AI-Native 6G: A New Era of Intelligent Networks
As the world accelerates toward a fully connected, intelligent future, the next generation of mobile communication—6G—is no longer just a vision but a rapidly approaching reality. Unlike its predecessors, 6G is not merely about faster speeds or lower latency. It is being designed from the ground up to be a cognitive, self-optimizing, and deeply intelligent network ecosystem. At the heart of this transformation is artificial intelligence (AI), which is no longer an add-on or optimization tool but a native, foundational element of the 6G architecture. This paradigm shift is the focus of groundbreaking research led by Wang Dong, Guan Wanqing, Zhang Haijun, and Long Keping from the University of Science and Technology Beijing (USTB), published in Radio Communications Technology.
Their comprehensive study, titled 6G Network Technology Based on Artificial Intelligence, presents a holistic framework for embedding AI into the very fabric of 6G networks. The research goes beyond conventional applications of machine learning in network optimization. Instead, it proposes a reimagined network architecture where AI is not just applied but integrated natively—across the physical layer, medium access control (MAC), protocol stack, and network slicing—enabling unprecedented levels of automation, efficiency, and adaptability.
The significance of this work lies in its timing and depth. As 5G networks continue to roll out globally, the telecommunications industry is already looking ahead to 6G, with standardization efforts expected to begin in earnest by 2025. The paper by the USTB team provides a timely and technically robust roadmap for how AI can be systematically embedded into 6G systems to meet the complex demands of future applications—from autonomous vehicles and smart cities to immersive extended reality (XR) and real-time holographic communications.
One of the central themes of the research is the concept of “AI-native” architecture. Traditional networks, including 5G, treat AI as an external intelligence layer—something that runs on top of existing infrastructure to improve performance. In contrast, the authors argue that 6G must be built with AI as a first-class citizen. This means designing network components, protocols, and interfaces with AI capabilities in mind from the outset. The result is a network that can perceive, learn, reason, and act in real time, without relying on manual intervention or pre-programmed rules.
The proposed AI-native architecture is built on three key pillars: intelligent resource management, AI-powered network maintenance, and AI-enhanced air interface technologies. Each of these components plays a critical role in transforming 6G from a passive communication pipeline into an active, intelligent platform.
The first pillar, intelligent resource management, addresses one of the most pressing challenges in modern wireless networks: the efficient allocation of limited radio resources. With the explosion of connected devices and diverse service requirements—from ultra-reliable low-latency communications (URLLC) to massive machine-type communications (mMTC)—static resource allocation schemes are no longer sufficient. The authors propose a dynamic, AI-driven approach that leverages deep reinforcement learning (DRL) algorithms such as Double Deep Q-Network (DDQN), Dueling DQN, and Asynchronous Advantage Actor-Critic (A3C) to enable real-time, adaptive resource scheduling.
These algorithms allow the network to learn optimal policies through continuous interaction with its environment. For example, in the context of network slicing—a key 5G/6G technology that enables the creation of multiple virtual networks on a shared physical infrastructure—the AI engine can dynamically allocate radio resources to different slices based on real-time traffic patterns, service level agreements (SLAs), and quality of service (QoS) requirements. The system observes the current state of the network (e.g., data throughput rates of various slices), selects an action (e.g., assigning a certain number of resource blocks), receives a reward (e.g., based on whether the target rate is met), and updates its policy accordingly. Over time, the AI agent converges to an optimal strategy that maximizes long-term performance while minimizing resource waste.
What sets this approach apart is its ability to handle uncertainty and variability. Traditional optimization methods often rely on deterministic models that assume perfect knowledge of network conditions. In reality, wireless environments are highly dynamic, with fluctuating channel conditions, user mobility, and traffic loads. AI-based methods, particularly reinforcement learning, excel in such stochastic environments by learning from experience rather than relying on fixed models. The authors demonstrate that their AI-driven scheduling algorithms can achieve performance close to that of traditional iterative optimization methods, but with significantly lower computational complexity—making them suitable for real-time deployment in large-scale networks.
The second pillar of the AI-native 6G vision is intelligent network maintenance. As 6G networks become more complex, with billions of connected devices and heterogeneous infrastructure spanning terrestrial, aerial, and satellite domains, traditional network operations and maintenance (O&M) models are becoming unsustainable. Manual configuration, periodic monitoring, and reactive troubleshooting are no longer viable. The authors propose a fully automated, AI-powered O&M framework that enables zero-touch network operations.
This framework leverages the massive amounts of operational data generated by the network—such as performance metrics, alarm logs, configuration records, and topology information—to enable three types of AI-driven learning: spatial, temporal, and logical. Spatial learning allows the AI system to understand the network’s topology and relationships between different network elements. By analyzing historical alarm patterns and device dependencies, the AI can quickly pinpoint the root cause of a failure, even in a highly distributed environment. This capability is particularly valuable in large-scale networks where a single fault can trigger a cascade of alarms across multiple nodes.
Temporal learning focuses on time-series analysis of network performance. By establishing baseline behavior patterns from historical data, the AI system can detect anomalies in real time. For example, if a base station’s throughput suddenly deviates from its normal range, the system can issue an early warning before the issue escalates into a full-blown outage. This predictive maintenance capability not only improves network reliability but also reduces operational costs by preventing service disruptions.
Logical learning involves understanding the causal relationships between different events. The AI system learns how specific alarm patterns correlate with different types of failures and their severity levels. Over time, it builds a knowledge base that enables more accurate and faster diagnosis. This is particularly useful in complex scenarios where multiple factors contribute to a problem, such as interference, hardware degradation, or software bugs.
