Al For 6G

Guangyi Liu

October, 2023

群内每日免费分享5份+最新资料

300T网盘资源+40万份行业报告

为您的创业、职场、商业、投资、亲子、网赚、艺术、健身、心理、个人成长 …… 全面赋能！

关注公众号获取更多资料

添加微信，备
注“入群”
立刻免费领取
200套知识地图+最新研报

致终身学习者社群

行业报告/思维导图/电子书/资讯情报

收钱文案、增长黑客、产品运营、品牌企划、营销战略、办公软件、会计财务、广告设计、摄影修图、视频剪辑、直播带货、电商运营、投资理财、汽车房产、餐饮烹饪、职场经验、演讲口才、风水命理、心理思维、恋爱情趣、美妆护肤、健身瘦身、格斗搏击、漫画手绘、声乐训练、自媒体打造、效率软件工具、游戏影音……

扫码先加好友，以备不时之需

Driving forces of 6G native AI

6G native AI architecture design 2 and key features

3 6G and Foundation Model

AI has become the core driving force for a new round of industrial transformation. The automation, digitalization and intelligence of the industry require ubiquitous intelligence.

Network autonomy needs AI

Operation & maintenance

Emergency communication

Customers need AI

Voiceprint recognition

Machine translation

Business needs AI

Medical Recognition

Security monitor

Smart coverage

Customized network

Smart Navigation

Personalized Recommend.

Robot rescue

Smart Manufacturing
The integration of 6G and AI includes two aspects: "AI for Networks" and “Networks for AI"

The Driving Force of AI for Networks

Mobile communication technology faces bottlenecks, requiring urgent technological innovation and interdisciplinary integration. AI is a key solution for enhancing network performance.

Traditional communication systems face performance bottlenecks

Conflicts arise between network operation efficiency, complexity, and cost

6G poses more challenging demand metrics

Current technology falls short of meeting 6G's needs

Difficulty in estimating larger scale MIMO channels

Dense base station deployment leads to increased interference

More complex system designs lead to increased energy consumption

Complex routing in heterogeneous equipment networks

Diverse communication scenario requirements are fragmented

air interface

More accurate channel information

More precise positioning

Enhanced interference cancellation capabilities

Enhanced energy efficiency, spectral capacity

network

Improved scene adaptation speed

More balanced traffic scheduling

Faster networking interference avoidance

More refined business identification

More accurate fault location

The Driving Force of Networks for AI

ITU extends 6G scenarios to ubiquitous intelligence. AI needs to be transformed into new capabilities and services for 6G communication networks to achieve AIaaS

ITU extends 6G scenarios to ubiquitous intelligence

Get AI anytime, anywhere

Low latency AI inference/training

Support mobile AI

AI service quality assurance

AI security and privacy protection

6G network inherently provides AI services

Challenges in the Integration of 5G Networks and AI

Fulfilling 6G and AI integration demands, the universality and efficiency of existing AI design methods driven by scenario use cases, plugins, or grafts need to be improved.

Scenario-driven AI

n Design separate AI models for specific air interface and network optimization use cases

Problem: AI models have low generalization, long development cycles, and high costs

n Design different AI service processes for different third-party AI scenarios

Internet of Vehicles

Networks for AI

High-speed intelligent following

smart factory

Real-time multi-agent collaboration

XR/VR

User movement prediction

Problem: The network struggles to rapidly deploy AI services for diverse scenarios

External or grafting AI

n Add AI servers or AI-related network functions to the network, such as NWDAF

Network Management

Problem：It‘s challenging to guarantee real-time, effective, and consistent data. Completing the entire AI process involves high trial and error costs.

n Cloud AI service providers provide best-effort AI services after users submit orders

Problem：Data is only uploaded to the cloud, making it difficult to efficiently leverage the ubiquitous resources within the network, which cannot guarantee the quality and security of AI services

To achieve ubiquitous intelligence, 6G network architecture requires "four transformations"

From external to native and integrated

From best-effort to Al service quality guarantee

From multi elements decoupling to collaboration

From scenario driven to capability driven

Traffic analysis

Antenna adjustment

Movement prediction

Traffic analysis

Antenna adjustment

Movement prediction

1 Driving forces of 6G native AI

6G native AI architecture design 2 and key features

3 6G and Foundation Model

Challenge: As the three fundamental components of AI (data, algorithms and computing) have gained significance on par with network connections, the design of the corresponding architecture, interfaces, and protocols should span the entire AI lifecycle.

