PaperlayerView exact revision on arXiv

Source-linked AI summary

Overview of AI and Communication for 6G Network: Fundamentals, Challenges, and Future Research Opportunities

Qimei Cui, Xiaohu You, Ni Wei, Guoshun Nan, Xuefei Zhang, Jianhua Zhang, Xinchen Lyu, Ming Ai, Xiaofeng Tao, Zhiyong Feng, Ping Zhang, Qingqing Wu, Meixia Tao, Yongming Huang, Chongwen Huang, Guangyi Liu, Chenghui Peng, Zhiwen Pan, Tao Sun, Dusit Niyato, Tao Chen, Muhammad Khurram Khan, Abbas Jamalipour, Mohsen Guizani, Chau Yuen

arXiv:2412.14538v4cs.NIcs.AIeess.SP

TL;DR

6G must integrate AI to address complex, dynamic communication demands while supporting intelligent applications. This review synthesizes AI–6G foundations, applications, three integration stages, challenges, and research opportunities, reporting benefits across network optimization and wireless functions but also major heterogeneity, complexity, sustainability, compatibility, and security constraints.

  • Problem

    Complex 6G environments require intelligent communication systems, but integrating AI across networks raises challenges involving heterogeneity, complexity, energy consumption, protocol compatibility, and security.

  • Method

    The paper provides a comprehensive review of AI and communication for 6G, covering AI-enabled wireless techniques, distributed intelligence, integration stages, challenges, and future opportunities.

  • Results

    The review reports AI-enabled gains in CSI feedback, beam management, localization, and privacy-preserving distributed learning for 6G networks.

  • Takeaways & Limitations

    AI–6G development spans AI4NET, NET4AI, and AIaaS, linking network optimization with network-supported AI operations and network-provided AI services.

  • Takeaways & Limitations

    Wireless network large models remain constrained by unresolved protocol compatibility and security threats involving privacy attacks, data tampering, and operational disruption.

Abstract

from arXiv · show

With the growing demand for seamless connectivity and intelligent communication, the integration of artificial intelligence (AI) and sixth-generation (6G) communication networks has emerged as a transformative paradigm. By embedding AI capabilities across various network layers, this integration enables optimized resource allocation, improved efficiency, and enhanced system robust performance, particularly in intricate and dynamic environments. This paper presents a comprehensive overview of AI and communication for 6G networks, with a focus on emphasizing their foundational principles, inherent challenges, and future research opportunities. We first review the integration of AI and communications in the context of 6G, exploring the driving factors behind incorporating AI into wireless communications, as well as the vision for the convergence of AI and 6G. The discourse then transitions to a detailed exposition of the envisioned integration of AI within 6G networks, delineated across three progressive developmental stages. The first stage, AI for Network, focuses on employing AI to augment network performance, optimize efficiency, and enhance user service experiences. The second stage, Network for AI, highlights the role of the network in facilitating and buttressing AI operations and presents key enabling technologies, such as digital twins for AI and semantic communication. In the final stage, AI as a Service, it is anticipated that future 6G networks will innately provide AI functions as services, supporting application scenarios like immersive communication and intelligent industrial robots. In addition, we conduct an in-depth analysis of the critical challenges faced by the integration of AI and communications in 6G. Finally, we outline promising future research opportunities that are expected to drive the development and refinement of AI and 6G communications.

1 Introduction

The paper frames AI–6G integration as necessary for intelligent, efficient wireless systems amid rising connectivity demands and unresolved data-management questions. It reviews foundational principles, integration issues, and future directions.

  • 6G is anticipated to provide ultra-low latency, higher data rates, greater reliability, and ubiquitous connectivity, motivating AI integration.
  • AI’s data-analysis and learning capabilities support intelligent network management, optimization, traffic prediction, and fault detection.
  • Semantic compression may reduce transmitted data and bandwidth, while AI-generated parameters, training data, and feedback may increase overall network data volume.
  • The review examines AI integration across 6G architecture, network elements, and functional processes, alongside technical enablers, applications, challenges, and research directions.
  • The article is organized around its review structure, with abbreviations collected in Appendix A.

1.1 Requirements and Challenges in the Post-5G Era

Post-5G networks face explosive traffic, constrained spectrum, rising energy and maintenance demands, and expanding application and security requirements. These pressures expose limitations in static management and existing security approaches.

