Wi-Fi Fingerprint-Based Indoor Positioning using a Transformer-Autoencoder Deep Learning Framework

Abstract

Wi-Fi fingerprinting is one of the cheapest and easiest methods to find your way about indoors because it works with a variety of different types of infrastructure. But problems like signal fluctuation, noise from the environment, and different types of devices still make positioning less accurate. In this study, we propose a deep learning framework that combines a Transformer encoder with an Autoencoder-based feature extractor to solve these problems. The Transformer module captures long-range dependencies and structural patterns across high-dimensional RSS vectors, while the Autoencoder compresses features and reduces noise in a strong way. Our architecture optimizes representation learning and sequence modelling together from start to finish, which is different from how traditional deep neural networks work. Tests on a UJIIndoorLoc Wi-Fi fingerprint dataset show that our method does much better than a baseline deep neural network in terms of both classification accuracy and mean localisation error. These results show that the Transformer-Autoencoder framework works well to make indoor positioning systems more accurate and reliable.