Hybrid Multipath Assisted Positioning and Fingerprinting Using Transformer Models

Multipath assisted positioning exploits the spatial information contained in wireless radio signals for localizing a receiver. While such schemes are computationally very complex, they can be used in the offline phase of a fingerprinting scheme. We have presented previously such a fingerprinting scheme, where the fingerprints are channel features, and their locations are estimated with a multipath assisted positioning scheme in the offline phase. The fingerprints and their locations are used to train a mixture density network (MDN). Within this paper, we investigate the benefit of a modern transformer model compared to a MDN.