Semi-supervised UE Positioning via CSI Assignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine-learning-assisted techniques for user equipment (UE) positioning in wireless networks require a large amount of labelled data, making them impractical due to the dynamic nature of wireless environments. Additionally, semi-supervised techniques are limited to indoor scenarios and lack high accuracy.
Innovation Solution
The proposed solution involves a semi-supervised machine learning algorithm that uses channel state information (CSI) measurements to estimate UE positions with minimal labelled data. This algorithm assigns CSI measurements from unknown UEs to known UEs closest to them, allowing for high-accuracy positioning without frequent retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised ML algorithms are used for UE positioning, then positioning accuracy is improved, but the requirement for large amounts of labelled data makes the system impractical
Solution Approach 1:
The patent introduces Channel State Information (CSI) measurements as an intermediary that bridges the gap between limited labelled data and accurate positioning. CSI measurements serve as additional features that capture environmental characteristics, enabling the semi-supervised ML algorithm to achieve high positioning accuracy without requiring large amounts of labelled position data.
Solution Approach 2:
The patent employs a semi-supervised ML algorithm that dynamically adapts to changing wireless environments. The system can incorporate newly collected labelled data over time to retrain and update the positioning model, allowing it to maintain accuracy in dynamic conditions without requiring complete retraining from scratch.
2Quantity of substance
If semi-supervised ML techniques are used for UE positioning, then data requirement is reduced, but positioning accuracy deteriorates due to limitation to indoor scenarios
Solution Approach 1:
The patent enhances the semi-supervised ML algorithm to handle both indoor and outdoor scenarios universally. By incorporating CSI measurements that capture multipath characteristics and environmental features, the algorithm can adapt to different propagation conditions and achieve high accuracy across diverse environments, not limited to indoor settings.
3Measurement precision
If supervised ML models are frequently retrained to adapt to dynamic wireless environments, then positioning accuracy is maintained, but system complexity and data acquisition burden increase
Solution Approach 1:
The patent employs a semi-supervised ML algorithm that dynamically adapts to changing wireless environments. The system can incorporate newly collected labelled data over time to retrain and update the positioning model, allowing it to maintain accuracy in dynamic conditions without requiring complete retraining from scratch.
Solution Approach 2:
The system enables network nodes to autonomously collect CSI measurements and labelled position data, and automatically update their positioning models without requiring external intervention or complex centralized coordination. This self-service capability reduces system complexity while maintaining adaptability to environmental changes.
Data Source
AI summary
According to the present disclosure, a first dataset comprising CSI measurements associated with first UEs and actual positions of the first UEs within a coverage region of a network node is received. Then, a second dataset comprising at least one CSI measurement associated with at least one second UE whose position within the coverage region of the network node is to be estimated is received. After that, based on the first and second datasets, the at least one CSI measurement associated with the at least one second UE is assigned to one of the first UEs that appears closest to the at least one second UE. Finally, the position of the at least one second UE within the coverage region of the network node is estimated by using a semi-supervised machine learning algorithm that receives the first and second datasets and the actual position of the closest first UE.