Positioning Configuration IDs for Consistent ML Training and Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wireless communication systems, particularly 5G NR, lack efficient methods for improving ML training and inference accuracy and efficiency in positioning configurations, leading to suboptimal performance in identifying similar configurations during training and inference processes.
Innovation Solution
Implementing a configuration identifier (ID) system to define and record reference signal configurations, enabling consistent use of configurations during ML training and inference, and associating these IDs with neural networks for precise positioning measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ML training and inference are performed without configuration identification, then the system is simpler to implement, but the accuracy and efficiency of positioning is suboptimal
Solution Approach 1:
The patent segments the configuration management system by introducing separate configuration identifiers (ConfigIDs) to distinguish different reference signal configurations. This segmentation enables the system to track and manage multiple configurations independently, improving positioning accuracy while maintaining manageable complexity through structured organization.
Solution Approach 2:
The patent introduces configuration identifiers as intermediary elements that mediate between the ML training/inference processes and the actual reference signal configurations. These ConfigIDs serve as references that link neural network models to specific configuration sets, enabling accurate positioning without requiring direct complex interactions between all system components.
2Adaptability or versatility
If multiple reference signal configurations are used without identification, then the system has higher adaptability, but it becomes difficult to track and manage which configurations are used for training versus inference
Solution Approach 1:
The patent uses configuration identifiers as distinguishing markers (analogous to color changes) to differentiate between various reference signal configurations. Each configuration or configuration set is assigned a unique ConfigID, enabling the system to clearly distinguish and track which configurations are used for training versus inference, thus preventing information loss while maintaining high adaptability.
3Reliability
If configuration identifiers are introduced for ML training, then the training accuracy improves, but the initial system complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-assigning and storing configuration identifiers before the ML training process begins. The system pre-establishes the mapping between ConfigIDs and reference signal configurations, allowing the training process to proceed with reliable configuration tracking without adding complexity during the training execution itself.
4Measurement precision
If consistent configuration identification is enforced during training and inference, then positioning performance improves, but the system requires more rigorous configuration management
Solution Approach 1:
The patent makes configuration identifiers universal by using the same ConfigID system throughout both training and inference phases. These identifiers serve multiple functions: they track configurations during training, ensure consistency during inference, and enable efficient configuration switching. This multi-functionality improves positioning precision while actually simplifying operations by providing a unified management approach.
Data Source
AI summary
Aspects presented herein may improve the efficiency and accuracy for ML training and inference, where entities associated with the ML training/inference may be able to identify whether same/similar sets of configurations are used both in the ML training and the ML inference. In one aspect, a UE receives a configuration for a set of positioning measurements from a network entity, where the configuration includes at least one configuration ID associated with an ML data collection or an ML inference for a set of RSs. The UE performs the set of positioning measurements based on the configuration and the set of RSs. The UE stores at least one parameter for the ML data collection or load a first NN based on the at least one configuration ID.