  • 1.1.1 Better Utilization of Spectrum Resources: Explosive traffic growth and massive device connectivity increase pressure on limited spectrum resources.
  • 1.1.1 Better Utilization of Spectrum Resources: Static spectrum management cannot dynamically allocate spectrum and power to real-time service demands, causing resource waste or shortages.
  • 1.1.2 Lower-Carbon Wireless Coverage: 5G consumes about three times more energy than 4G, while requiring three to four times as many base stations for comparable coverage.
  • 1.1.3 More Intelligent Network O&M: Current network O&M relies largely on reactive manual troubleshooting, producing low automation and high maintenance costs.
  • 1.1.3 More Intelligent Network O&M: Virtualization introduces additional fault points, ambiguous supplier responsibility, and longer edge fault-detection response times.
  • 1.1.4 More Diverse Application Scenarios: Post-5G networks must support diverse industries and scenarios, including mobile communication, IoT, smart cities, and satellite internet.
  • 1.1.5 More Comprehensive Security: Existing security evolution emphasizes authentication, encryption, and integrity checks, while standards for mMTC and uRLLC lag behind eMBB.

1.2 6G Vision

The 6G vision links physical and virtual worlds through ubiquitous intelligence, sensing, communication, and computing. IMT-2030 defines diverse usage scenarios, development goals, and capability indicators, with AI integration as a central capability.

  • 1.2.1 Development Goals and Capability Indicators of 6G: 6G is envisioned as infrastructure connecting human, physical, and virtual worlds through digital twins and ubiquitous intelligence.
  • 1.2.2 Usage Scenarios of 6G: IMT-2030 defines six scenarios: immersive communication, hyper-reliable low-latency communication, massive communication, ubiquitous connectivity, AI and communication, and integrated sensing and communication.
  • 1.2.2 Usage Scenarios of 6G: These scenarios target applications including immersive XR and holographic communication, industrial robotics, remote care, smart cities, underserved-region connectivity, and distributed AI.
  • 1.2.2 Usage Scenarios of 6G: Integrated sensing and communication supports wide-area multidimensional sensing for navigation, activity detection, and motion tracking.
  • 1.2.1 Development Goals and Capability Indicators of 6G: 6G development goals include inclusivity, ubiquitous connectivity, sustainability, innovation, security/privacy/resilience, standardization/interoperability, and accessibility.
  • 1.2.3 Capabilities of 6G: AI-related 6G capabilities progress from self-optimization, to support for AI operations, to AI services for users and equipment.

1.3 Overview of AI and Communication Integration in 6G

The paper presents AI and wireless communication as mutually reinforcing: AI adds autonomy and intelligence to networks, while networks provide infrastructure for AI. It organizes this convergence into AI for Network, Network for AI, and AI as a Service.

  • 1.3.2 Vision of AI and 6G Integration: 6G is envisioned as an integrated platform combining communication, sensing, computing, intelligence, and storage to provide customized services.
  • 1.3.1 Drivers of Integrating AI into Wireless Communication: AI can push wireless-communication limits through big-data analysis, prediction, adaptation, and modeling of complex network states.
  • 1.3.1 Drivers of Integrating AI into Wireless Communication: AI can predict congestion, device failures, and user-behavior changes, enabling advance resource allocation such as temporary base-station capacity or routing adjustments.
  • 1.3.1 Drivers of Integrating AI into Wireless Communication: AI’s adaptivity enables real-time communication-parameter adjustment as environmental and network conditions change.
  • 1.3.2 Vision of AI and 6G Integration: AI4NET uses AI to improve network performance, efficiency, and user experience, while NET4AI provides support for efficient, real-time AI training and inference and for data security and privacy.
  • 1.3.2 Vision of AI and 6G Integration: AIaaS uses network connectivity, computing, data, and models to construct distributed, efficient, energy-efficient, and secure AI services.
  • 1.3.2 Vision of AI and 6G Integration: LLMs may support 6G through contextual understanding, reasoning, generalization, model decomposition, distribution, and cross-layer network services.

2 AI for Network

AI for Network (AI4NET) applies AI to improve network performance, efficiency, and user service experience without fundamentally changing the existing network architecture. Its applications span communication optimization, resource allocation, automated operation and maintenance, and future 6G enhancements.

