Source-linked AI summary
Artificial Intelligence-Enabled Intelligent 6G Networks
Helin Yang, Arokiaswami Alphones, Zehui Xiong, Dusit Niyato, Jun Zhao, Kaishun Wu
TL;DR
Rising traffic, diverse applications, and stringent 6G requirements motivate a more systematic approach to intelligent network design. The paper proposes a four-layer AI-enabled architecture and reviews AI applications for 6G optimization. It concludes by identifying computation efficiency, robustness, hardware, and energy management as future research directions.
Problem
Earlier AI-enabled wireless-network studies did not systematically connect environmental sensing, data analysis, knowledge discovery, and 6G performance optimization.
Method
The paper proposes a four-layer AI-enabled 6G architecture comprising sensing, data mining and analytics, control, and application layers.
Results
The paper presents AI-enabled applications for mobile edge computing, mobility and handover management, and spectrum management, alongside future research directions.
Takeaways & Limitations
The proposed architecture is intended to support diverse services, optimize network performance, and guarantee seamless connectivity.
Abstract
from arXiv · showhide
With the rapid development of smart terminals and infrastructures, as well as diversified applications (e.g., virtual and augmented reality, remote surgery and holographic projection) with colorful requirements, current networks (e.g., 4G and upcoming 5G networks) may not be able to completely meet quickly rising traffic demands. Accordingly, efforts from both industry and academia have already been put to the research on 6G networks. Recently, artificial intelligence (AI) has been utilized as a new paradigm for the design and optimization of 6G networks with a high level of intelligence. Therefore, this article proposes an AI-enabled intelligent architecture for 6G networks to realize knowledge discovery, smart resource management, automatic network adjustment and intelligent service provisioning, where the architecture is divided into four layers: intelligent sensing layer, data mining and analytics layer, intelligent control layer and smart application layer. We then review and discuss the applications of AI techniques for 6G networks and elaborate how to employ the AI techniques to efficiently and effectively optimize the network performance, including AI-empowered mobile edge computing, intelligent mobility and handover management, and smart spectrum management. Moreover, we highlight important future research directions and potential solutions for AI-enabled intelligent 6G networks, including computation efficiency, algorithms robustness, hardware development and energy management.
I. INTRODUCTION
6G networks are motivated by rising traffic, diverse applications, stringent requirements, and the need for seamless connectivity across heterogeneous environments. The article proposes an AI-enabled architecture to support intelligent network optimization and services.
- 6G is being pursued to support high-quality services, emerging applications, and connectivity for massive numbers of smart terminals.
- The integrated space–air–ground–underwater network is presented as a potential 6G architecture for near-instant and seamless super-connectivity.
- Compared with 5G, 6G must meet ultrahigh data rates, ultralow latency, ultrahigh reliability, and seamless connectivity.
- Because 6G is large-scale, complex, dynamic, and heterogeneous, it calls for a flexible, adaptive, agile, and intelligent architecture.
- Prior work applied AI to wireless networks but did not systematically connect environmental sensing, data analysis, knowledge discovery, and performance optimization for 6G.
- The proposed architecture has sensing, data mining and analytics, control, and application layers supporting resource management, automatic adjustment, service provisioning, and self-configuration, self-optimization, and self-healing.
II. AI-ENABLED INTELLIGENT 6G NETWORKS
The paper frames AI as a means to perform performance optimization, knowledge discovery, sophisticated learning, organization, and decision making in complex 6G networks. It organizes the architecture into four intelligent layers and reviews major AI techniques.
- AI techniques are positioned to help 6G networks perform performance optimization, knowledge discovery, sophisticated learning, structure organization, and complicated decision making.
- The proposed architecture consists of intelligent sensing, data mining and analytics, intelligent control, and smart application layers.
- Machine learning includes supervised, unsupervised, and reinforcement learning, while deep learning uses multilayer artificial neural networks.
- Supervised learning builds models from labeled data for classification or regression, whereas unsupervised learning discovers patterns and useful features from unlabeled data.
- Reinforcement-learning agents map situations to actions through environmental interaction to maximize long-term reward.
A. Intelligent Sensing Layer
The intelligent sensing layer collects dynamic data from physical environments, but large-scale 6G sensing faces uncertainty, high dimensionality, and simultaneous spectrum-monitoring demands. AI methods are proposed for accurate, real-time, and robust sensing.
