UWB Channel Impulse Responses for Positioning in Complex Environments: A Detailed Feature Analysis.

Area of Science:

• Signal Processing and Machine Learning for Localization

Background:

• Classical positioning methods struggle in industrial environments due to complex radio wave propagation (reflections, diffractions, absorptions).
• Existing data-driven methods leverage ultra-wideband (UWB) radio systems and channel impulse responses (CIR) to capture environmental signal properties for positioning.

Purpose of the Study:

• To develop and evaluate a feature-based localization approach using machine learning on CIR signals.
• To assess the approach's effectiveness in complex environments without requiring precise time synchronization.

Main Methods:

• Investigated various signal features derived from CIR, based on complex propagation models.
• Qualitatively assessed features based on spatial relationships and their contribution to position accuracy.
• Quantitatively evaluated features using hierarchical classification on datasets from environments with varying complexity.

Main Results:

• Features derived from CIR demonstrated a clear relationship with the environment, improving positional accuracy in complex settings.
• Achieved classification accuracies exceeding 90% for region sizes as small as 0.1 m².
• Successfully distinguished between different screwing processes on a car door using CIR measures, adaptable to environmental changes via retraining.

Conclusions:

• The proposed feature-based machine learning approach enables highly accurate localization and classification in complex environments.
• The method requires minimal infrastructure (1-2 tags) and is adaptable to new or changing environments without system calibration or reference installations.