Source-linked AI summary
The Roadmap to 6G -- AI Empowered Wireless Networks
Khaled B. Letaief, Wei Chen, Yuanming Shi, Jun Zhang, Ying-Jun Angela Zhang
TL;DR
The paper addresses how wireless networks should evolve beyond 5G to support ubiquitous AI services amid growing mobile and IoT applications. It conceptualizes AI-empowered 6G technologies and methodologies for network design and optimization, emphasizing an architecture and communication techniques that integrate AI across the network. The resulting roadmap identifies new 6G features and discusses enabling technologies while recognizing that the presented picture is partial.
Problem
Growing mobile and IoT intelligence requires wireless networks to support AI applications at edge devices with limited communication, computation, hardware, and energy resources.
Method
The article develops a forward-looking 6G roadmap covering AI-empowered architecture, AI-centric communication techniques, and AI-enabled network design and optimization.
Results
The article identifies new 6G features and discusses enabling technologies for AI-empowered architecture and communication techniques.
Takeaways & Limitations
6G is presented as an intelligent information system in which AI supports mobile AI applications and network operation across the network.
Abstract
from arXiv · showhide
The recent upsurge of diversified mobile applications, especially those supported by Artificial Intelligence (AI), is spurring heated discussions on the future evolution of wireless communications. While 5G is being deployed around the world, efforts from industry and academia have started to look beyond 5G and conceptualize 6G. We envision 6G to undergo an unprecedented transformation that will make it substantially different from the previous generations of wireless cellular systems. In particular, 6G will go beyond mobile Internet and will be required to support ubiquitous AI services from the core to the end devices of the network. Meanwhile, AI will play a critical role in designing and optimizing 6G architectures, protocols, and operations. In this article, we discuss potential technologies for 6G to enable mobile AI applications, as well as AI-enabled methodologies for 6G network design and optimization. Key trends in the evolution to 6G will also be discussed.
I. INTRODUCTION
The paper envisions 6G as a transformation from connected things to connected intelligence, supporting new AI-oriented services and using AI to shape network design and operation.
- 6G is envisioned to move wireless evolution from “connected things” to “connected intelligence” with more stringent requirements.
- 6G adds Computation Oriented Communications, Contextually Agile eMBB Communications, and Event Defined uRLLC beyond 5G service types.These services address distributed computation, context-aware adaptation, and extreme or emergency events, respectively.
- Computation Oriented Communications selects rate-latency-reliability operating points according to available resources to achieve computational accuracy.
- The article conceptualizes 6G as an intelligent information system both driven by and driving modern AI technologies.Its roadmap is based on standardization plans and 5G status, while also illustrating KPIs, service types, and a potential architecture.
- AI is expected to improve operators’ situational awareness and enable closed-loop optimization for the new 6G service types.The envisioned applications include smart cities, connected infrastructure, autonomous driving, IoT, and integrated space-air-ground networks.
II. THE ARCHITECTURE OF 6G NETWORKS
The proposed 6G architecture embraces network intelligentization, subnetwork evolution, and intelligent radio.
- The potential 6G architecture incorporates network intelligentization, subnetwork evolution, and intelligent radio.
A. From Network Softwarization to Network Intelligentization
The paper argues that 6G must progress beyond softwarization toward an AI-native architecture that can adapt to complex, heterogeneous network demands.
- Softwarization alone is considered insufficient because 6G must support communications, caching, computing, wireless power transfer, sensing, analytics, and storage.
- An AI-native architecture is proposed to make the network smart, agile, and able to learn and adapt to changing network dynamics.
- The architecture is intended to evolve into a network of subnetworks supporting flexible upgrades.
- Intelligent radio and algorithm-hardware separation are proposed to accommodate heterogeneous and upgradable hardware capabilities.
B. A Network of Subnetworks – Local vs Global Evolution
6G is envisioned as a network of subnetworks whose local components can evolve independently while maintaining coordination and network-level control.
- B. A Network of Subnetworks – Local vs Global Evolution: Local 6G subnetworks may upgrade individually, even within neighboring cells or a single cell, to adapt to environments and user demands.
- B. A Network of Subnetworks – Local vs Global Evolution: Each subnetwork should collect and analyze local wireless, user, and mobility data to upgrade itself dynamically using AI methods.
- B. A Network of Subnetworks – Local vs Global Evolution: Game and learning approaches are proposed to maintain inter-subnetwork coordination when local PHY or MAC protocols change.
- B. A Network of Subnetworks – Local vs Global Evolution: A relatively stable control plane is needed to support network-level evolution and identify suitable strategies for each subnetwork.
C. Towards Intelligent Radio (IR)
6G’s intelligent radio framework separates transceiver algorithms from hardware, allowing algorithms to adapt to diversified and upgradeable capabilities. It uses an intermediate operating system to estimate hardware properties and configure algorithms through AI.
- 6G’s hardware advances motivate separating transceiver algorithms from hardware so systems can adapt to diversified and upgradeable devices.Previous generations jointly designed hardware and algorithms, limiting agile adaptation as hardware capabilities changed.
