NIaaS Framework for Interdependent Telecom Network Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing telecom platforms lack a holistic AI framework, leading to increased development time and costs, limited intelligence sharing among apps, and challenges with data heterogeneity, inter-dependencies, and model heterogeneity, which impact the effectiveness of AI services in mobile telecom systems.
Innovation Solution
A Network Intelligence-as-a-Service (NIaaS) framework that provides a unified AI platform for mobile telecom systems, addressing data heterogeneity, feature engineering, model heterogeneity, and life cycle management through services like data aggregation, feature selection, model training, and closed-loop control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If AI services are built as individual apps without a common framework, then each app can be developed independently, but development time and cost increase while intelligence sharing is limited
Solution Approach 1:
The patent implements a common AI framework that provides universal services including data processing, ML training, inference, life cycle management, and continuous performance monitoring. This framework can be leveraged by multiple AI apps, reducing development time and cost while maintaining the ability to develop apps independently through standardized interfaces and APIs.
2Adaptability or versatility
If a common AI framework is implemented, then intelligence sharing and capability binding are improved, but system complexity increases
Solution Approach 1:
The common AI framework is segmented into distinct functional modules including data processing services, ML training services, inference services, life cycle management services, and continuous performance monitoring services. Each module operates independently with well-defined interfaces, reducing overall system complexity while enabling intelligent sharing and binding across multiple apps.
3Loss of information
If heterogeneous network data is collected from multiple sources, then comprehensive analysis capability is improved, but data validation and feature engineering difficulty increase
Solution Approach 1:
The patent introduces an intermediary data processing layer that sits between heterogeneous network data sources and the AI/ML models. This layer provides standardized data validation, feature engineering, and transformation services that convert diverse data formats into unified structures, reducing the difficulty of detecting and measuring while maintaining comprehensive data collection from multiple network sources.
4Measurement precision
If AI models are trained with comprehensive features, then model accuracy is improved, but training time and computational resources increase
Solution Approach 1:
The patent implements a feature selection mechanism that identifies and prioritizes the most critical features for each specific AI/ML model training task. Rather than training all models with every available feature, the system selectively applies feature sets based on model requirements, data availability, and performance targets, reducing training time and computational resources while maintaining necessary model accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system and method for modelling, generating intelligence of and optimization of interdependent network entities in a wireless telecom network with a computer using machine learning that includes receiving network entity-specific data relating to a network function and imputing missing values in the received network entity-specific data. Groups of interdependent network entities are associated with each other to generate aggregated data to determine an impact each interdependent network entity has on the group and for predicting Key Performance Indicators (KPIs) of the interdependent network entities based on the aggregated data. From this information, optimizations to improve the KPIs are generated and tested, where control actions are implemented based on the outcome of the tested optimizations.