Self-Supervised Learning for Wireless Localization

Abstract

In this chapter, we provide an overview of several data-driven techniques for wireless localization. We initially discuss shallow dimensionality reduction (DR) approaches and investigate a supervised learning method. Subsequently, we transition into deep metric learning and then place particular emphasis on a transformer-based model and self-supervised learning. We highlight a new research direction of employing designed pretext tasks to train AI models, enabling them to learn compressed channel features useful for wireless localization. We use datasets obtained in massive multiple-input multiple-output (MIMO) systems indoors and outdoors to investigate the performance of the discussed approaches.