par Hannotier, Cédric 
Président du jury De Doncker, Philippe
Promoteur Quitin, François
Publication Non publié, 2025-04-24
Président du jury De Doncker, Philippe
Promoteur Quitin, François
Publication Non publié, 2025-04-24
Thèse de doctorat
| Résumé : | As cities grow larger and more complex, the demand for smarter urban management has increased. This has led to particular interest in localization technologies using wireless networks. They play a key role in applications like emergency services, traffic optimization, and the development of smart cities. In this context, time difference of arrival (TDOA)-based localization is a promising solution. It estimates the position of a target based on the difference in the arrival times of signals received by multiple anchors, without requiring the target cooperation. However, real-world deployments of TDOA systems face several challenges, such as outliers, time synchronization errors, and the complexity of tracking dynamic systems. Therefore, this thesis proposes novel solutions to these challenges, enhancing both localization accuracy and tracking robustness of TDOA-based localization systems in practical deployments.Accurately capturing phenomena occurring on real localization system is paramount for validating algorithms. Hence, chapter 2 provides a detailed overview of experimental campaigns conducted to evaluate TDOA localization in vehicular environments. These setups include multiple anchors deployed along roadways, with measurements captured under various vehicular scenarios. The experimental data serve as the foundation for developing the algorithms discussed throughout the manuscript.Then, chapter 3 introduces the fundamentals of TDOA-based localization and highlights the influence of anchor placement on the localization accuracy. It also discusses how hyperbolic geometry helps understand the localization process under noisy conditions and introduces the first major challenge encountered in real-world TDOA systems: the presence of outliers. Outliers are often caused by factors such as multipath component (MPC), non-line-of-sight (NLOS) propagation, and synchronization errors. These outliers can severely degrade localization accuracy, especially when anchor placement is suboptimal.To address this issue, chapter 4 proposes a novel localization algorithm, Intersection+DBSCAN, which is robust to outliers. Traditional pseudo-range multilateration algorithms fail to handle outliers, typically requiring assumptions about error distribution or prior knowledge of outliers. In contrast, the Intersection+DBSCAN algorithm utilizes geometric consensus from the intersections of TDOA hyperbolas with density-based spatial clustering of applications with noise (DBSCAN) to filter out outliers. This method provides more accurate localization, without requiring prior information about the outliers. Experimental results demonstrate that Intersection+DBSCAN improves localization accuracy by reducing the influence of outliers, making it a practical solution for real-world TDOA systems.Chapter 5 tackles the challenge of time synchronization in TDOA-based localization. Accurate synchronization of clocks between anchors is essential for accurate TDOA-based localization. However, practical systems often suffer from clock imperfections in the local oscillator (LO) used by the anchors. Conventional methods of synchronization, such as using a reference anchor, are limited by the rate of synchronization. To mitigate this limitation, the chapter proposes a novel carrier frequency offset (CFO)-assisted TDOA method, which leverages the relation between CFO and time drift in radio frequency (RF) transceivers. The proposed method significantly improves synchronization accuracy in both controlled and real-world experiments. Additionally, this method is extended to estimate target velocity by analyzing Doppler shifts in CFOs. While velocity estimation proves challenging due to measurement noise and slow target speeds, the method shows promise in determining the target’s direction.Finally, chapter 6 evaluates two Kalman filters (KFs) structures – conventional and cascaded – for vehicular tracking in TDOA system. The conventional approach, which integrates all system states and measurements into a single filter, provides accurate tracking but is sensitive to model inaccuracies and measurement errors, which are prevalent in real-world applications. In contrast, the cascaded KF structure divides the tracking problem into smaller, specialized filters, reducing the impact of model inaccuracies and enhancing robustness. Although the cascaded structure sacrifices some information by compartmentalizing the problem, it performs better than the conventional structure when dealing with model and measurement imperfections. This structure is further enhanced by integrating the Intersection+DBSCAN algorithm to shield the tracking process from the effects of outliers, demonstrating improved tracking accuracy, especially under moderate synchronization update rates. |