- Intelligent sensing gathers dynamic, diverse, and scalable data through devices or human crowds interfacing directly with physical environments.
- AI-enabled sensing targets high accuracy, real-time operation, and robustness because 6G requires ultrahigh reliability and ultralow latency.
- Fuzzy SVM and nonparallel hyperplane SVM address environmental uncertainty, while CNN-based cooperative sensing improves accuracy with low complexity.
- Large-scale spectrum sensing is difficult because many devices sense simultaneously, producing high-dimensional search problems.
- SVM and DNN models classify spectrum feature vectors into spectrum idle and spectrum busy classes and adaptively update the models.
B. Data Mining and Analytics Layer
The data mining and analytics layer processes massive raw data, reduces dimensionality and abnormalities, and discovers knowledge that can guide 6G management and adaptation.
- The data mining and analytics layer processes and analyzes massive amounts of raw data to discover useful information and form valuable knowledge.
- Dimensionality reduction transforms higher-dimensional data into lower-dimensional subspaces, decreasing computing time, storage space, and model complexity.
- PCA and ISOMAP can compress channel information, traffic flows, images, and videos into useful variables and filter abnormal data.
- Data analytics extracts meaningful features and patterns from data collected across physical, cyber, and social environments.
- Discovered knowledge can support resource management, protocol adaptation, architecture slicing, cloud computing, and signal processing.
- Examples include understanding UAV mobility patterns, establishing satellite–ground channel path-loss models, and predicting device behavior.
C. Intelligent Control Layer
The intelligent control layer uses learning, optimization, and decision-making to adapt complex 6G network parameters, resources, and architectures. AI supports self-configuration, self-optimization, self-organization, and self-healing across diverse network-control tasks.
- Learning: Learning uses existing knowledge and experience to improve device or service-center behavior toward optimal 6G operation.
- Intelligent control: AI-based control helps 6G networks achieve self-configuration, self-optimization, self-organization, and self-healing.SDN/NFV integration supports fast learning, quick adaptation, and intelligent softwarization, cloudization, virtualization, and slicing.
- Optimization: Traditional optimization faces suitability limits because 6G networks are significantly dynamic and complex.The layer targets global objectives such as QoS, QoE, connectivity, and coverage through AI-enabled optimization.
- Optimization: AI techniques optimize 6G network parameters, resources, and architectures instead of relying on traditional tedious computation.This supports agile adaptation to services and devices in dynamic networks.
- Decision-making: Decision-making enables agents to explore new knowledge, exploit existing knowledge, and select actions satisfying service requirements.Applications include precoding, routing, spectrum selection, and network association.
D. Smart Application Layer
The smart application layer delivers application-specific services and evaluates them across quality and resource-efficiency dimensions. AI supports automated services and smart-domain applications while managing devices and infrastructures for network self-organization.
- Service provisioning: The layer delivers application-specific services to people and evaluates provisioned-service performance before feeding results back into the intelligence process.
- Smart applications: AI-enabled management supports automated services, smart cities, smart industry, smart transportation, smart grids, and smart health.
- Smart applications: The layer manages smart devices, terminals, and infrastructures through AI techniques to realize network self-organization.
- Service evaluation: Service evaluation considers QoS, QoE, collected-data quality, learned-knowledge quality, and resource-efficiency costs.Cost metrics include spectrum, computational, energy, and storage efficiency.
- Service evaluation: These evaluation metrics support intelligent resource management, automatic network slicing, and smart management functions.
III. ARTIFICIAL INTELLIGENCE TECHNIQUES FOR 6G NETWORKS
AI techniques provide intelligence for 6G through mobile edge computing, mobility and handover management, and spectrum management. In MEC, centralized cloud and edge resources support learning, prediction, optimization, and real-time control.
- Scope: AI applications for 6G include AI-empowered mobile edge computing, intelligent mobility and handover management, and smart spectrum management.
- Deep reinforcement learning: DRL searches for resource-management policies under high-dimensional observation spaces, while experience replay improves learning efficiency and accuracy.This supports high-quality services for edge devices.
- MEC architecture: The AI-empowered MEC framework combines central cloud computing with edge computing.Cloud servers can run complex centralized algorithms, while edge servers use lightweight algorithms under limited capability.
- Edge intelligence: RL-based edge resource management learns environment dynamics and makes real-time control decisions without historical knowledge.States include mobility, requirement dynamics, and resource conditions; actions include energy management, resource allocation, and task scheduling.
