Gaussian Approximation-Based, Deep Neural Network-assisted Precise Indoor Localization

In this work, we develop a novel indoor localization system powered by a Gaussian Approximation (GA) based, Deep Neural Network-assisted Precise Indoor Localization (GAB-DNNPIL) algorithm and investigate its performance. We focus on an energy-efficient, precise, and adaptable signal processing-based localization perspective. This work uses a range-based factor, which works on the distance between the prior location of the user equipment (UE) and the reference point location in the lookup table (LUT). This factor scales the GA of access points (APs), eliminating spatial equivocation in localization. The GAB-DNN-PIL method does not require UEs to know their previous travel speed or direction despite using past locations. However, it accounts for temporal variations of the received signal strength indication (RSSI) by considering its Gaussian mean when computing distance. Our numerical experiments show that the proposed system has an average localization error of only 0.50 meters. Compared to other benchmarking methods, the proposed GAB-DNN-PIL method outperforms them by having an impressive 95% accuracy, making it an exceptional solution for indoor localization and a useful benchmark for the next generation of localization systems.