Indoor Positioning by Deep Q-Network in VLC Environment

Abstract

—With the recent development of the Fourth Industrial Revolution, Internet of Things technology has been widely adopted. In addition, key technologies such as big data, artificial intelligence, and wireless communication are being combined. Positioning technology that uses these technologies is essential for locating human devices in modern industries. Although the Global Positioning System can provide relatively precise positioning outdoors, its performance is limited indoors due to propagation loss. Hence, various wireless signal-based indoor positioning technologies, such as WiFi, Bluetooth, ultra-wideband, and Visible Light Communication (VLC) are being studied. In this study, positioning in indoor VLC environments is analyzed using Deep Q-Network (DQN). Each element of reinforcement learning and the agent's action and reward function are set to increase positioning accuracy. Deep Q-Network (DQN) training is then performed to derive positioning performance. The simulation results show that the proposed model attains a positioning resolution of less than 15 cm and achieves a processing speed of less than 0.03 seconds to obtain the final position in the Visible Light Communication (VLC) environment.