Application of a temporal convolutional network algorithm fused with channel attention module for UWB indoor...

Area of Science:

• Robotics and Automation
• Signal Processing
• Machine Learning

Background:

• Ultra-wideband (UWB) technology is valuable for indoor positioning but suffers accuracy degradation in non-line-of-sight (NLOS) conditions, particularly with dynamic human movement.
• Existing neural network approaches can struggle with long-range dependencies and feature salience in complex, time-series positioning data.

Purpose of the Study:

• To develop and evaluate a novel Temporal Convolutional Network with a Channel Attention Module (TCN-CAM) for enhanced UWB indoor positioning performance.
• To address the limitations of UWB accuracy in dynamic NLOS environments through advanced deep learning techniques.

Main Methods:

• Proposed a TCN-CAM model integrating causal and dilated convolutions for capturing long-range temporal dependencies and mitigating vanishing gradients.
• Incorporated a Channel Attention Module (CAM) to adaptively focus on salient features in complex positioning scenarios.
• Conducted simulations and field experiments to validate the TCN-CAM algorithm's effectiveness.

Main Results:

• The TCN-CAM algorithm achieved high positioning accuracy and stability, with a mean error as low as 3.32 cm.
• Demonstrated significant improvements in positioning accuracy compared to LSTM-AM (76.12%), CNN-CAM (25.06%), and conventional TCN (19.42%).

Conclusions:

• The proposed TCN-CAM method substantially enhances the robustness and performance of UWB-based indoor positioning systems.
• TCN-CAM offers a superior solution for accurate and stable UWB positioning in challenging NLOS and dynamic environments.