个

Resource layer: provide underlying resources

Network function layer: provide specific network function/ network service capabilities

Application and service layer: provide corresponding support f customers' business needs.

Data plane:
manages network data and provides
data services
Computing plane:
manages computing and provides
computing services

Intelligent plane: provides the operating environment for full life-cycle of native AI.

Unlike 5G network, new data plane, smart plane, and computing plane will be defined in 6G network, and traditional control plane and user plane are expected to be extended as well.

Traditional QoS systems primarily emphasize session and connection performance, lacking comprehensive support for diverse requirements；The QoAIS indicator system incorporates security, privacy, autonomy, and resource overhead as new evaluation dimensions to form a standardized AI service quality evaluation system.

QoAIS Guarantee Mechanism

Complete and Unified QoAls Indicator System

Standardized QoAIs Evaluation System

Appropriate Data Collection and Processing

Closed Loop Feedback

Real-Time Response

Designing a native AI protocol that integrates computing and communication is necessary to meet AI‘s connectivity and distributed computing service needs. It is achieved through three dimensions: Management Plane, Control Plane and User Plane

Control Plane: Three Modes of Deep Convergence of Computing and Communication

Computing requirements for 6G native AI

High computational efficiency

Low energy consumption and latency

Meet the differentiated QoAIS needs

CEB：Computing Execution Bearer CCB：Computing Connection Bearer CS：Computing Session = CEB + CCB

Mode 1

Mode 2

Mode 3

Coordination

xNB

Connection control

Computing control

Connection control

Computing control

Converged control

CCB

CEB

CCB

CEB

CCB

CEB

Computing Task Data Transmission&Execution

Task1 CCB CCB Task2

CCB Task3 CCB

CEB

to ensure the continuous achievement of QoAlS goals

Management Plane

Functional arrangement QoS analysis

to ensure better network-level performance

User Plane

collaborative design of computing and communication protocol

Communication and Computing Capability to be Interchangeable

The massive training data demand and high risk of trial and error for AI in the network require network digital twins to achieve on-demand data generation and reliable AI and verification

Data generation and optimization

Reduce the cost of data collection and transmission;
Solve problems such as difficulty in obtaining traditional real data;
Technology: Data Augmentation in GANs;

Pre validation of AI

Intended to complete performance pre validation without affecting network operations ;

2 . Re d u c e p o t e n t i a l r i s k s t h a t deci s i ons may l ead to , s u c h a s deteriorating network performance;

1 Driving forces of 6G native AI

6G native AI architecture design 2 and key features

3 6G and Foundation Model

As AI enters the era of general intelligence, the emergence of Foundation Models promises a profound transformation in the integration of 6G and AI

Networks for Foundation Models

Foundation Models for Networks

The network serves as a platform to support or provide Foundation Models services

• Provide rich environmental data for foundation models

Offer intent-based services to users

Achieve global collaborative control of intelligent terminals

Foundation Models will enhance mobile network services in aspects such as operations, execution, and verification

Domains	Requirements	Impact onNetworks
NetworkOperations	Multi-modal MachineLearning，LanguageUnderstanding,TextGeneration	Small
NetworkMaintenance	Non-standard DataGovernance，DataAlignment,NaturalLanguage Understanding,Code Generation	Medium
NetworkRunning	Non-standard DataGovernance,ImageGeneration，VideoGeneration	Large

Detecting Failures and Generating Solutions

Orchestrating and Scheduling Task Workflows

• Playing a Vital Role in the Validation Phase

6G native AI facilitates the training of foundation models by providing links and data services during the training process, and supports the inference process with links, computation, and model decomposition/distribution services

Features

AI training services

Foundation models training often needs high-speed fiber optic connections in data centers, making radio network deployment challenging.

Services

Collecting user and network data, preprocessing it, and managing traffic to support model training

Potential gains

6G networks process data efficiently, reducing data transmission and improving cloud AI training for models

Future issues

The required special data analysis techniques? How to efficiently schedule data in a distributed?

AI inference services

Foundation models require significant storage space and powerful AI inference chips, which cannot be met by a single base station.

With proper model segmentation, models can be deployed in wireless networks to offer AI inference services.

