A Generative Adversarial Network (GAN) Fingerprint Approach Over LTE

Recent advancements in communication technologies have significantly enhanced localization techniques, improving both accuracy and operating modes. Initially, localization methods relied on global navigation satellite systems, offering high accuracy but proving inefficient in Non-Line-of-Sight scenarios. Furthermore, the absence of a passive mode, where the user can be localized without explicitly requesting it, renders these methods unsuitable for applications like passive tracking systems. Fingerprinting methods, a pattern matching techniques based on signal power estimation from target devices and distance estimation from reference points, can be seen as a valid and promising alternative. However, these methods face limitations due to extensive measurement campaigns needed to establish accurate sampling systems within specific areas and the substantial amount of data required for machine learning algorithms to achieve optimal performance. This study introduces a novel fingerprinting method capable of passive operation, involving all smartphones within a designated area, suitable for both indoor and outdoor scenarios. The proposed solution leverages Generative Adversarial Networks (GANs) to augment fingerprinting datasets, enhancing machine learning models' capabilities. Additionally, the offline phase's cost-effectiveness is improved by integrating a Bayesian system as a secondary machine learning component.