A Deep Learning Approach to Position Estimation from Channel Impulse Responses

Area of Science:

• Robotics and Automation
• Signal Processing
• Machine Learning

Background:

• Radio-based locating systems are crucial for industrial digitalization, enabling continuous tracking in production, manufacturing, and warehousing.
• Time difference of arrival (TDoA) systems use synchronized antennas to trilaterate mobile tag positions based on signal time-of-flight (ToF).
• Multipath propagation in industrial environments significantly hinders accurate ToF extraction, limiting traditional locating system performance.

Purpose of the Study:

• To investigate the application of deep learning (DL) for direct mobile object position estimation.
• To evaluate the effectiveness of DL in overcoming multipath propagation challenges in industrial environments.
• To compare the performance of DL-based positioning against state-of-the-art methods.

Main Methods:

• Utilizing deep learning (DL) models to process raw channel impulse responses (CIR) from radio signals.
• Directly estimating mobile object positions from CIR data, bypassing traditional ToF extraction.
• Conducting experiments in industrial environments characterized by significant multipath propagation.

Main Results:

• The proposed DL-based position estimation method demonstrates robust performance in harsh multipath environments.
• DL-based positioning significantly outperforms existing state-of-the-art approaches.
• The method also shows superior accuracy in line-of-sight (LoS) situations compared to conventional techniques.

Conclusions:

• Deep learning offers a powerful solution for accurate mobile object localization in challenging industrial settings.
• DL effectively mitigates the negative impacts of multipath propagation on radio-based locating systems.
• The DL approach provides a superior alternative to traditional TDoA methods, enhancing industrial process digitalization.