Device-Invariant Cellular-Based Indoor Localization System Using Deep Learning

The demand for a ubiquitous and accurate indoor localization service is continuously growing. Cellular-based systems, by definition, have been shown to be a perfect selection to provide a ubiquitous localization service. The main barrier towards achieving this goal is the heterogeneity of the many different types and models of cell phones which result in variations of the measured received signal strength (RSS) even from the same location at the same time. This is particular to fingerprinting-based localization where different types of phones may be used between the system training and tracking times. The performance of the current cellular-based solutions drops significantly. In this paper, we propose a deep learning-based system that leverages cellular measurements from training devices to provide consistent, fine-grained performance across unseen tracking phones with milliwatts of power consumption. The proposed system incorporates different components to extract the device-invariant features and improve the deep model's generalization and robustness, achieving device-transparent operation. Evaluation of the proposed system in a realistic testbed using three different Android phones with different form factors and sensing capability shows that it can achieve a consistent localization accuracy. This is better than the state-of-the-art indoor cellularbased systems by at least 65%. Our experiments show the promise of this method, yielding maximum median error typically within only 0.39 meter of training and testing with the same phone.