  • AI4NET uses AI to enhance network performance, efficiency, and user service experience.The approach focuses on optimizing specific communication functions while preserving network architectural stability.
  • AI can optimize signal modulation and demodulation, allocate network resources for load balancing, and automate operation and maintenance.These functions aim to improve transmission accuracy and efficiency while reducing operational costs.
  • Wireless AI capabilities include feature extraction, prediction, adaptation, optimization, real-time processing, correlation, and scene clustering.These capabilities can be deployed at base stations and the core network to support communication functions.
  • AI4NET applications in 6G are expected to become more abundant and sophisticated as deep learning matures and connectivity and computing infrastructures converge.The paper illustrates applications through air-interface performance enhancement and network operation and maintenance improvement.

2.1 Air Interface Performance Enhancement

AI-based methods enhance several 6G air-interface functions, including CSI feedback, OFDM reception, beam management, and wireless localization. These methods compress or reconstruct signal information, learn complex channel relationships, and reduce processing or search burdens while maintaining accuracy.

  • 2.1.1 AI-Based CSI Feedback Algorithm: AI-based CSI feedback compresses high-dimensional channel state information into compact representations to reduce feedback overhead and reconstruct channels accurately.CsiNet uses convolutional neural networks to extract, compress, and reconstruct channel features, outperforming compressed sensing in feedback accuracy and computational complexity.
  • 2.1.2 AI-Based OFDM Receiver: AI-based receivers extract features directly from signals and demodulate them in a data-driven manner, simplifying conventional receiver processing.The approach can learn channel characteristics in complex environments and compensate for signal distortion.
  • 2.1.2 AI-Based OFDM Receiver: Fully connected deep neural networks improve existing modular OFDM receivers by learning a mapping from input X to output Y and minimizing loss against true labels.Training updates network parameters through back propagation.
  • 2.1.3 AI-Based Beam Management: AI beam management selects optimal beams without scanning all beam pairs, reducing conventional scanning overhead through learned input-output mappings.Hierarchical and jointly trained codebook methods use coarse-to-fine probing and beam prediction.
  • 2.1.3 AI-Based Beam Management: 4 times faster AP selection and over tenfold faster beam selection were reported for a location-context DNN beam-alignment method.Other deep-learning beam-alignment methods reduced beam-sweeping complexity by ten times while maintaining high alignment accuracy.
  • 2.1.3 AI-Based Beam Management: Beam-management deployment still requires high-quality labels and stronger generalization across different base-station environments.Training overhead and complex propagation settings remain practical concerns.
  • 2.1.4 AI-Based Wireless Localization: AI and channel data can map channel responses to position coordinates in line-of-sight and non-line-of-sight coexistence scenarios, improving localization accuracy.This supports high-precision positioning in complex environments where indoor GPS is unavailable.

2.2 Network O&M Efficiency Improvement

AI-based methods improve 6G network operation and maintenance by predicting traffic, optimizing energy and parameters, and adapting resource orchestration to changing network conditions. These approaches target efficiency, service quality, sustainability, and reduced communication overhead.

  • AI for Network O&M: AI supports real-time network monitoring and dynamic allocation of bandwidth and power to optimize wireless-network performance.AI analyzes large data volumes and continuously adjusts network variables and resources.
  • AI-Based Traffic Prediction: Long short-term memory models outperform conventional autoregressive moving-average models for nonlinear and complex traffic prediction tasks.Traffic prediction enables advance identification of fluctuations and peak periods for dynamic allocation and scheduling.
  • AI-Based Traffic Prediction: Federated learning reduces traffic-prediction communication overhead and security risks by transmitting model parameters rather than raw traffic data.Transfer learning further accelerates training by exploiting correlations between traffic patterns in different regions.
  • AI-Based BS Energy Conservation: Threshold-based base-station shutdown can cause frequent state handoffs and service interruptions during peak periods.Reinforcement-learning approaches instead adapt sleep-mode duration and base-station states to changing loads and channel conditions.
  • AI-Based BS Energy Conservation: Integrating reinforcement learning, graph neural networks, and photovoltaic solar energy supports energy conservation and carbon-emission optimization in wireless networks.Renewable energy can also reduce dependence on conventional power grids and improve network energy efficiency.
  • AI-Based Network Parameter Optimization: AI-based parameter optimization learns nonlinear network characteristics from extensive data and enables real-time tuning for changing scenarios.The optimization scope includes base-station, threshold, mobility, and spectrum-related parameters.
  • AI-Based Resource Orchestration: AI-based resource orchestration uses communication and computing at the edge and device sides to move data processing and decisions closer to devices.Future work emphasizes adaptability, performance evaluation, and interpretability for complex and diverse scenarios.

