Real-Time Indoor Visible Light Positioning (VLP) Using Long Short Term Memory Neural Network (LSTM-NN) with Principal...

Area of Science:

• Indoor Positioning Systems
• Optical Wireless Communications
• Machine Learning for Localization

Background:

• Emerging applications like AR/VR, IoT, and AMR demand high-accuracy indoor positioning.
• Visible Light Positioning (VLP) offers a promising solution for real-time tracking.
• Received Signal Strength (RSS) based VLP is simple but prone to errors, especially at cell boundaries.

Purpose of the Study:

• To propose and demonstrate a real-time VLP system with enhanced accuracy.
• To mitigate positioning errors using a combination of LSTM-NN and PCA.
• To improve the reliability of indoor tracking for various applications.

Main Methods:

• Implemented a real-time Visible Light Positioning (VLP) system.
• Utilized a Long Short-Term Memory Neural Network (LSTM-NN) for positioning.
• Integrated Principal Component Analysis (PCA) with LSTM-NN to reduce positioning errors.

Main Results:

• Achieved an average positioning error of 5.912 cm using LSTM-NN alone.
• Reduced average positioning error to 1.806 cm with LSTM-NN and PCA, a 69.45% improvement.
• 95% of experimental data showed errors <5 cm with LSTM-NN and PCA, compared to >15 cm with LSTM-NN alone.
• Demonstrated the system's capability to predict direction and trajectory of moving receivers.

Conclusions:

• The proposed VLP system effectively enhances indoor positioning accuracy.
• The combination of LSTM-NN and PCA significantly improves precision, particularly at unit cell boundaries.
• The system shows potential for real-time tracking and trajectory prediction in demanding applications.