In 6G networks, deploying models closer to users can reduce latency

How to balance increased inference latency with reduced transmission latency in 6G networks? Are techniques like model segmentation, compression, and acceleration feasible for models? data be effectively scheduled between nodes? 16

Foundation Models for Networks face significant challenges due to the abundance of structured data and unclear commonalities among different network problems, unlike ChatGPT

Exploring in phases, beginning with the exploration of network operations ai general models Progressing from small-scale to large-scale and from offline to real-time, ultimately investigating the feasibility of unification

Network operation and maintenance data is mainly available at minute/hour intervals from a consistent source, while network operational data is more complex due to varying time intervals, standardization, and data sources, making it harder to acquire.

Data openness and standardization

Difficult data acquisition

Poor data quality

Data openness

Continuously curating and accumulating intelligent network datasets, open to the public, to build a series of innovative smart network ecosystems, and support research

standardization

Collaborate with the industry to jointly formulate new data collection standards and develop a dynamic data collection granularity scheme tailored to specific needs

Industry-wide collaborative data openness

6GANA collaborates with multiple organizations, including the Nine Heavens platform, to release four major datasets, creating an industry data sharing ecosystem to support network AI research!

通过实测和仿真，本数据集基于信道特征向量数据，探索基于AI的信道状态信息压缩反馈问题，利用AI技术实现快速准确的CSI压缩反馈能力。

CSI CompressionFeedback Dataset本数据集旨在研究基于AI的RAN切片算法，建立RAN切片模型，研究面向业务SLA保障的高效RAN切片资源配置方法，提高切片满足率与切片配置效率。

中国移动China Mobile

Intelligent RAN Slicing Dataset

本数据集对NetworkAI的计算服务场景进行建模，供用户研究任务调度的策略算法。

优秀的调度策略是保障无线通信系统传输性能的决定性条件，本数据集旨在研究基于AI的cell free调度策略。

Network AI Scheduling Technology Research Dataset

HUAWEI

Radio Resource Scheduling Dataset

Establishing a comprehensive evaluation system for AI-enabled networks to ensure overall practicality, balance, and systematicity of technical solutions.

Metrics

Requirements

Drivers

Creating a holistic metric system balancing capability, efficiency, and quality. Covering customization, generalization, universality, real-time performance, reliability, and cost-effectiveness

Evaluation Assurance System

Designing an engineering evaluation method encompassing the entire lifecycle and all elements

Including training inference, element weights, comprehensive objectives, etc.

Tools

Support

Methods

Proposing evaluation tools that cater to different stages while balancing precision and efficiency. Such as high-precision network simulation platforms, digital twin networks, and network management statistics.

Challenges

Feasibility of Quantitative Metrics

Validity of Evaluation Methods

Effectiveness of Assessment Tools

Use Cases

How to explore potential application scenarios for models and unearth their value in various contexts?

How to evaluate the value and benefits of AI enabled networks?

• How to build a unified network AI foundation model?

How to establish interpretable theoretical models for AI to ensure the effectiveness and reliability of decisions in the network?

Data/Model

Computing Capability

How to address issues such as data fragmentation and challenges in acquiring device data?

How to utilize digital twin networks to generate high-quality data and validate network AI foundation model?

How to leverage ubiquitous and mobile computing capabilities to empower foundation models such as ChatGPT and semantic models?

Architecture

• How to refine the unified architecture of Network for AI and AI for Networks to achieve efficient design of intelligent interfaces, functions, and processes in the intelligence or computing plane?

How does the architecture support the distributed training of network AI foundation model?

6GANA: A civil and open international communication platform for network AI

The road ahead for a successful 6G: Together for a shared future !

International conference MWC 2023、 uk Seminar on 6G Technology， 6GANA Annual Hexa-X、6GIC、TII！

6GANA releases 17 white papers (7 in English) to illustrate the latest global technical consensus on network AI!

Thank You！

群内每日免费分享5份+最新资料

300T网盘资源+40万份行业报告

为您的创业、职场、商业、投资、亲子、网赚、艺术、健身、心理、个人成长 …… 全面赋能！

关注公众号获取更多资料

添加微信，备
注“入群”
立刻免费领取
200套知识地图+最新研报

致终身学习者社群

行业报告/思维导图/电子书/资讯情报

刘光毅：AI for 6G-英-22页-WN11