- Cloud intelligence: Centralized AI supports service recognition, traffic and behavior prediction, security detection, and customized traffic-flow decisions.AI-based clustering can obtain MEC server association instead of relying on individual decisions, reducing participants.
B. Intelligent Mobility and Handover Management
AI-based mobility and handover management addresses the high dynamics, dimensionality, and service demands of 6G networks. DRL and predictive learning optimize mobility decisions and handover parameters to reduce latency and failures while maintaining connectivity.
- Challenges: Mobility and handover management are challenging in 6G because networks are highly dynamic, multi-layered, and large-dimensional.Vehicles and UAVs also impose high-speed, delay-sensitive, reliable, and continuous communication requirements.
- Predictive mobility management: AI techniques predict mobility and movement states, then predict vehicle trajectories to optimize handover parameters and avoid connectivity failures.
- DRL-based management: DRL learns handover strategies online from device or UAV mobility behavior while minimizing transmission latency and guaranteeing reliable wireless connectivity.
- DRL-based management: UAV agents sense link quality, location, and velocity, then select mobility and handover actions using rewards based on connectivity, latency, and capacity.
- Outcomes: DRL can reduce latency and handover failure probability while providing better services for ground devices.
- Vehicular networks: Deep-learning mobility prediction and fuzzy Q-learning handover optimization can learn high-speed vehicular patterns and avoid frequent or failed handovers.LSTM uses previous and future mobility contexts to learn sequences of future time-dependent handovers.
C. Intelligent Spectrum Management
The paper presents deep learning-based flexible spectrum management for 6G, using spectrum data to discover usage characteristics and provide real-time strategies for massive connectivity and diverse services.
- Flexible spectrum management: AI-enabled spectrum management supports massive connectivities and diverse services by intelligently assigning different spectrum bands.The framework is intended to match spectrum resources to traffic requirements in real time.
- Learning framework: The learning framework uses input, hidden or training, and output layers to process spectrum datasets and generate suitable management decisions.Input modules analyze current and previous spectrum utilization characteristics before producing output decisions.
- Learning framework: Offline training stores trained models, past experience, and developed rules to support smooth online spectrum management decisions.This separates model preparation from real-time spectrum management.
- Band-specific allocation: Visible light and THz bands can support high-capacity transmissions with large bandwidth, while low-frequency bands can broadcast short satellite–ground messages.The framework maps spectrum bands to different traffic-transmission requirements.
IV. FUTURE RESEARCH DIRECTIONS
The paper identifies future challenges for AI-enabled 6G networks in computation, robustness, hardware, and energy management. It highlights the need for efficient, adaptable learning and coordinated hardware–algorithm design.
- Computation Efficiency and Accuracy: Efficient AI learning schemes must address massive high-dimensional data, limited computing resources, computational complexity, and training accuracy.The paper describes improving computation efficiency and accuracy as a significant research challenge.
- Robustness, Scalability, and Flexibility of Learning Frameworks: Robust, scalable, and flexible learning frameworks remain an open issue because 6G network conditions, participants, and service requirements are highly dynamic.Relevant uncertainties include base-station associations, wireless channels, network topologies, mobility, and changing QoS and QoE requirements.
- Hardware Development: Hardware development is challenging because mmWave and THz components have high energy consumption and expensive cost, while some terminals have limited storage and computing energy.The paper points to hardware–algorithm collaboration, GPUs, and transfer learning as potential directions.
- Energy Management: Energy management is critical because undersea, air, and space infrastructures and some sensors cannot connect to power stations, while 6G must connect massive low-power devices.Potential AI-assisted approaches include managing consumption, energy harvesting, and wireless power transfer.
V. CONCLUSION
The paper proposes an AI-enabled intelligent architecture for 6G networks and reviews AI applications across network deployment and management. It concludes by identifying promising research directions and potential solutions.
- Architecture: The proposed AI-enabled architecture aims to support diverse services, optimize network performance, and guarantee seamless connectivity.The architecture is presented as the paper’s central design contribution.
- AI-enabled applications: Reviewed applications include AI-empowered mobile edge computing, intelligent mobility and handover management, and smart spectrum management.These applications address different aspects of 6G network deployment and management.
- Future directions: The paper highlights promising future research directions and potential solutions for AI-enabled 6G networks.The conclusion follows the discussion of unresolved challenges and research opportunities.