Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Error Mitigation for TDoA UWB Indoor Localization using Unsupervised Machine Learning

Phuong Bich Duong, Ben Van Herbruggen, Arne Broering, Adnan Shahid, Eli De Poorter

TL;DR

The paper tackles the challenge of multipath-induced errors in cm-level UWB indoor localization by introducing an unsupervised DEC-based framework for anchor selection. An autoencoder learns latent CIR features, which are clustered into multiple quality-based groups; low-quality clusters are discarded before position estimation. The key contributions are the DEC-based clustering with KL-divergence optimization, a cluster-quality scoring mechanism using first/peak path distances, and demonstrated MAE and 95th percentile reductions across full trajectories and dense multipath regions, without requiring labeled data. The approach enables practical, distributed indoor localization by reducing labeling burdens and leveraging anchor-level signal selection to improve robustness in real-world environments.

Abstract

Indoor positioning systems based on Ultra-wideband (UWB) technology are gaining recognition for their ability to provide cm-level localization accuracy. However, these systems often encounter challenges caused by dense multi-path fading, leading to positioning errors. To address this issue, in this letter, we propose a novel methodology for unsupervised anchor node selection using deep embedded clustering (DEC). Our approach uses an Auto Encoder (AE) before clustering, thereby better separating UWB features into separable clusters of UWB input signals. We furthermore investigate how to rank these clusters based on their cluster quality, allowing us to remove untrustworthy signals. Experimental results show the efficiency of our proposed method, demonstrating a significant 23.1% reduction in mean absolute error (MAE) compared to without anchor exclusion. Especially in the dense multi-path area, our algorithm achieves even more significant enhancements, reducing the MAE by 26.6% and the 95th percentile error by 49.3% compared to without anchor exclusion.

Error Mitigation for TDoA UWB Indoor Localization using Unsupervised Machine Learning

TL;DR

The paper tackles the challenge of multipath-induced errors in cm-level UWB indoor localization by introducing an unsupervised DEC-based framework for anchor selection. An autoencoder learns latent CIR features, which are clustered into multiple quality-based groups; low-quality clusters are discarded before position estimation. The key contributions are the DEC-based clustering with KL-divergence optimization, a cluster-quality scoring mechanism using first/peak path distances, and demonstrated MAE and 95th percentile reductions across full trajectories and dense multipath regions, without requiring labeled data. The approach enables practical, distributed indoor localization by reducing labeling burdens and leveraging anchor-level signal selection to improve robustness in real-world environments.

Abstract

Indoor positioning systems based on Ultra-wideband (UWB) technology are gaining recognition for their ability to provide cm-level localization accuracy. However, these systems often encounter challenges caused by dense multi-path fading, leading to positioning errors. To address this issue, in this letter, we propose a novel methodology for unsupervised anchor node selection using deep embedded clustering (DEC). Our approach uses an Auto Encoder (AE) before clustering, thereby better separating UWB features into separable clusters of UWB input signals. We furthermore investigate how to rank these clusters based on their cluster quality, allowing us to remove untrustworthy signals. Experimental results show the efficiency of our proposed method, demonstrating a significant 23.1% reduction in mean absolute error (MAE) compared to without anchor exclusion. Especially in the dense multi-path area, our algorithm achieves even more significant enhancements, reducing the MAE by 26.6% and the 95th percentile error by 49.3% compared to without anchor exclusion.
Paper Structure (18 sections, 6 equations, 3 figures, 3 tables)

This paper contains 18 sections, 6 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Data Collection and Preprocessing
  3. Methodology
  4. Input Data
  5. Deep Embedded Clustering
  6. Autoencoder
  7. Clustering Layer
  8. Training Procedure
  9. Determining Cluster Quality Scores
  10. Determining Cluster Quality
  11. Cluster Selection
  12. Evaluation of position estimate
  13. Evaluation
  14. Clustering Performance Visualization and Cluster Score
  15. Position estimate
  16. ...and 3 more sections

Figures (3)

  • Figure 1: The architecture of the cluster selection model. The input is the raw s (B1). The (B2) is used to learn the salient features of represented at code layer. Then, the code layer is used as input for clustering layer (B3). The output of clustering layer is the clusters fed to the clustering score (B4). The clustering score outputs different cluster quality levels corresponding to the $\mu_{j}$ and $\sigma _{j}$ values of each cluster. Later, cluster selection (B5) outputs only the good quality clusters.
  • Figure 2: input data clustered to 9 clusters using (a) k-means, (b) AE+k-means, (c) DEC+GMM with iteration = 1500, (d) DEC+k-means with iteration = 1500
  • Figure 3: Impact of excluding additional number of anchors from clusters on .