3 Network for AI

The Network for AI stage requires 6G networks to support AI operations as a coordinated service rather than only embedding isolated AI functions. Existing 5G advances remain limited by insufficient systematicity, interpretability, access-network flexibility, and end-to-end coordination.

  • 3 Network for AI: 5G has standardized AI-related use cases and developed module-level plug-in functions under 3GPP, but these functions lack systematicity and interpretability.These limitations constrain AI capability scalability and model generalization.
  • 3 Network for AI: Cloud-native, software-defined, virtualized architectures and NWDAF provide 5G foundations for AI-enabled network operation.Their deployment remains constrained by technological maturity, security concerns, and system operation-and-maintenance complexity.
  • 3 Network for AI: The access network largely retains a siloed communication-oriented base-station design without clearly defined AI functional components.Coordination of AI service functions between the core network and access network remains unaddressed, limiting support for intelligent services.

3.1 NET4AI Architecture over 6G

NET4AI frames 6G as an integrated platform that supports AI through communication, data, computing, security, and management capabilities. Its architecture combines distributed network-native computing, enhanced control and user planes, AI-specific connectivity, and privacy-preserving intelligence.

  • Architecture vision: 6G architectures are evolving beyond connectivity toward distributed, endogenous intelligence and integrated multi-dimensional services.Industry and research proposals emphasize broader network capabilities, including task-centric operation, edge intelligence, and customized AI services.
  • NET4AI architecture: NET4AI provides AI capabilities with communication connectivity, data, computation, security, and management and orchestration support for AI-native 6G.The architecture emphasizes interactions between these five network capabilities and AI services.
  • Network functions: 6G control and user planes are extended to manage and forward services across communication, computing, data, and related resources.The control plane supports service-oriented management and policy control, while the user plane enables agile service forwarding and processing.
  • Connection for AI: AI connectivity carries both AI signaling, such as service requests and computing-power requirements, and AI data, including model inputs, outputs, and parameters.These categories reflect the communication support required for AI services and collaborative AI operations.
  • Computation for AI: Distributed network-native computing equips network elements with control, forwarding, and computing capabilities to support large-scale collaborative AI services.This endogenous computing power can respond to mobility and network changes and support applications including immersive cloud XR, holographic communication, and digital twins.
  • Security for AI: NET4AI combines distributed machine learning with privacy-protection methods to strengthen data privacy and create an efficient, secure data ecosystem.This security approach addresses the need for inherently trustworthy and self-driving protection in 6G networks.

3.2 Key enabling technologies

NET4AI organizes enabling technologies around distributed intelligence, with federated learning, multi-agent reinforcement learning, and split learning addressing privacy, adaptability, and resource constraints. These paradigms distribute computation and learning across network nodes while supporting privacy-preserving and efficient AI services.

  • NET4AI assigns enabling technologies across five capability/service planes, focusing on technologies closely related to AI needs.
  • Distributed Intelligence and Federated Learning: Distributed intelligence reduces bandwidth pressure and central-cloud computation by processing AI tasks collaboratively without uploading all raw data.
  • Distributed Intelligence and Federated Learning: Federated learning trains models locally and shares model parameters rather than raw data, with FedAvg aggregating client updates into a global model.
  • Distributed Intelligence and Federated Learning: Vertical FL supports participants with different features but the same users, enabling complementary data integration and knowledge transfer through shared model parameters.
  • Distributed Intelligence and Federated Learning: MARL supports adaptive decision-making among multiple agents, while SL partitions models between clients and servers for resource-constrained devices and privacy preservation.
  • Summary and lessons learned: FL suits privacy-sensitive AI services, MARL supports online learning in observable interactive environments, and SL supports collaborative learning under device constraints but faces convergence and heterogeneity challenges.

4 Wireless Network Large Model

Wireless network large models require domain-specific designs because wireless data are heterogeneous, privacy-sensitive, and tied to stringent inference-speed and accuracy requirements. The paper outlines layered model construction, distributed inference support, and challenges involving data, protocols, and security.

  • Comparisons Between LLM and Wireless Network Large Model: Wireless network large models cannot directly reuse traditional LLMs because wireless data span heterogeneous structured and unstructured protocol, network, and application information.
  • Comparisons Between LLM and Wireless Network Large Model: Wireless network large models require critically high inference speed and accuracy because outdated or slightly erroneous predictions can disrupt communication operations.
  • Wireless Network Large Model: A wireless network large model integrates communication-domain knowledge to provide network management, operations, fault detection, and diagnostic services.
  • Potential Approaches to Constructing a Wireless Network Large Model: The proposed construction philosophy combines an evolution route, dataset construction, and computational support, with layered models spanning universal, domain-specific, and scenario-customized functions.
  • Challenges of Wireless Network Large Model Development: Development challenges include heterogeneous training data, compatibility with existing protocol stacks, and security threats involving sensitive network information and disruptive attacks.

