LOS Detector Retraining Through Minority-Class Sample Collection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Inaccurate time-of-arrival (TOA) measurements due to imbalanced line-of-sight (LOS) training datasets in wireless communication systems lead to incorrect distance computations, which can be mitigated by enhancing the accuracy of LOS detection using machine-learning-based methods.
Innovation Solution
A method for user equipment (UE) to detect imbalances in the LOS training dataset and request additional signal samples for the minority class, and a network element to facilitate these measurements during a configured observation window, thereby reconfiguring the LOS detector with a more balanced dataset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine-learning-based LOS detection is used to improve TOA measurement accuracy, then measurement precision is improved, but the reliability deteriorates due to imbalanced training datasets
Solution Approach 1:
The patent changes the composition parameter of the training dataset by dynamically requesting additional signal samples for the minority class (NLOS conditions) when imbalance is detected. This parameter change in dataset composition improves both measurement precision and reliability by ensuring the machine learning model is trained on balanced representations of all propagation conditions.
2Stability of the object's composition
If additional signal samples are collected for the minority class, then dataset balance is improved, but loss of time increases due to extended observation windows
Solution Approach 1:
The patent implements a dynamic observation window that adapts its duration based on the detected level of class imbalance. When severe imbalance is detected, the observation window is extended to collect more minority class samples. When balance is adequate, the window remains short. This dynamic adjustment resolves the contradiction by making the time investment proportional to the actual need for dataset balancing.
Solution Approach 2:
The system continuously monitors the balance status of the training dataset and provides feedback to adjust the observation window duration. This feedback mechanism ensures that additional time is only spent collecting samples when actually needed, optimizing the trade-off between dataset balance and time consumption.
Data Source
AI summary
There is provided a method for a UE of a wireless communication system, the method comprising: detecting an imbalance in a line-of-sight, LOS, training dataset for training a LOS detector; determining a minority class of the LOS training dataset associated with the detected imbalance; transmitting, to a network element of the wireless communication system, a request message indicating the determined minority class and requesting measurement activation to perform one or more measurements, by the UE, to obtain one or more additional signal samples for the minority class; and receiving, from the network element, an activation message causing the UE to perform the one or more measurements during a configured observation window.