Machine and Deep Learning for Indoor UWB Jammer Localization

TL;DR

The paper addresses indoor UWB jammer localization under environmental changes, where models trained in one room configuration fail in a modified layout due to domain shift. It introduces two UWB datasets capturing source and target domains and benchmarks multiple ML/DL baselines for both classification and regression tasks. To tackle domain shift, it adapts a domain-adversarial ConvNeXt autoencoder (A-CNT) with a gradient-reversal layer to align CIR-derived features across domains, achieving substantial improvements in the target environment. The results show a mean localization error of 34.67 cm in the target domain and 56% of samples within 30 cm, representing a major performance gain over non-adversarial transfer and the best source-trained baseline, thereby demonstrating the practical value of adversarial feature alignment for robust indoor jammer localization. The work provides public datasets and code to benchmark domain adaptation approaches and highlights the importance of adversarial alignment for transferring jammer localization capabilities across dynamic indoor layouts.

Abstract

Ultra-wideband (UWB) localization delivers centimeter-scale accuracy but is vulnerable to jamming attacks, creating security risks for asset tracking and intrusion detection in smart buildings. Although machine learning (ML) and deep learning (DL) methods have improved tag localization, localizing malicious jammers within a single room and across changing indoor layouts remains largely unexplored. Two novel UWB datasets, collected under original and modified room configurations, are introduced to establish comprehensive ML/DL baselines. Performance is rigorously evaluated using a variety of classification and regression metrics. On the source dataset with the collected UWB features, Random Forest achieves the highest F1-macro score of 0.95 and XGBoost achieves the lowest mean Euclidean error of 20.16 cm. However, deploying these source-trained models in the modified room layout led to severe performance degradation, with XGBoost's mean Euclidean error increasing tenfold to 207.99 cm, demonstrating significant domain shift. To mitigate this degradation, a domain-adversarial ConvNeXt autoencoder (A-CNT) is proposed that leverages a gradient-reversal layer to align CIR-derived features across domains. The A-CNT framework restores localization performance by reducing the mean Euclidean error to 34.67 cm. This represents a 77 percent improvement over non-adversarial transfer learning and an 83 percent improvement over the best baseline, restoring the fraction of samples within 30 cm to 0.56. Overall, the results demonstrate that adversarial feature alignment enables robust and transferable indoor jammer localization despite environmental changes. Code and dataset available at https://github.com/afbf4c8996f/Jammer-Loc

Figures (6)

• Figure 1: Left: Schematic layout of the experimental environment showing the positions of the four UWB receivers (RX 1–4), the transmitter (TX) mounted on a TurtleBot robot with a 1 m extension arm, and the 52 distinct jammer positions (red dots) within a 3 m×5 m area. Right: The actual indoor setup, with the sender (TX), receivers (RX), and marked jammer locations on the floor corresponding to the schematic.
• Figure 2: Fine-tuning convergence over 200 epochs: (a) loss components $L_{\mathrm{rec}}$, $L_{\mathrm{reg}}$, $L_{\mathrm{dom}}$; (b) weight schedules for $\alpha$ and $\lambda_{\mathrm{ft}}$
• Figure 3: Confusion matrices of the three deep learning models evaluated on the source dataset (52 classes).
• Figure 4: Left: K-means clustering of the 3000 hold-out ground-truth $(x,y)$ points into five spatial zones. Marker “$\times$” shows each zone centroid and area is proportional to visit count. Right: Schematic of the modified experimental layout with four UWB receivers (RX 1–4), the TurtleBot-mounted transmitter (TX) on a 1 m arm, and the 16 jammer positions (red dots) in a 3 m×5 m area.
• Figure 5: Side-by-side visualization of domain shift across taps. Left: Per-tap Wasserstein distance (EMD), with shaded regions indicating multiple intervals where $\text{EMD} \geq 0.1$. Right: Mean feature values for source and target domains, with shaded regions where the absolute difference $|\Delta\text{mean}| \geq 0.1$. A dashed green curve shows $|\Delta\text{mean}|$ per tap. Shaded areas are bounded by the actual metric values for visual fidelity.