Precise positioning is a strict requirement for demanding and safety-related use cases such as cooperative Intelligent Transportation Systems (ITSs) and Internet of Things (IoT), where Global Satellite Navigation System (GNSS) fails to meet the accuracy requirements or is unavailable (e.g., highway tunnels or indoor environments). Next-generation wireless networks (e.g., 5G advanced and 6G) offer a valid positioning solution in GNSS-denied scenarios, especially if considering large-bandwidth technologies. Although large-bandwidth wireless networks can enable high precision localization, complex maneuvering in highly dynamic scenarios, and harsh propagation conditions (e.g., multipath, signal blockage) can degrade the performance of the positioning system. Therefore, to guarantee the required performance in demanding scenarios, augmentation methodologies are mandatory. Bayesian filtering, paired with data fusion and data-driven approaches, is the widespread solution for augmenting wireless positioning systems. This thesis focuses on the design and experimental validation of localization algorithms for next-generation wireless networks, where the design is rooted in the theory of advanced Bayesian filtering, Bayesian inference, and model-based deep learning. The first part of the research tackles vehicular localization, starting with the design of advanced Bayesian filters for localization in highly dynamic and GNSS-denied scenarios such as highway tunnels. Then, two novel Bayesian filters are proposed; the first one is based on the model-based deep learning Deep Unfolding (DU) framework, targeting the reduction of computational load of an accurate iterative algorithm, to make it suitable for a racetrack scenario. The second filter is a new version of the Variational Bayes Multiple Model (VBMM) that can deal with non-linearities in the motion and measurement models, supporting precise localization in highway tunnels. The second part of the thesis concerns the validation of non-Bayesian and Bayesian positioning techniques in IoT applications, starting from the full design of a plug-and-play wide-band localization system for sports applications, where novel solutions for system calibration and anchor self-localization are developed. Then, the IoT research moves to indoor environments, where two new Bayesian filters are designed, both rooted in the theory of the Stein Variational Gradient Descent (SVGD). To strengthen the validity of the proposed methods, all methodologies developed in this thesis are tested with real-world data gathered during intensive experimental campaigns. Results obtained during this thesis research prove the effectiveness of the designed positioning algorithms and their validity in supporting critical use cases such as autonomous navigation and industrial IoT positioning.
Un posizionamento preciso è un requisito stringente per casi d’uso difficili e legati alla sicurezza, come i sistemi di Intelligent Transportation Systems (ITs) cooperativi oppure di Internet of Things (IoT), dove la tecnologia Global Satellite Navigation System (GNSS) non riesce a raggiungere i requisiti di accuratezza richiesti o non è disponibile (come in tunnel autostradali o canyon urbani). Le reti wireless di nuova generazione (e.g. 5G e 6G) costituiscono una valida soluzione per scenari in assenza di GNSS, in particolare se vengono considerate le tecnologie a larga banda. Nonostante le reti a larga banda possano offrire una localizzazione precisa, le manovre complesse in scenari altamente dinamici e condizioni di propagazione del segnale sfavorevoli (e.g. multi-cammino o blocco del segnale) possono degradare le prestazioni del sistema di posizionamento. Quindi, per garantire le prestazioni richieste negli scenari complessi, dei metodi di miglioramento sono necessari. I filtri Bayesiani, affiancati da metodologie di fusione dei dati e data-driven, costituiscono una soluzione diffusa per migliorare i sistemi di posizionamento con reti wireless. Questa tesi si concentra sullo sviluppo e sulla validazione sperimentale di algoritmi di localizzazione per reti wireless di nuova generazione, dove la loro ideazione ha radici nella teoria dei filtri Bayesiani avanzati, dell’inferenza Bayesiana e del model-based deep learning. La prima parte della ricerca riguarda la localizzazione veicolare, iniziando con lo sviluppo di filtri Bayesiani avanzati per la localizzazione in ambienti estremamente dinamici e in assenza del GNSS come le gallerie autostradali. In seguito, sono stati proposti due nuovi filtri Bayesiani: il primo è basato sulla metodologia di model-based deep learning chiamata Deep Unfolding (DU), con l'obiettivo di ridurre il costo computazionale di un algoritmo iterativo, per renderlo adatto per lo scenario del circuito automobilistico. Il secondo filtro è una nuova versione del Variational Bayes Multiple Model (VBMM) che può gestire le non-linearità dei modelli di moto e delle misure, supportando la localizzazione precise nei tunnel autostradali. La seconda parte della tesi riguarda la validazione di tecniche di posizionamento non-Bayesiane e Bayesiane in applicazioni IoT, iniziando dall’ideazione di un sistema di localizzazione plug-and-play a larga banda per applicazioni sportive, dove sono state sviluppate nuove soluzioni per la calibrazione del sistema e per l’auto-localizzazione delle ancore. In seguito, la ricerca sull’IoT si è spostata in ambienti indoor, dove sono stati ideati due nuovi filtri Bayesiani, basati entrambi sulla teoria dello Stein Variational Gradient Descent (SVGD). Per rafforzare la validità dei metodi proposti, tutte le metodologie sviluppate durante la tesi sono state testate su dei reali collezionati durante intensive campagne sperimentali. I risultati ottenuti durante la tesi provano l’efficacia degli algoritmi di posizionamento sviluppati e la loro validità nel supportare casi d’uso critici come la navigazione autonoma e la localizzazione IoT in ambito industriale.
High-precision positioning in next generation wireless networks
Piavanini, Marco
2025/2026
Abstract
Precise positioning is a strict requirement for demanding and safety-related use cases such as cooperative Intelligent Transportation Systems (ITSs) and Internet of Things (IoT), where Global Satellite Navigation System (GNSS) fails to meet the accuracy requirements or is unavailable (e.g., highway tunnels or indoor environments). Next-generation wireless networks (e.g., 5G advanced and 6G) offer a valid positioning solution in GNSS-denied scenarios, especially if considering large-bandwidth technologies. Although large-bandwidth wireless networks can enable high precision localization, complex maneuvering in highly dynamic scenarios, and harsh propagation conditions (e.g., multipath, signal blockage) can degrade the performance of the positioning system. Therefore, to guarantee the required performance in demanding scenarios, augmentation methodologies are mandatory. Bayesian filtering, paired with data fusion and data-driven approaches, is the widespread solution for augmenting wireless positioning systems. This thesis focuses on the design and experimental validation of localization algorithms for next-generation wireless networks, where the design is rooted in the theory of advanced Bayesian filtering, Bayesian inference, and model-based deep learning. The first part of the research tackles vehicular localization, starting with the design of advanced Bayesian filters for localization in highly dynamic and GNSS-denied scenarios such as highway tunnels. Then, two novel Bayesian filters are proposed; the first one is based on the model-based deep learning Deep Unfolding (DU) framework, targeting the reduction of computational load of an accurate iterative algorithm, to make it suitable for a racetrack scenario. The second filter is a new version of the Variational Bayes Multiple Model (VBMM) that can deal with non-linearities in the motion and measurement models, supporting precise localization in highway tunnels. The second part of the thesis concerns the validation of non-Bayesian and Bayesian positioning techniques in IoT applications, starting from the full design of a plug-and-play wide-band localization system for sports applications, where novel solutions for system calibration and anchor self-localization are developed. Then, the IoT research moves to indoor environments, where two new Bayesian filters are designed, both rooted in the theory of the Stein Variational Gradient Descent (SVGD). To strengthen the validity of the proposed methods, all methodologies developed in this thesis are tested with real-world data gathered during intensive experimental campaigns. Results obtained during this thesis research prove the effectiveness of the designed positioning algorithms and their validity in supporting critical use cases such as autonomous navigation and industrial IoT positioning.| File | Dimensione | Formato | |
|---|---|---|---|
|
PHD_MP_TESI_FINAL.pdf
accessibile in internet per tutti
Dimensione
33.25 MB
Formato
Adobe PDF
|
33.25 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/253897