Hybrid 5G User Equipment Localization With ML Position Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing non-machine learning (non-ML) based positioning methods in 5G networks struggle to achieve the high accuracy and reliability required for new use cases, such as collision avoidance of vehicles, with regulatory bodies demanding less than 1-meter accuracy in 95% of service areas, while current ML methods face challenges in reducing signaling overhead and improving computational efficiency.
Innovation Solution
A hybrid positioning method combining non-ML and machine learning (ML) techniques, where non-ML methods provide an initial coarse location estimate, followed by a ML model, such as a DNN-based autoencoder, to refine the position, utilizing hybrid learning and minimizing labeled data requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If non-ML based positioning methods are used, then device complexity is reduced, but localization accuracy deteriorates and cannot meet the less than 1-meter accuracy requirement
Solution Approach 1:
The positioning system is segmented into two distinct stages: a non-ML based positioning stage that provides coarse location estimates, and an ML-based refinement stage that improves accuracy. This segmentation allows each component to be optimized independently, with the non-ML part handling basic positioning with low complexity and the ML part providing accuracy enhancement only when needed.
Solution Approach 2:
The non-ML based positioning method performs preliminary positioning to obtain coarse location estimates before the ML-based refinement is applied. This preliminary action provides a starting point that reduces the search space for the subsequent ML refinement, improving overall efficiency while meeting accuracy requirements.
2Measurement precision
If ML based positioning methods are used to improve accuracy, then signaling overhead and computational requirements increase
Solution Approach 1:
The ML-based refinement is applied partially - only when the non-ML based positioning accuracy is insufficient to meet the required accuracy threshold. This selective application reduces the overall computational burden and signaling overhead while maintaining the required accuracy level for critical cases.
Solution Approach 2:
The coarse location estimate from the non-ML based positioning method serves as an intermediary that bridges the gap between simple positioning and accurate positioning. This intermediary provides a preliminary result that reduces the complexity of the subsequent ML refinement by narrowing the search space and providing initial constraints.
3Measurement precision
If ML refinement is applied to all positioning cases, then localization accuracy is maximized, but signaling overhead increases
Solution Approach 1:
The system applies ML refinement partially based on the accuracy requirements of each positioning case. When the non-ML based positioning already meets the accuracy threshold, the ML refinement is skipped, thereby reducing signaling overhead. The decision to apply refinement is made selectively rather than universally.
Solution Approach 2:
The system uses feedback from the non-ML based positioning accuracy assessment to determine whether ML refinement is necessary. This feedback mechanism allows the system to adaptively adjust the positioning process, applying additional processing only when the initial results are insufficient, thus optimizing the balance between accuracy and signaling overhead.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method is provided for determining, at a first stage, a first position of a communications device, inputting, from the first stage, the first position to a machine learning model at a second stage, and determining, at the second stage, at least based on the first position from the first stage, a second position of the communications device by using the machine learning model. The first position is determined at the first stage by using one of at least a non-machine learning model and a machine learning model.