AI-native networks and the Implications for 6G

By Iain Gillott, Senior Research and Technical Advisor, WIA

As the wireless world debates what 6G will look like, how it will work and what spectrum bands it will use, one thing seems certain: Artificial Intelligence (AI) will be incorporated into the new standards to a degree not seen before.  Rather than use AI to improve an existing infrastructure, as is being done with 5G, AI will form a foundational pillar of the new 6G infrastructure with AI deeply embedded across all network layers. Researchers today refer to these new architectures as AI-native networks.

This final blog in this AI series explores in more detail how AI is likely to evolve in the Radio Access Network (RAN) and how it could shape 6G architectures and services, specifically:

• AI as a native foundation of 6G;
• Real-time closed loop automation;
• Distributed learning and AI collaboration;
• AI-enhanced network sensing and environment awareness;
• Ethical AI and governance in 6G; and
• Strategic implications.

AI as a Native Foundation of 6G

With 5G, AI is bolted onto existing architectures and networks – 5G was not designed, or even imagined, to use AI to the degree anticipated today.  But 6G is being designed with AI as a native, built-in function from the very start. The likely implications of this are that:

• AI will define how the network operates at a core level, not just optimize existing operations and parameters;
• AI models will likely be responsible for autonomous service creation, zero-touch management, and intent-based orchestration – in essence, the 6G network will define, manage and operate itself;
• Similarly, the fundamental 6G RAN design will be driven by AI models that optimize the physical layer functions, including waveform selection, coding schemes and spatial multiplexing strategies.

As with today’s AI models, continuous learning will also apply to the AI-native 6G network. Embedded feedback loops will enable real-time adaptation and self-improvement of the model, which will then be applied to the network.

Real-Time Closed-Loop Automation

Today’s 5G networks use Self Optimizing Network (SON) concepts. (This effort was actually started back in 4G LTE). In 6G, these concepts will extend to self-evolving networks where network architectures, resource allocation and even software functions are dynamically and continuously updated and refined.

One of the main goals for 6G is the implementation of fully closed-loop, real-time network automation such that:

• AI agents continuously observe network behavior;
• Decision-making and enforcement occur autonomously;
• Performance is validated and tuned with minimum human intervention.

This means that the RAN will fundamentally change from being simply a passive layer that transmits and receives data to and from end-user devices into a context-aware system driven by AI.  

Distributed Learning and AI Collaboration

Just as the RAN is distributed across the country, 6G will require massively distributed AI systems. These distributed AI models will be:

• Trained across thousands of edge nodes using federated learning to minimize privacy and security risks;
• Continuously updated through multi-agent learning – the various RAN elements will work together across the network to improve the performance of the AI model; and
• Able to adapt to changing environments without the need to retrain a new AI model from scratch.

The AI fabric incorporated into the 6G network therefore will be distributed to provide reliability, adaptability and scalability to meet dynamic service demands as the 6G network, and the applications it enables, evolve.

AI-Enhanced Network Sensing and Environment Awareness

Network sensing will be integrated into 6G infrastructure such that the RAN senses its own environment, as well as transmitting and receiving data. This means that:

• The RAN likely becomes a distributed sensor grid, capable of detecting motion, presence and object shapes in its immediate vicinity and further afield via reflected signals; and
• Machine learning models will interpret these signals to potentially enable services such as traffic monitoring, intrusion detection and real-time environmental mapping.

These sensing and detection functions will mean that 6G will enable a range of new cross-industry use cases, from smart buildings and infrastructure monitoring to precision agriculture and autonomous vehicle coordination.  Rather than using a GPS signal (as many devices do today) which can be blocked, the RAN itself will be used to determine location and movement.

Ethical AI and Governance in 6G

The extensive use of AI in upcoming 6G networks raises new questions of ethics and governance.  In short, procedures and frameworks will be required that enable:

• Fairness in algorithmic decision-making such as resource allocation to ensure that all parts of the network have access to the necessary capacities and capabilities;
• Explainability and accountability for autonomous actions – network changes and actions must be audited and accountable;
• Global standards for secure, privacy-preserving AI operations.

Standards organizations such as ITU, ETSI, and IEEE are already working on principles for trustworthy AI in telecommunications, which will be critical for 6G. The goal is to enable these trust and accountability frameworks at the start of 6G network design and implementation and not to have to add the required features later on.

Conclusions

For mobile network operators (MNOs), vendors and policymakers, the transition to AI-native RAN in 6G is likely to result in:

• Increased reliance on cross-domain expertise in AI, telecom and software engineering. RAN engineers will be as familiar with AI models and concepts as they are with waveforms and interference models today;
• Investment in AI infrastructure and edge computing capacity. AI models will need to be processed at the edge of the network to provide the necessary performance; and
• Adoption of open ecosystems that support rapid deployment of new AI models and services across the entire network.

As this series of articles has discussed, AI algorithms and models can be used today across the RAN to optimize a variety of functions.  In future 6G networks, AI will be used natively as a fundamental foundation of the network.  The RAN has always been a key part of the overall efficiency of a mobile network and key to end user experience. As networks have become more complex, so the RAN has become both more important and harder to deploy, tune and manage.  AI helps address these challenges today and ultimately will allow:

• Higher spectral efficiency, allowing more data to flow through the same bandwidth;
• Improved user experience, with reduced interference and fewer dropped sessions; and
• Operational flexibility, enabling MNOs to support more use cases within the same spectrum portfolio.

Future 6G networks embed AI as a native feature, with edge-based training, inference and closed-loop control fully integrated into the network fabric.  Deploying 6G will not be easy. There are many decisions to be made and steps that need to be taken, least of all the identification and licensing of sufficient and suitable spectrum.  But for 6G, AI will be a fundamental building block.