Environmental Cross-Validation of NLOS Machine Learning Classification/Mitigation with Low-Cost UWB Positioning...

Area of Science:

• Robotics
• Signal Processing
• Machine Learning

Background:

• Radio frequency indoor positioning systems suffer from multipath propagation, degrading accuracy.
• Ultra-wideband (UWB) ranging is particularly susceptible to errors from secondary signal paths.
• Positioning algorithms using raw ranging data without accounting for multipath face significant errors.

Purpose of the Study:

• To analyze the performance of localization systems combining algorithms with machine learning.
• To classify and mitigate propagation effects like non-line-of-sight (NLOS) using ML.
• To evaluate system performance in cross-scenarios with distinct training and testing environments.

Main Methods:

• Implementing machine learning techniques for classification and mitigation of propagation effects.
• Utilizing low-cost ultra-wideband (UWB) devices for data acquisition.
• Testing the system in real-world cross-scenarios where training and testing data originate from different environments.

Main Results:

• Machine learning techniques demonstrate suitability for detecting non-line-of-sight (NLOS) ranging values.
• The proposed approach shows effectiveness even when training and testing data are from disparate scenarios.
• Performance analysis indicates improved localization accuracy through ML-based mitigation.

Conclusions:

• Machine learning is a viable tool for enhancing the robustness of UWB indoor positioning systems.
• ML-based mitigation of multipath effects, specifically NLOS, is crucial for accurate localization.
• The cross-scenario validation confirms the adaptability and practical applicability of the developed techniques.