LTE and 5G RAN Traffic Control with Reusable AI Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI/ML models for mobile communication networks are not scalable and require substantial time and cost to develop, as they are tied to specific geographic areas and lack flexibility in managing increasing UE and RAN resources.
Innovation Solution
Abstract performance characteristics and traffic patterns of UEs and carriers to form AI/ML models, enabling scalable and reusable control by grouping UEs and carriers based on common characteristics and time information, allowing model application to similar environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI/ML models are built for specific geographic areas, then model accuracy for that area is improved, but scalability to new areas deteriorates
Solution Approach 1:
The patent segments the AI/ML modeling process into two distinct phases: a training phase that builds generalizable models from historical data, and an application phase that adapts these models to specific geographic areas and time periods. This segmentation allows models to be developed once and reused across multiple locations, resolving the contradiction between accuracy and scalability.
Solution Approach 2:
The patent creates abstract representations (copies) of network performance characteristics and traffic patterns that can be replicated across different geographic areas. By forming AI/ML models based on these abstracted patterns rather than area-specific details, the same models can be copied and applied to new areas, achieving both accuracy and scalability.
2Ease of operation
If custom AI/ML models are developed for each area, then local performance control is improved, but development time and cost increase
Solution Approach 1:
The patent performs preliminary action by pre-training AI/ML models using historical network data during an offline training phase. This preliminary model development eliminates the need for time-consuming custom model creation when deploying to new areas, significantly reducing development time while maintaining local performance control through the application phase.
Solution Approach 2:
The patent creates universal AI/ML models that can function across multiple geographic areas and time periods rather than area-specific models. This multi-functionality allows a single model to serve multiple locations, reducing overall development time and cost while still enabling local performance optimization through parameter adjustment in the application phase.
3Measurement precision
If comprehensive network data is collected for model training, then model accuracy is improved, but data processing complexity increases
Solution Approach 1:
The patent extracts only the essential performance characteristics and traffic pattern features from comprehensive network data during the training phase, rather than processing all raw data. This extraction of key features maintains model accuracy while significantly reducing data processing complexity when applying models to new areas.
Data Source
AI summary
Aspects of the subject disclosure may include, for example, grouping user equipment (UEs) in a radio access network (RAN) according to performance data and UE time information, forming UE groups, grouping carriers of the RAN according to traffic patterns and carrier time information, forming carrier groups, combining selected UE groups and selected carrier groups based on common time information, forming combinations, building machine learning (ML) models for each combination of the combinations, providing current UE performance information and current carrier traffic information to the ML model, and receiving, from the ML model, a network modification recommendation to improve one or more key performance indicators (KPIs) of the RAN. Other embodiments are disclosed.