Towards Scalable Uncertainty Aware DNN-based Wireless Localisation

TL;DR

This paper introduces scalable DNN methods for wireless localization that quantify uncertainty, improving reliability under varying conditions and non-line-of-sight scenarios, especially in massive MIMO environments.

Contribution

It presents variational and scalable DNN approaches to measure uncertainty, highlighting the importance of data and model uncertainty in wireless localization.

Findings

Data uncertainty captures NLOS and propagation variability.

Model uncertainty enhances overall localization reliability.

Methods are effective in massive MIMO scenarios.

Abstract

Existing deep neural network (DNN) based wireless localization approaches typically do not capture uncertainty inherent in their estimates. In this work, we propose and evaluate variational and scalable DNN approaches to measure the uncertainty as a result of changing propagation conditions and the finite number of training samples. Furthermore, we show that data uncertainty is sufficient to capture the uncertainty due to non-line-of-sight (NLOS) and, model uncertainty improves the overall reliability. To assess the robustness due to channel conditions and out-of-set regions, we evaluate the methods on challenging massive multiple-input multiple-output (MIMO) scenarios.