The authors emphasize that this intelligent maintenance system is not a standalone tool but an integral part of the AI-native architecture. It is supported by a distributed computing infrastructure that spans the cloud, fog, and edge layers. This multi-tiered computing model ensures that AI processing can be performed close to the data source, reducing latency and bandwidth consumption. For example, real-time anomaly detection can be performed at the edge, while long-term trend analysis and model training can be offloaded to the cloud. The fog layer acts as an intermediary, handling intermediate processing tasks and enabling coordination between edge and cloud nodes.
The third and perhaps most revolutionary pillar of the AI-native 6G framework is the integration of AI into the air interface—the physical and data link layers that govern wireless transmission. Traditionally, these layers have been designed based on mathematical models and fixed protocols. However, the authors argue that AI can fundamentally transform how wireless signals are processed, transmitted, and received.
One of the most promising applications is in channel estimation and feedback. In massive MIMO (Multiple-Input Multiple-Output) systems, which are expected to be a cornerstone of 6G, the number of antennas can reach into the hundreds or even thousands. This creates an enormous challenge for channel state information (CSI) acquisition, as the amount of feedback data grows quadratically with the number of antennas. The authors propose using deep learning models, such as the Channel-State-Information-Net (CsiNet), to compress and reconstruct CSI more efficiently. By treating the channel matrix as a two-dimensional image, convolutional neural networks (CNNs) can learn to extract and encode the most relevant features, significantly reducing the feedback overhead without sacrificing accuracy.
Another key application is in beamforming and power allocation. In dynamic environments with high user mobility and changing interference conditions, traditional beamforming algorithms struggle to maintain optimal performance. AI-based approaches, particularly deep reinforcement learning, can adapt beam patterns in real time based on feedback from the network. Each node can learn to adjust its beam direction and power level to maximize signal quality while minimizing interference to neighboring cells. This results in higher spectral efficiency and better user experience.
The authors also explore the potential of end-to-end learning for the entire communication chain. Instead of designing each layer of the protocol stack independently, AI can be used to jointly optimize the physical, MAC, and network layers. This holistic approach allows the system to discover novel transmission strategies that may not be apparent from traditional signal processing theory. For instance, AI could learn to modulate signals in ways that are robust to specific types of noise or interference, or to dynamically adjust packet sizes and retransmission strategies based on real-time channel conditions.
A critical enabler of this AI-native vision is the AI engine—a centralized intelligence module that orchestrates all AI-driven functions across the network. The AI engine is not a single monolithic system but a distributed, modular platform that can be deployed across the cloud, fog, and edge. It consists of four main components: data acquisition, data preprocessing, pre-trained AI models, and machine learning algorithms. The engine collects vast amounts of network data, cleans and normalizes it, and then applies various AI models to generate actionable insights.
The AI engine serves as the “brain” of the 6G network, providing intelligent control services to other network components. It supports interoperability between different layers and domains, enabling seamless coordination between the radio access network (RAN), core network, and transport network. For example, when a user moves from one cell to another, the AI engine can predict the handover event, pre-allocate resources in the target cell, and optimize the handover parameters to minimize latency and packet loss.
The research also highlights the importance of a global, fine-grained network measurement system. To make accurate decisions, the AI engine must have access to real-time, high-resolution data about the network’s state. The authors propose a unified measurement architecture that leverages in-band telemetry and programmable data planes to collect detailed performance metrics with minimal overhead. This measurement system provides the foundation for AI-driven optimization, enabling the network to continuously monitor, analyze, and adapt to changing conditions.
One of the most compelling aspects of the paper is its practical orientation. While many discussions about AI in 6G remain theoretical, the USTB team provides concrete algorithms, system designs, and performance evaluations. They demonstrate how AI can be used to solve real-world problems such as network slicing, dynamic spectrum sharing, and energy-efficient resource allocation. Their work bridges the gap between academic research and industrial application, offering valuable insights for network operators, equipment vendors, and standards bodies.
The implications of this research extend far beyond technical performance. By making networks more intelligent and autonomous, AI-native 6G has the potential to democratize access to advanced communication services. It can enable truly ubiquitous connectivity, where even remote and underserved areas benefit from high-quality, adaptive networks. It can also reduce the environmental impact of telecommunications by optimizing energy consumption and resource utilization.
Moreover, the AI-native approach opens up new business models and service opportunities. Network operators can offer AI-powered services such as real-time traffic prediction, intelligent content delivery, and personalized user experiences. Enterprises can leverage AI-driven network slicing to create private 6G networks tailored to their specific needs, from factory automation to smart healthcare.
In conclusion, the work by Wang Dong, Guan Wanqing, Zhang Haijun, and Long Keping represents a significant step forward in the evolution of mobile communication. Their vision of AI-native 6G is not just about incremental improvements but a fundamental rethinking of what a network can be. By embedding intelligence into every layer of the system, they are paving the way for a future where networks are not just fast and reliable, but truly smart.
As 6G development progresses, the ideas presented in this paper are likely to influence the direction of global standardization efforts. The integration of AI into the core design of 6G networks will require close collaboration between academia, industry, and regulatory bodies. But if the vision outlined by the USTB team is realized, the result will be a network that is not only technologically advanced but also more efficient, sustainable, and human-centric.
The journey to 6G has only just begun, but one thing is clear: artificial intelligence will be at its core. And with researchers like Wang, Guan, Zhang, and Long leading the way, the future of communication is looking smarter than ever.
Wang Dong, Guan Wanqing, Zhang Haijun, Long Keping, University of Science and Technology Beijing, Radio Communications Technology, doi:10.3969/j.issn.1003-3114.2021.06.007