Deep Learning-Based Indoor Localization Using Multi-View BLE Signal.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Accurate indoor localization is crucial for various applications.
• Existing methods often struggle with accuracy and robustness in dynamic environments.
• Bluetooth Low Energy (BLE) offers a low-power solution for proximity sensing and localization.

Purpose of the Study:

• To develop a novel Deep Neural Network (DNN)-based indoor localization system.
• To estimate the position of a BLE transmitter using received signal characteristics at multiple Anchor Points (APs).
• To achieve high localization accuracy and robustness in diverse indoor scenarios.

Main Methods:

• Utilizing received signal strength indicator (RSSI) and in-phase/quadrature (IQ) components of BLE signals.
• Simultaneously estimating Angle of Arrival (AoA) at APs using supervised learning on simulated data.
• Employing various machine learning architectures for AoA estimation with different AP subsets.
• Integrating AoA estimates into a positioning engine using the least squares (LS) algorithm for final tag positioning.

Main Results:

• Achieved a localization accuracy of 70 cm in simulated room scenarios.
• Demonstrated generalization capabilities, showing robustness to changes in room content and AP configuration.
• Some architectures exhibited the ability to distribute computational load across APs.

Conclusions:

• The proposed DNN-based method offers a promising approach for accurate and robust indoor localization using BLE signals.
• The system's adaptability to environmental and configuration changes highlights its practical applicability.
• Potential for distributed computation enhances scalability and efficiency for large-scale deployments.