Distributed xLM Architecture for CSI Prediction Without UE Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing generative artificial intelligence (Gen AI) systems in communications networks rely on centralized architectures with separate AI engines and large language models, leading to inefficiencies and high communication overhead, particularly in managing channel state information (CSI) feedback in wireless networks.
Innovation Solution
Implementing distributed AI engines within localized language models (xLM) that enable self-service automation, allowing for faster and more efficient network optimization by generating CSI heatmaps and predicting CSI without requiring UE feedback, thus reducing complexity and increasing spectrum efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If centralized architecture with separate AI engine and LLM is used, then system structure is simplified, but communication overhead increases and efficiency decreases
Solution Approach 1:
The patent merges the AI engine and LLM into a single integrated model, eliminating the need for separate components and their intercommunication. This consolidation reduces communication overhead while maintaining the functional capabilities of both the AI engine and language model.
Solution Approach 2:
The patent segments the centralized AI system into distributed edge AI agents deployed across multiple network elements. This distribution eliminates the single-point centralized architecture, reducing communication overhead between central controller and network elements while enabling localized autonomous decision-making.
2Measurement precision
If UE feedback is required for CSI, then measurement precision is improved, but feedback overhead increases and spectrum efficiency decreases
Solution Approach 1:
The edge AI agents perform autonomous channel estimation and CSI prediction without requiring explicit feedback from UEs. The system serves itself by using uplink signals for channel estimation and generating CSI predictions independently, eliminating the need for dedicated feedback resources while maintaining accurate channel state information.
Solution Approach 2:
The patent implements implicit feedback mechanisms where the network estimates channel state from uplink signals and uses this information for downlink resource allocation. This approach replaces explicit CSI feedback with implicit channel estimation, reducing feedback overhead while maintaining measurement precision through intelligent signal processing.
3Extent of automation
If distributed edge AI agents are deployed, then automation and efficiency are improved, but device complexity increases
Solution Approach 1:
The edge AI agents are designed as universal, multi-functional components that can be deployed across different network elements (base stations, core network functions). Each agent performs multiple tasks including channel estimation, CSI prediction, resource allocation, and optimization, reducing the need for specialized components and simplifying the overall architecture despite the distributed nature.
Data Source
AI summary
Aspects of the subject disclosure may include, for example, AI engines having localized intelligence distributed throughout the air interface of a communications network. Various embodiments herein generate a CSI heat map based on data from multiple UEs. After a CSI heat map is established, Gen AI may be utilized to predict the CSI for a UE without requiring the UE to send it. Other embodiments are disclosed.