5 AI as a Service

AI as a Service (AIaaS) envisions 6G networks delivering end-to-end AI capabilities for distributed applications, supported by multidimensional service-quality evaluation. It targets immersive communication, industrial robotics, healthcare, and intelligent transportation.

  • AIaaS Vision: 6G AIaaS trains or infers AI within the network to provide real-time, secure, private, and energy-conscious services tailored to application scenarios.The network supplies communication, computation, datasets, and foundational models for large-scale distributed AI deployment.
  • Immersive Communication: AIaaS supports immersive communication through translation, emotion analysis, personalized suggestions, multimodal learning, digital-twin validation, and multi-device interaction.These capabilities target more realistic and efficient virtual communication while supporting rapid model training and updates.
  • Industrial Robots: AIaaS provides industrial robots with data collection, multi-agent training, model distribution, parameter exchange, remote control, cross-system learning, and digital-twin verification.The described services support collaboration among robots, terminals, and network resources across industrial environments.
  • Smart Healthcare: 6G AI supports smart healthcare by connecting medical institutions, aggregating models and case data, synchronizing information, and improving federated and group learning.The paper relates these capabilities to predictive diagnostics, treatment actions, and medical-resource allocation.
  • Intelligent Transportation: Deep learning and multisensor fusion support integrated perception and decision-making for vehicles, roadside infrastructure, and cloud platforms in intelligent transportation.The described services include AI data provision, computation offloading, environment prediction, and path planning.
  • QoAIS: QoAIS evaluates networked AI services across performance, connectivity, computation, data, security, and orchestration rather than only model-internal KPIs.Its broader scope includes model generalizability and robustness alongside network-specific service dimensions.

5.3 Key Capabilities for Supporting AIaaS

Supporting AIaaS requires coordinated data, distributed computing, and security capabilities across the 6G network. These capabilities must be integrated with scheduling and user-specific service requirements.

  • Data Support: AIaaS requires real-time data acquisition and storage from network sensing, IoT devices, and user terminals while preserving privacy and supporting task-specific models.Maintaining a direct connection between users and their data supports model training and inference services.
  • Computing Support: Distributed CPUs, GPUs, and cache resources across network nodes, base stations, devices, and third-party systems form the computational backbone of AIaaS.Software-defined networking can discover, tag, virtualize, and manage these resources.
  • Security Support: AIaaS security must protect dynamic cross-domain connections and the full service lifecycle through authentication, secure data fusion, privacy-preserving aggregation, and related safeguards.The passage identifies secure acquisition and device authentication as protections for input integrity.
  • Integrated Platform: AIaaS integration must jointly address data management, computational scheduling, security, and multidimensional user-specific service experience.The paper frames these requirements as part of an intelligent, adaptive, and efficient network architecture.
  • Operational Flow: Applying AI models to communication requires advance data collection, deployment at the appropriate network position, and sufficient inference performance to retain application value.

6 Standardization Process

Standardization efforts for AI and wireless networks span 3GPP, ITU-T, ETSI, and regional 6G initiatives. Their work covers architectures, data, interfaces, use cases, and AI/ML lifecycle alignment.

  • 3GPP: 3GPP has standardized network-intelligence and automation work through NWDAF, data-collection procedures, network-management analytics, and AI/ML model and function management.
  • 3GPP: 3GPP studies apply AI/ML to NR air-interface performance, including CSI feedback, beam management, positioning enhancement, and AI/ML lifecycle management.The targeted improvements include throughput, robustness, accuracy, reliability, and reduced overhead.
  • 3GPP: RAN3 pursued an intelligent RAN functional architecture by enhancing data collection and identifying standardization implications for NG-RAN nodes and interfaces.
  • 3GPP: Cross-3GPP coordination studies investigate AI/ML consistency across CT, RAN, and SA working groups, with a technical report planned for June 2025.
  • Other Organizations: European and international initiatives include RAN-DAF for real-time RAN data analysis, ITU-T studies on ML architectures and protocols, and ETSI ENI work on context-aware closed-loop network operations.
  • 6G Initiatives: Regional 6G programs, including China IMT-2030, have promoted research, international cooperation, and proposals for mobile-communication and AI integration.

