AI/ML Beam Management with Preconfigured Reference Signal Sets

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing beam management methods in NR are inefficient and resource-intensive, particularly in managing beam sets and determining appropriate reference signals, leading to increased power consumption and latency in UE operations.

Innovation Solution

Implementing artificial intelligence and machine learning (AI/ML) for model management in beam management by preconfiguring reference signal resource sets and dynamically transmitting/receiving appropriate reference signals based on UE context, allowing faster beam set measurement and indication.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of energy

If traditional beam management methods are used, then beam sets can be managed, but power consumption and latency increase

Engineering Contradiction:
Improvepower consumptionVSAvoidbeam management efficiency
Core Design Contradiction:
Loss of energyVSProductivity

Solution Approach 1:

The base station preconfigures multiple reference signal resource sets corresponding to different AI/ML models before UE operations. This preliminary preparation allows UEs to directly use these pre-configured resources for beam management without performing additional information exchange, thereby reducing both power consumption and latency in beam management operations

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If additional information exchange is performed for beam set determination, then accurate beam selection is achieved, but latency increases

Engineering Contradiction:
Improvebeam selection accuracyVSAvoidlatency
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The base station preconfigures multiple reference signal resource sets corresponding to different AI/ML models before UE operations. This preliminary preparation allows UEs to directly use these pre-configured resources for beam management without performing additional information exchange, thereby reducing both power consumption and latency in beam management operations

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

AI/ML models serve as intermediaries between the base station's beam configuration and UE beam selection. These models process beam management data and provide optimized beam set recommendations, enabling accurate beam selection while minimizing the need for additional information exchange between base station and UE

Inventive Principle:
Principle #24Intermediary (Mediator)

3Adaptability or versatility

If multiple reference signal resource sets are configured for different AI/ML models, then model management flexibility is improved, but device complexity increases

Engineering Contradiction:
Improvemodel management flexibilityVSAvoidconfiguration complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The base station implements a universal reference signal resource set configuration framework that can accommodate multiple AI/ML models with different requirements. Each resource set is designed to be multi-functional, supporting various beam management scenarios through a unified configuration approach, thereby maintaining flexibility while controlling complexity

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4611272A1Method and device for managing model in beam management using artificial intelligence and machine learning
Publication Date: 2025.09.03 KT CORP
  • EP4611272A1 patent drawingFigure 1
  • EP4611272A1 patent drawingFigure 2
  • EP4611272A1 patent drawingFigure 3

AI summary

Provided are a method and device for managing a model in beam management using artificial intelligence and machine learning. The method may include: receiving configuration information about one or more reference signal (RS) resource sets configured to respectively correspond to one or more AI/ML models or functionalities used for a certain purpose; receiving instruction information about selected one RS resource set, which is selected from the one or more RS resource sets; and measuring signal intensity or signal quality of a reference signal based on the selected one RS resource set.