what to do.”

These are the practical objectives which underpin ETSI ENI’s remit, which is to define an AI-based context-aware system based on the ‘observe-orient-decide-act’ control model. This year, it will develop use cases and requirements for telco network AI, which it will model in 2018. It will then move on to testing and automation in 2019-2020.

In other words, this is not an overnight process. The technologies and processes are immature, with many unknowns, and few telcos are at the stage of planning AI-driven projects for the near future. For many, it will be part of a wider progress towards SDN and 5G, which will take place over as many as 10 years. And AI and machine learning are of their nature slow to perfect because the models need to learn until they reach the stage where they can make quicker, better decisions than a team of trained experts. “You need to design a model and spend a bit of time teaching the model how to do things,” Forbes said.

He added: “We don’t want to jump in the deep end”, cautioning that the dream of “fully AI-based autonomics” is far away, though ENI is taking steps in that direction.  “The ultimate aim is to improve the user experience and simplicity to have intent-based network indications in line with customer demand for new revenue,” he concluded.

To avoid reinventing the wheel, ENI will work with other ETSI initiatives like NFV, Multi-Access Edge Computing (MEC) and Next Generation Protocol (NGP), as well as other standards groups including IETF, Metro Ethernet Forum, 3GPP and Broadband Forum.

At the heart of ENI’s approach will be context awareness – the goal of many 5G-related projects, which envisage a whole new wave of mobile experiences, and revenue-generating services, which derive from the networks’ precise knowledge of a user’s context, preferences and habits.

The group says it will define an AI-enabled context aware system based on the ‘observe-orient-decide-act’ control model, which allows a system to adjust the services it is offering according to changes in user requirements, environmental conditions and business goals. This system will result in standardized models for using AI to manage a network and intelligently control service delivery and user experience.

The automated system will learn from previous experience and so add intelligence to automated network planning and monitoring, reducing opex, improving capacity usage, and boosting user experience, to a greater extent than just using automated technologies like self-optimizing network (SON).

The group will also study network telemetry, big data mechanisms, machine learning, and how to simplify and scale complex device configuration and monitoring.
“The unique added value of the ETSI ISG ENI approach is to define new metrics to quantify the operator’s experience; this enables the optimization and adjustment of the operator’s experience over time, taking advantage of machine learning and reasoning,” said Forbes when the ISG was announced in April.

While standards will be welcome, especially to encourage multivendor, interoperable networks, some operators are already embarking on this AI-driven path. Telefonica has deployed its first service operation centers for intelligent management of its networks, based on real time, AI-aided analysis of customer experience. These initial centers are in Argentina, Chile and Germany and the Spanish-based operator says they are a first step towards applying AI broadly to operations, proactively spotting incidents and taking action to improve customer experience. The system will be implemented in all Telefonica’s markets over time.

The technology underpinning this is Huawei’s Smartcare SOC (Service Operation Center) solution. This technology forms a bridge between network resource assets management and customer assets management, said Telefonica, providing per-service per-user (PSPU) visibility of customer experience.

Enrique Blanco, Telefonica’s global CTO, said in April: “These SOCs are the first step in bringing customer experience to the next level. In the near future the application of artificial intelligence to networks will maximize capacity and solve any problems before end users even notice anything. The final objective is to manage the network automatically to avert any potential problems.”

He added: “Machine learning is also becoming critical as operators virtualize their infrastructure: networks are becoming dynamic and exponentially more complicated to manage as the control is delegated to the network’s edge. As Telefonica moves towards the next generation of networks, intelligence and analytics are key, to turn data into a knowledge that enables real life innovation.”

Full context awareness combined with automated, AI-enabled network management is at the heart of many 5G visions. The UK-based 5G Innovation Center, for instance, last year showed off its Flat Distributed Cloud (FDC) framework for 5G.  This has context awareness at its core and uses information about the user and network status to provide better quality of experience over its dynamic and distributed cloud architecture, directing resources where they are most needed.

FDC is able to understand a user’s (or ‘thing’s) context and deliver just the right amount of capacity, latency and so on required for that particular situation. That would achieve one of the key aims of 5G – to support a user experience which is perceived to be always sufficient (as opposed to over-delivering, for a given usage, on some occasions and then disappointing on others).

The architecture will enable the network to access and act upon context information automatically and, based on that, make decisions about the right levels of QoS. The user would share information with the network via a user profile, which would then provide the ‘5 Ws’ – what, when, where, who, why – to support more intelligent, personalized decisions by systems such as SON.