7 Challenges of AI and communication for 6G

AI–6G integration faces challenges spanning model reliability, interpretability, network dynamics, heterogeneity, complexity, resources, energy, latency, and human collaboration. These constraints arise from demanding real-time services and dynamic distributed environments.

  • AI Reliability and Stability: AI models may lose consistent performance under unexpected anomalies, rare edge cases, or user-behavior shifts in dynamic, heterogeneous 6G systems.
  • AI Generalization and Network Dynamics: Wireless variability in user mobility, position, interference, and changing network conditions makes generalization to new data difficult.
  • Interpretability: Deep-learning black boxes hinder understanding of decisions, parameters, and potential bias, limiting explainability and trustworthiness.The paper discusses gradient visualization and t-SNE as approaches for improving interpretability.
  • Network Heterogeneity: Communication, computation, data, and model heterogeneity complicate collaborative learning and require optimized coordination across diverse edge nodes.The cited examples optimize training-node sampling and approximate decision boundaries across models.
  • Network Complexity: The scale, high-dimensional data, decentralization, and synchronization demands of 6G complicate real-time AI training, inference, deployment, and aggregation.
  • Resource Scarcity: Power, computation, bandwidth, and channel constraints can compromise federated-learning accuracy and convergence speed, requiring trade-offs in multidimensional resource allocation.
  • Sustainability: AI computation across networks, cloud centers, and edge devices carries high energy costs that may increase carbon emissions if unchecked.Pruning, quantization, neural architecture search, energy-aware models, optimized processing, and renewable energy are identified as responses.
  • Real-Time Requirements: Millisecond- or microsecond-level applications conflict with large AI models whose distributed deployment and intensive computation introduce additional latency.

8 Future Works

Future work targets adaptive, flexible, sustainable, secure, and ubiquitous AI-enabled 6G networks, while addressing computational, privacy, and robustness challenges. It also explores AI-driven innovations across emerging network technologies and lightweight LLM deployment.

  • Adaptive Learning Mechanisms: Adaptive learning must handle heterogeneous devices, changing traffic, real-time processing, scalability, and robustness across varying network states.Deep reinforcement learning is identified as one promising approach.
  • Network Architecture Design: Future architectures should dynamically adapt to AI applications by combining edge computing with central cloud resources and federated learning.Edge deployment supports latency-sensitive applications, while federated learning shares model updates rather than local data.
  • Green AI and 6G Networking: Green AI research should predict component energy consumption and dynamically adjust network infrastructure to reduce energy use.The proposed scope includes base stations and data centers.
  • AI for Ultra-Low Latency Communications: Research should develop lightweight edge AI for ultra-low-latency communication and adaptive cybersecurity systems for real-time threat detection.The cybersecurity direction uses machine learning to analyze network traffic and identify anomalies indicating possible breaches.
  • AI-Driven 6G Innovations: AI-driven innovation spans IRS, ISAC, O-RAN, NTN, near-field communication, ubiquitous computing, and LLM-driven cognitive networking.Key challenges include dynamic topology, latency, privacy, computational overhead, model compression, and robust decisions under changing conditions.

9 Conclusions

The conclusion presents AI and communications integration as central to 6G’s evolution into an infrastructure combining communication, sensing, computation, intelligence, and storage. It organizes this trajectory into AI4NET, NET4AI, and AIaaS, culminating in network-supported AI services and intelligent applications.

  • 9 Conclusions: 6G evolution toward an integrated platform combining communication, sensing, computation, intelligence, and storage depends on deep AI–communications integration.The conclusion frames this integration as foundational to the network’s development.
  • 9 Conclusions: The three-stage trajectory moves from AI optimizing networks, through networks supporting AI, to networks offering AI as a service.The stages are named AI4NET, NET4AI, and AIaaS.
  • 9 Conclusions: Future 6G service paradigms will integrate AI capabilities into application scenarios and use wireless network large models to advance intelligent, autonomous, efficient systems.The conclusion identifies these models as pivotal drivers of 6G systems.

Appendix A

Appendix A provides a list of abbreviations used in the paper.

  • Appendix A: Table A1 is titled “A list of abbreviations.”It serves as the appendix’s abbreviation reference.
  • Appendix A: The appendix identifies abbreviation entries in Table A1.The supplied passage establishes the table’s reference function.
Loading 2412.14538v4…