Deep Learning for 3D Indoor Localization: A CNN Approach with 802.11az Fingerprinting

Vehicle Indoor location tracking is a crucial technology for various applications, including intelligent buildings, robotics, and the Internet of Things (IoT). This research presents a deep learning method that employs Convolutional Neural Networks (CNNs) for Three-Dimensional (3D) indoor positioning, using 802.11az Wi-Fi fingerprinting. The proposed technique utilizes Channel Impulse Response (CIR) fingerprints generated through ray-tracing methods to gather detailed features of the wireless channel. In our method, CIR fingerprints are collected from multiple Access Points (APs) and enhanced through techniques like data augmentation, outlier removal, and normalization to boost model generalization. A sophisticated CNN architecture is designed to extract spatial information from Wi-Fi fingerprints, establishing strong connections between the received signals and their 3D location coordinates. The model is trained on both synthetic and real-world datasets and evaluated using cross-validation techniques. Our experimental results indicate that positioning based on CNNs significantly outperforms traditional machine learning approaches. Specifically, increasing Wi-Fi bandwidth (from CBW20 MHz to CBW180 MHz) and implementing MIMO configurations reduce positioning errors from 2.5 meters to 0.6 meters, achieving sub-meter localization accuracy in over 90% of cases. The analysis of the Cumulative Distribution Function (CDF) further corroborates that enhanced bandwidth and multiple antennas improve localization accuracy. Additionally, a comparative evaluation of 1×1 and 4×4 MIMO configurations highlights the performance gains achieved through spatial diversity. In conclusion, the proposed CNN-based system demonstrates that deep learning can significantly enhance Wi-Fi fingerprinting for indoor positioning, making it a viable solution for accurate localization in complex indoor environments. Future research may focus on optimizing neural architectures, facilitating real-time adjustments, and integrating beam-forming techniques to further elevate positioning efficacy.