Research on high-precision Bluetooth and UWB positioning algorithms and performance optimization under Non-Line-of-Sight (NLOS) conditions via machine learning

  • Xufei YANG

Student thesis: MRes Thesis

Abstract

With the development of the Internet of Things (IOT) and intelligent environments, the demand for high-precision indoor positioning is increasing. However, the multipath effects and non-line-of-sight (NLOS) propagation in complex environments severely restrict the positioning accuracy. To combat this challenge, the thesis examined two major positioning technologies: Bluetooth RSSI and Ultra Wideband (UWB). This study aims to use machine learning techniques to simultaneously improve the robustness and accuracy of two types of positioning schemes in complex scenarios.

In terms of Bluetooth positioning, this thesis proposes an RSSI sequence positioning framework based on bidirectional long short-term memory network (Bi LSTM) and attention mechanism. This approach has improved the expressivity of RSSI data through time series modeling and key feature weighting. In addition, Kalman filtering (KF) and trajectory smoothing were applied to ensure the continuity and stability of the positioning results. In terms of UWB positioning, this thesis proposes a noise learning and compensation methods for explicitly modeling of NLOS errors. This method separates and corrects NLOS noise through data-driven methods to recover effective ranging signals that approximate line-of-sight (LOS), and further performs high-precision coordinate regression.

To validate the effectiveness of the proposed approaches, this thesis established a Bluetooth and UWB data acquisition platform, constructed a real dataset, and conducted experiments in typical office environments. The results show that the Bluetooth model achieves a root mean square error (RMSE) of 0.82 m in dynamic trajectory prediction, outperforming conventional regression and deep learning baselines while significantly improving robustness. Meanwhile, the UWB scheme effectively mitigates ranging offsets in NLOS scenarios, attaining an average error estimation (AEE) of approximately 0.10 m, which represents a substantial improvement in positioning accuracy compared with existing methods.

In summary, this paper realizes high precision positioning in difficult-to-resolve indoor environments powered by a data platform, a deep learning enhanced Bluetooth RSSI positioning model, and a UWB compensation framework for error modeling of NLOS. This provides valuable references for future research and implementation of multimodal fusion positioning in practice.
Date of Award15 Jan 2026
Original languageEnglish
Awarding Institution
  • University of Nottingham
SupervisorSen Yang (Supervisor) & C.F. Kwong (Supervisor)

Cite this

'