Magnetic-Field-Based Indoor Positioning Using Temporal Convolutional Networks.

Area of Science:

• Computer Science
• Signal Processing
• Geophysics

Background:

• Traditional magnetic-field positioning relies on fingerprint databases, which are time-consuming to search.
• Deep learning models like Long Short-Term Memory (LSTM) have been applied but face challenges with training time and performance degradation.

Purpose of the Study:

• To propose a novel temporal convolutional network (TCN)-based system for magnetic-field indoor positioning.
• To enhance the accuracy and robustness of magnetic positioning by leveraging magnetic-field sequence patterns.

Main Methods:

• Magnetic-field sequences were preprocessed using coordinate transformation, smoothing filtering, and first-order differencing.
• A Temporal Convolutional Network (TCN) model was developed and trained for position prediction.
• The TCN model's performance was compared against Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models.

Main Results:

• The TCN-based system demonstrated high accuracy and robustness in magnetic-field positioning.
• The proposed method is applicable to heterogeneous smartphones.
• TCN models outperformed LSTM and GRU models in positioning tasks.

Conclusions:

• The TCN-based approach offers a more efficient and accurate solution for magnetic-field indoor positioning compared to existing methods.
• The TCN effectively extracts features from magnetic-field sequences, improving positioning performance.
• This method provides a robust and versatile solution for indoor localization using magnetic fields.