This thesis addresses the study and implementation of a noise cancellation system aimed at improving the decoding of TISCODE (Transmit In Sound Code) signals, a recently developed technology that employs short sound sequences to transmit digital information to mobile devices. The goal is to make recognition more robust and to minimize the occurrence of false positives, thereby ensuring a more reliable and stable user experience in real-world scenarios, often characterized by high levels of environmental noise. The work initially focused on the main noise reduction techniques from a theoretical perspective, ranging from linear and statistical solutions (such as Wiener filtering and spectral subtraction) to more advanced approaches based on deep learning. Several filtering methods were then implemented and tested, evaluating their effectiveness on datasets of TISCODE signals contaminated with both real and synthetic noise. Particular attention was given to defining parameters that balance the number of codes considered with the guarantee of achieving zero false positives in the most conservative cases. The experimental results show that the integration of noise cancellation techniques leads to a significant improvement compared to the absence of filtering, allowing the identification of configurations capable of maintaining recognition rates close to 100%. These outcomes provide a solid foundation for future developments, which may include the adoption of more sophisticated approaches, the optimization of the methods discussed, and their efficient implementation on mobile devices, with the ultimate goal of strengthening the reliability of the TISCODE system across the various application scenarios in which it may be employed.
Questa tesi affronta lo studio e l'implementazione di un sistema di cancellazione del rumore per migliorare la decodifica dei segnali TISCODE (Transmit In Sound Code), una tecnologia di recente sviluppo che utilizza brevi sequenze sonore per trasmettere informazioni digitali a dispositivi mobili. L'obiettivo è rendere il riconoscimento più robusto e ridurre al minimo l'occorrenza di falsi positivi, così da garantire un'esperienza utente più affidabile e stabile in scenari reali, spesso caratterizzati da elevata rumorosità ambientale. Si è trattato inizialmente delle principali tecniche di riduzione da un punto di vista teorico, dalle soluzioni lineari e statistiche (come i filtri di Wiener e la sottrazione spettrale) fino ad approcci più avanzati basati su deep learning. Sono quindi stati implementati e testati diversi metodi di filtraggio, valutandone l'efficacia su dataset di segnali TISCODE contaminati da rumori reali e sintetici. Particolare attenzione è stata rivolta alla definizione di parametri utilizzati per bilanciare il numero di codici considerati con la garanzia di ottenere zero falsi positivi nei casi più conservativi. I risultati sperimentali mostrano come l'integrazione di tecniche di cancellazione del rumore porti a un netto miglioramento rispetto all'assenza di filtraggio, permettendo di individuare configurazioni in grado di mantenere percentuali di riconoscimento prossime al 100%. Questi esiti costituiscono un solido punto di partenza per futuri sviluppi, che potranno includere l'adozione di approcci più sofisticati, l'ottimizzazione dei metodi trattati e la loro implementazione efficiente su dispositivi mobili, al fine di consolidare l'affidabilità del sistema TISCODE nei molteplici scenari applicativi in cui esso potrà essere impiegato.
Tecniche di rimozione del rumore per sistemi IoT basati su trasmissione audio
ZARDO, FEDERICO
2024/2025
Abstract
This thesis addresses the study and implementation of a noise cancellation system aimed at improving the decoding of TISCODE (Transmit In Sound Code) signals, a recently developed technology that employs short sound sequences to transmit digital information to mobile devices. The goal is to make recognition more robust and to minimize the occurrence of false positives, thereby ensuring a more reliable and stable user experience in real-world scenarios, often characterized by high levels of environmental noise. The work initially focused on the main noise reduction techniques from a theoretical perspective, ranging from linear and statistical solutions (such as Wiener filtering and spectral subtraction) to more advanced approaches based on deep learning. Several filtering methods were then implemented and tested, evaluating their effectiveness on datasets of TISCODE signals contaminated with both real and synthetic noise. Particular attention was given to defining parameters that balance the number of codes considered with the guarantee of achieving zero false positives in the most conservative cases. The experimental results show that the integration of noise cancellation techniques leads to a significant improvement compared to the absence of filtering, allowing the identification of configurations capable of maintaining recognition rates close to 100%. These outcomes provide a solid foundation for future developments, which may include the adoption of more sophisticated approaches, the optimization of the methods discussed, and their efficient implementation on mobile devices, with the ultimate goal of strengthening the reliability of the TISCODE system across the various application scenarios in which it may be employed.| File | Dimensione | Formato | |
|---|---|---|---|
|
Zardo_Federico.pdf
accesso aperto
Dimensione
3.58 MB
Formato
Adobe PDF
|
3.58 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/92528