Employing Signal Statistics for Universal Fingerprinting Solution

Ensuring accurate, reliable, and effortless localization capabilities becomes one of the key requirements of the upcoming 6G and beyond wireless networks, while user equipment expands beyond traditional smartphones to countless Internet of Things (IoT) devices, vehicles, or drones. Location awareness becomes a necessity for smooth operation, security, and safety, while fingerprinting-based methods are able to ensure reliability and accuracy. k-Nearest Neighbors (k-NN) remains to this day one of the most popular localization algorithms, while its main drawbacks include increased complexity when operating on voluminous data, and requires exhaustive hyperparameter sweeping to find optimal performance. In this work, we propose a localization system denoted σ-MESS, which reduces the volume of the dataset, accelerates the positioning speed, and improves the positioning performance, while at the same time alleviates the requirement for finding optimal parameters for k-NN. The method is evaluated on 13 openly available indoor positioning datasets, reducing the achieved positioning error by 15% and positioning time by 87.5% on average, when compared to the k-NN with the same hyperparameters. We further compare the achieved results with the ones achieved in recently published papers outperforming numerous solutions.