Evaluation of Free-Calibration Methods Applied to Fingerprinting-Based Radiolocalization using Machine Learning

Daniel C Cunha, Douglas T R P Silva

DOI: 10.14209/sbrt.2024.1571032714

Evento: XLII Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2024)

Keywords: Fingerprinting Indoor localization Machine learning Free-calibration methods

Abstract

One of the main problems of fingerprinting (FP)-based radiolocalization systems is the heterogeneity of mobile devices. This problem usually causes variations in the collected radio frequency (RF) signals due to a set of heterogeneity elements, such as RF chipsets, antennas, hardware drivers, and operating systems, resulting in larger location prediction errors. This work proposes a combined calibration method to correct discrepancies of the RF signal levels collected helping to reduce the prediction errors of the FP-based localization systems. The combined calibration method presented better performance than its component methods in all cases of the generalized and homogeneous scenarios and, partially, in the heterogeneous scenarios. The results showed that, in generalized scenarios, the FP-based localization system using the combined calibration method reduced the average prediction error in the range of 7 to 22%.

Download