Compared to previous types of networks, 5G networks are both more in need of automation and more susceptible to automation.Automation tools are still being developed and machine learning is not yet common in operator-class networks, but rapid changes are expected.
New standards from 3GPP, ETSI, ITU and the open source software community provide for increased use of automation, artificial intelligence (AI) and machine learning (ML). And the activities of key providers add confidence to the vision and promise of artificially intelligent network operations.
“The growing complexity and the need to solve recurring problems in 5G and future radio systems require new automation solutions that take advantage of state-of-the-art artificial intelligence and machine learning techniques that increase system efficiency, ”Eric Ecudden, Ericsson’s Chief Technology Officer (CTO), wrote recently.
In 2020, Ericsson engineers demonstrated machine learning software that organizes virtual machines on a web server. They reported that during a 12-hour stress test, their software reduced idle cycles to 2%, from baseline to 20%. Such efficiency gains could enhance the collections of modern computers and computers inside cloud 5G infrastructure.
Given that 5G core networks are evolving to increase reliance on software and shared computing resources, Ericsson’s demonstration suggests that the widespread use of AI solutions can help carriers use infrastructure as efficiently as possible while processing a combination of types of traffic that change dynamically and implement various service-level agreements.
The Nokia White Paper describes the potential roles of AI and ML in virtually all phases of a service provider’s operations. Last month, Nokia announced the availability of its software activation platform, whose features include a tool for using AI and ML in endpoints that work with both open radio access networks (O-RAN) and application-level services. Nokia’s platform provides data that is important for the development of machine learning for software-defined radio stations.
Carriers and third parties may develop software for the Nokia platform that comes with some samples that are currently under commercial testing. One of the included “xApp” relies on machine learning methods for traffic management – roughly speaking, a kind of load balancing for service-oriented radio channels.
The need for machine learning arises from the expected challenges in managing future 5G networks. Future deployments are likely to have a capacity to transmit traffic of an order of magnitude larger than existing infrastructure. Many vendors, researchers and developers expect to need machine learning to use 5G technology effectively.
Opportunities to use machine learning appear with increasing dependence on local resources in telecommunications networks. Carriers are also experiencing the same powerful currents that are pushing many industries to “software”, use virtual machines, DevOps principles, and other global vectors for intelligent automation.
Telecommunications providers and advanced researchers are developing machine learning software that, for example, controls smart antennas with a split time, sets and reassigns bandwidth in the package core, and orchestrates assignments for virtual machines on the end computer.
In essence, the software plays a game designed to predict traffic congestion and use the least amount of traffic resources in accordance with service level agreements. The expected result would improve the availability of resources to serve additional customers at times when the workload is at its peak. When loads are reduced, the software can cause the hardware to operate in standby mode to save power.
Rule-based scripts and statistical models can achieve some of these goals, but hand-crafted algorithms face challenges. A large number of parameters determine a connection event in a 5G network – more than in previous generations. Therefore, machine learning can be a requirement, not just an optimization tool, for resource efficiency in full-scale 5G operations.
Varieties of AI tasks in the cellular network
Recent reports explore a number of wireless communication applications that researchers and developers of machine learning are working on, which provides many candidate technologies for the carrier’s roadmaps.
From a business lifecycle perspective, there are opportunities for machine learning developments to accelerate network planning and design, operations, marketing, and other responsibilities that typically require an intelligent person. The developers focus on network management features, including fault management, configuration, accounting, performance, and security (FCAPS).
From a network technology perspective, machine learning applications in the R&D phases can affect each layer of the communication stack, from levels of physical layers and data communication layers, through media access, transport, switching, session , presentation and application layers.
On the lower layers of radio access networks, common computers process signals on the main band and they plan and generate directed radio beams by synchronizing many antenna elements. Machine learning systems can alleviate congestion by setting optimal modulation parameters and quickly scheduling beams that are calculated to meet immediate requirements.
At the higher layers of the communication stacks, the software provides opportunities to use and reuse virtual network functions (VNF) in dynamic combinations to cope with changes in traffic patterns. For example, smart systems can scale (automatically scale) temporary combinations of resources to support large video conferencing and reallocate those resources to other jobs after the event.
In packet core networks, smart selection is among the astronomical number of ways to mix and match network functions to reduce idle time while keeping customers satisfied. In radio access networks, smart tweaks to power levels, character sets, frame sizes, and other parameters promise to push the largest capacity out of the available spectrum.
Cybersecurity and confidentiality measures can also benefit from machine learning. In theory, intelligent domain isolation can automatically open and close access according to knowledge encoded in large databases, such as event logs. Distributed training methods can run on end computers and user devices, keeping personal data separate from centralized databases.
Self-managed network?
“As driverless cars require more time and resources to develop than some expected, the vision for fully autonomous networks seems to remain distant.”
Michael Gold
Juniper’s slogan “self-governing network” expresses a vision for autonomous communication services similar to autonomous vehicles. Many other network technology developers are embracing similar ideas. Engineers and traders often describe intention-based network (IBN), one-touch provision and management of network and zero-touch services.
Most providers are likely to use one of these phrases or a similar phase. They all refer to a subset of network operations that can be performed autonomously or almost. In fact, many software-defined network technology concepts rely on rule-based systems, a programming strategy developed by the artificial intelligence community decades ago.
Imaginatively, the future network manager may use natural language to revise the bandwidth limitation policy. But beware of the hustle and bustle around network automation. In some corporate networks, zero-touch nodes are configured automatically when a technician turns on a new stand. In contrast, the installation of a carrier class fiber bonding unit remains complicated.
As driverless cars require more time and resources to develop than some expected, the vision of fully autonomous networks seems to remain distant. One of the main challenges is to acquire and analyze an abundance of telemetry data in the networks of service providers.
Many systems do not expose the data required by data-driven machine learning systems to predict and respond to changes in traffic loads. Systems that provide telemetry use various protocols and data structures that complicate AI software development. Providers may see that telemetry data has a high value as intellectual property and is worthy of encryption.
The Nokia White Paper for 2020 advocates a multi-stage roadmap for managing opportunities and risks. Nokia recognizes that AI is a rarity in modern networks. More often, expert human network managers create, implement, and often adjust statistical and rule-based models that manage automated systems in telecommunications networks.
In between today’s model-driven practices and the future vision for stand-alone networks, Nokia sees the emergence of intent-driven network management processes activated by closed-loop automated systems. Automated resource orchestration will free human network managers to focus on business needs, service creation, and DevOps.
Open questions
“In a sense, the changing technological landscape is challenging network professionals to keep up with new developments. In another sense, artificial intelligence tools in different fields are usually performance enhancers, not redundancy generators. ”
Michael Gold
Does AI threaten the work of network managers? In a sense, the changing technological environment often challenges network professionals to keep up with new developments. In another sense, artificial intelligence tools in various fields are usually performance enhancers, not redundancy generators. Similarly, for physicians and lawyers, AI is a tool rather than a threat.
It seems that one or another player in the industry is always buzzing about “smart grids”. AT&T has been at this for the longest time, initially using the phrase in the 1980s to describe an early network computing initiative. Expectations of artificial intelligence in networks have been focused and redirected many times over the years. This time it may be different. Are we there yet?
Now that computers control or make up almost all network nodes, the software looks more agile on all layers of the communication stack. Business development will determine which AI and ML developments contribute the most to business results and customer experience, and which nodes in the network provide maximum leverage for machine learning software to add value.
