UE AI/ML Capability Reporting for Scenario-Specific Model Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The challenge of developing a universal machine learning (ML) model for diverse deployment scenarios in the air interface is significant, as existing approaches face higher training/inference complexities and performance issues due to varying conditions, making it difficult to achieve optimal performance across different configurations.

Innovation Solution

Implementing multiple ML models for specific scenarios, each with unique identifiers, parameter lists, data sets, and registration IDs, allowing for dynamic switching and reporting capabilities to network entities, thereby enabling efficient configuration and operation of ML models based on specific conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a universal ML model is developed to cover different scenarios/configurations, then the model can be applied across multiple deployment scenarios, but the training and inference complexities increase significantly

Engineering Contradiction:

Improvemodel applicability across scenariosVSAvoidtraining and inference complexities

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent divides the universal model into multiple scenario-specific sub-models, each optimized for particular deployment conditions. Instead of training one large universal model that attempts to handle all scenarios, the system segments the functionality into smaller, specialized models (e.g., beam prediction models for different spatial domains, CSI feedback models for various channel conditions), thereby reducing the complexity of training and inference for each individual model while maintaining overall system versatility.

Inventive Principle:

Principle #1Segmentation

2Adaptability or versatility

If a universal ML model is developed to cover different scenarios/configurations, then the model can be applied across multiple deployment scenarios, but the performance may deteriorate due to inability to work well in all scenarios

Engineering Contradiction:

Improvemodel applicability across scenariosVSAvoidmodel performance across scenarios

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent implements local quality by training separate ML models tailored to specific local scenarios or configurations. Each model is optimized for its target scenario (e.g., indoor vs. outdoor beam prediction, different antenna configurations, specific channel conditions), ensuring high performance for that particular scenario rather than attempting to provide average performance across all scenarios. The system then selects the appropriate local model based on current deployment conditions, maximizing reliability for the active scenario.

Inventive Principle:

Principle #3Local quality

3Reliability

If multiple scenario-specific ML models are implemented, then the performance for each specific scenario is optimized, but the device complexity and number of models to manage increases

Engineering Contradiction:

Improvemodel performance for specific scenariosVSAvoidnumber of models to manage

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent introduces dynamic model selection mechanisms that automatically choose the appropriate scenario-specific model based on current deployment conditions. Instead of statically managing all models simultaneously, the system dynamically loads and switches between models depending on factors such as detected channel conditions, spatial domain requirements, or network configuration. This dynamic approach reduces the effective complexity by having only the relevant model(s) active at any given time, while still maintaining the capability to optimize performance across multiple scenarios.

Inventive Principle:

Principle #15Dynamics

4Adaptability or versatility

If capability information including multiple ML models with unique identifiers is reported to network entities, then the network can configure the most suitable models, but the signaling overhead and information processing complexity increases

Engineering Contradiction:

Improvenetwork configuration flexibilityVSAvoidsignaling overhead

Core Design Contradiction:

Adaptability or versatilityVSLoss of information

Solution Approach 1:

The patent extracts and reports only the essential capability information from the full model descriptions to the network entity. Instead of transmitting complete model parameters, training data, or all available model details, the UE reports condensed capability information such as model identifiers, supported scenarios, and key performance characteristics. This extraction approach allows the network to receive sufficient information for configuration decisions while minimizing signaling overhead and information processing complexity.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS20260032427A1Capability reporting for multi-model artificial intelligence/machine learning user equipment features

Publication Date: 2026.01.29 NOKIA TECHNOLOGIES OY

AI summary

The present invention provides apparatuses, methods, computer programs, computer program products and computer-readable media for capability reporting for multi-model AI/ML UE features. The method comprises generating user equipment capability information, including identifying at least one machine learning model available at the user equipment for a predetermined scenario, assigning a unique identification to each of the at least one machine learning model, associating the at least one machine learning model having the unique identification with at least one of a parameter list, a data set and a registration ID, and reporting the generated user equipment capability information to a network entity.