Enhanced Wi-Fi Access Point Positioning Using Hexagonal CNN With Mobile Data and Urban Information

Abstract

Wi-Fi-based localization has many advantages for personal mobile devices as it works well indoors or in urban environments while consuming much less energy than global positioning system-based localization. The position of Wi-Fi access points (APs) is critical for the accuracy of Wi-Fi-based localization. However, the AP positions are often incorrect or unavailable, making it significantly challenging to use Wi-Fibased localization for critical position-based services. In this article, we propose novel techniques that significantly enhance the Wi-Fi AP positioning by leveraging daily-collected real-world mobile data collected from six million users over a month. The proposed approach, namely Hexa U-Net, includes novel data processing by incorporating the received signal strength indicator and urban information. We also propose a novel loss function called hex-loss to train the proposed Hexa U-Net. Our evaluation results show that the proposed approach achieves 25 times higher accuracy for the Wi-Fi AP positioning compared to the simple deep neural network-based approach and 2.1 times higher accuracy compared to the state-of-the-art square gridbased convolutional neural network.