<
//

Autonomous networks

Wireless networks are becoming more flexible, spectrally efficient and cost effective. 5G adds the capacity and latency to support a wider range of use cases, device types and network architectures. But this comes with a price tag: increased complexity.

We have to learn to manage this increased complexity if we want to take advantage of the benefits of technology evolution – 5G, edge, network slicing, virtualization and so on – and of the pervasive connectivity that IoT needs. Managing complexity requires a fundamental change in how we operate and manage networks, which goes beyond the adoption of 5G or other technologies and requires a new end-to-end architecture.

As we move from manually driven operations that focus on specific network elements – the EMS approach – to more advanced end-to-end operations, management and analysis in closed-loop, fully autonomous networks, the underlying network architecture will have to evolve.

For a successful evolution, we need to ensure that the foundations of this new architecture are in place as we move to 5G – or indeed, before that, as we plan to move from 4G to 5G. The end-to-end network transformation will not happen overnight, and will not be completed during the initial stages of 5G rollouts.

It is not only the network architecture that has to change. Operators and vendors also need to introduce a new culture that breaks down the existing business-unit silos, and that rewards retraining, innovation and risk-taking among employees.

The six-level model for the assessment of network intelligence under evaluation at the ITU and promoted by GSMA, IEEE and TMF provides a useful framework to articulate how automation, AI and ML can take us to autonomous networks, and how networks will change in the process. The model is based on the SAEJ3016 Automatic Driving Levels, developed for self-driving vehicles.

Both in the vehicle and in the wireless network cases, the evolution cannot be easily reduced to an increasing adoption of automation and AI/ML – e.g., the percentage of operations that are automated – but should be thought of as a sequence of qualitatively different phases that affect different parts of the network and different levels of network management differently:

  • Tasks that involve context awareness and analysis are among the first to be automated (Level 1), because they require the collection and processing of large data sets. Manual analysis quickly becomes inadequate to cope with the growing amount of data that operators can now collect and analyze.
  • Initially non-real-time optimization and prediction will still be conducted and overseen by humans, using rules and algorithms based on human expertise. Starting at Level 2, it will gradually shift to AI/ML.
  • Decision making and execution based on non-real-time optimization and prediction will subsequently become automated at Level 3.
  • Real-time prediction and interference cannot be done in scale and at a sufficiently low time resolution by humans, and it will require AI and ML (Level 4).
  • Exception handling will be the last component be introduced (Level 4), because it requires the network to act against rules, algorithms or automated learning that were previously created and that are currently in use. The network will have to correctly decide what needs to be changed or temporarily set aside. This is a difficult task that requires very advanced and reliable AI and ML tools.
  • With fully autonomous networks (Level 5), networks will move to an unsupervised, closed-loop learning and intent-based architecture. Moving beyond rules and algorithms based on historical data, the final shift from automated networks to fully autonomous networks is the most ambitious one. Wireless networks will not only use automated processes for basic operations, they will also dynamically adapt to new environments and conditions, and optimize network performance and service quality in real time.

For a network to be fully autonomous automation and intelligence have to be introduced and managed end-to-end.

In the early stages, this will happen at the element level and be confined to specific features – for instance, to optimize massive MIMO or SON. This is a necessary first step to ensure that the transition from human-based to machine-based works smoothly and to refine the approach to the transition.

In later stages, the scope will extend to network subsystems, such as the RAN, the core and the OSS, but the timeline may vary. For instance, real-time operation and optimization may be introduced in the RAN ahead of the core, because more architectural changes are needed for real-time processing in the core than in the RAN, and real-time processing is more valuable in the RAN.

Eventually automation and intelligence will have to extend to the end-to-end system to ensure that the overall network performance is optimized. It is crucial to avoid a situation where elements or subsystems in the network introduce inconsistencies that result in optimized performance at the element and subsystem levels, but not at the end-to-end level. These inconsistencies may lead to suboptimal QoE or network inefficiencies.

At the final stage, use cases such as detection of security threats through anomaly detection, service assurance, real-time traffic management and policy will gain prominence as they require end-to-end automation and intelligence.

An autonomous network will also make it possible for operators to develop, provision and optimize new revenue-generating services that require real-time management of network slicing or URLLC – for instance, for mission-critical applications that demand high reliability, low latency and stringent security.

Much work lies ahead for wireless networks to become automated, intelligent and, eventually, autonomous. The six-level model indicates a path that can guide us through this evolution. We have to examine multiple dimensions, act at multiple levels within the networks, and develop a new network architecture to extract the benefits from the complexity that 5G brings alongside its advanced capabilities.

We thank Huawei for sponsoring this blog.