- Intelligent radio inserts an operating system between device hardware and transceiver algorithms.The framework treats transceiver algorithms as software running over the operating system.
- The operating system estimates RF-chain, phase-shifter, ADC, and antenna capabilities, measures analog parameters, and configures transceiver algorithms using AI.
- Intelligent radio evaluates hardware contributions and bottlenecks, helping manufacturers allocate hardware costs and reducing implementation time and development costs.
III. AI-ENABLED TECHNOLOGIES FOR 6G
The paper presents AI as an indispensable tool for learning, reasoning, and decision making in highly heterogeneous 6G networks. This motivates data-driven planning and operation for dynamic network environments.
- 6G’s heterogeneity spans infrastructures, radio technologies, RF devices, computing and storage resources, and application types.The resulting systems must intelligently use communications, computing, control, and storage resources across network layers and platforms.
- Data-driven network planning and operation can support real-time adaptability to dynamic network environments.
- AI is advocated as an indispensable tool for intelligent learning, reasoning, and decision making in 6G wireless networks.
A. Big Data Analytics for 6G
The paper identifies big data analytics and AI-based optimization as central tools for managing complex 6G networks and learning communication strategies across hardware and channel effects.
- Big Data Analytics: Big data analytics comprises descriptive, diagnostic, predictive, and prescriptive analytics for understanding network performance, traffic, channels, and user perspectives.Descriptive analytics enhances operators’ and service providers’ situational awareness.
- 6G’s scale, density, and heterogeneity make conventional model-dependent optimization approaches inadequate.
- Intelligent Wireless Communication: Existing communication designs divide the end-to-end signal-processing chain into independent blocks whose simplified models do not holistically capture real-world systems.
- Intelligent Wireless Communication: An intelligent PHY layer can self-learn and self-optimize by combining sensing, data collection, AI, and domain-specific signal processing.
IV. 6G FOR AI APPLICATIONS
6G is intended to support mobile AI across end devices, network edges, and cloud data centers while addressing communication, computation, storage, power, privacy, and hardware constraints. The section discusses distributed learning, inference, and hardware-efficient wireless designs.
- Communication for Distributed Machine Learning: Wireless capacity and latency are key bottlenecks for mobile AI applications because privacy requirements keep training data on devices and distributed learning requires frequent communication.
- 6G will provide platforms that optimize communication, computation, and storage across end devices, network edges, and cloud data centers for ubiquitous mobile AI.
- Communication for Distributed Machine Learning: Communication is the key bottleneck for scaling distributed training and inference across the cloud, network edge, and end devices.
- Communication for Distributed Machine Learning: Federated learning keeps training data on devices while learning a shared global model, but limited bandwidth constrains global model aggregation.
- Hardware-Efficient Wireless Design: The proposed hybrid beamforming structure reduces phase-shifter requirements from 576 or 144 adaptive units to 30 fixed units, with 15 sufficient in a second simulation.
- Communication for Distributed Machine Learning: Mobile edge computing distributes inference across cloud, edge, and end devices to address device computation, bandwidth, storage, power, and privacy constraints.Heterogeneous computing capabilities and communication bandwidths make resource allocation challenging.
V. HARDWARE-AWARE COMMUNICATIONS FOR 6G
6G hardware-aware communications must address costly, power-hungry, and resource-constrained hardware through holistic, adaptive design. The section proposes hardware-algorithm co-design and application-aware communications as key principles.
- Hardware constraints will critically shape 6G network design as millimeter-wave and possibly Terahertz systems increase component cost and power consumption while IoT devices remain resource-constrained.The proposed paradigm jointly considers communication, sensing, and inference.
- Hardware-efficient transceivers should use fewer costly components while remaining compatible with existing signal-processing algorithms.
- Application-aware communications will be essential for IoT-like scenarios, while intelligent communications should adapt to heterogeneous hardware constraints.
A. Hardware-Algorithm Co-design
Hardware-algorithm co-design targets efficient transceivers and joint device-edge processing under demanding propagation, hardware, and IoT resource constraints. A mmWave hybrid beamforming case study illustrates the approach.
- Higher data rates require higher frequencies and large antenna arrays, increasing hardware components, cost, and power consumption.
- The proposed mmWave hybrid beamforming structure approaches fully digital beamforming performance with far fewer phase shifters than other hybrid structures.It uses a small number of phase shifters with fixed phase while preserving basic hybrid-beamforming design principles.
- IoT devices face limited computing power, energy, storage, and communication budgets, motivating joint optimization of sampling, communication, and local processing.
- Joint edge-device processing should account for local processor, storage, and channel states and integrate with edge computing.
C. Intelligent Communications for Heterogeneous Hardware Constraints
6G’s heterogeneous hardware makes conventional redesign for each transceiver architecture inefficient. Intelligent communications and transfer learning are presented as ways to adapt designs across hardware settings.
- Increasing hardware heterogeneity complicates communication protocol and algorithm design and may degrade communication efficiency.
- Transfer learning can transfer designs across analog, hybrid, and 1-bit digital mmWave transceiver architectures instead of relying on separate hand-crafted designs.
- The article presents an AI-empowered architecture and AI-centric communication techniques for 6G networks.