Performance and limits of KNN-based positioning methods for GSM networks over leaky feeder in underground tunnels

Localisation techniques have long been of major importance for safety systems and a lot of research has been conducted in the distributed computing field regarding its functionality and reliability. In the specific scenario of long yet narrow tunnels existing at CERN, localisation methods will enable a number of applications and processes to substantially reduce human intervention. In this article, we evaluate the use of fingerprinting techniques with GSM signal available throughout the LHC tunnel via a radiating cable and compare some methods to estimate the location. In the tests, 16 variants of the K-Nearest Neighbour algorithm, employing different distance weighting methods and fingerprint grouping functions, are taken into consideration and their performance is assessed with a specific rating algorithm. The existing GSM infrastructure and tunnel conditions seem to be favourable to the adoption of these fingerprinting methods. Nevertheless, significant variations in the signal have been observed which might be traced back to the presence of bulky equipment and different operational states of the accelerator. The performance limits of these fingerprinting methods are discussed for the current scenario and, based on that, an outlook for future research is given aiming at improving the system's accuracy under such challenging conditions.