Due to the sheer complexity of electromagnetic spectrum usage, detecting anomalous activity in wireless spectrum is a difficult undertaking. Anomalies in the wireless spectrum can take a variety of forms, ranging from the presence of an undesirable signal in a licensed band to the absence of an expected signal, making human tagging of anomalies difficult and inefficient. There are several applications for spectrum monitoring that makes addressing this issue crucial. among which we can mention military, border control, internet of things, regulatory bodies and so on. This work focuses to propose a solution to address mentioned problems, specifically it focuses on the data collection, preprocessing, and visualization needed for an algorithm to be developed to address the following concerns: • spectrum monitoring for anomaly detection; • anomaly localization; • interpretable feature extraction; Also, this work discusses a comprehensive algorithm that, in a semi-supervised manner, can address mentioned issues with high level of accuracy and that is applicable on the entire spectrum. Furthermore, this work includes implementation of the core component of the proposed algorithm that in essence is a variational auto encoder. In this work a spectrum anomaly detector with interpretable features is discussed through a power spectral density (PSD)-based adversarial autoencoder (AAE) anomaly detector for wireless spectrum anomaly detection. The adversarial autoencoder (AAE) is a smart concept that combines the autoencoder architecture with generative adversarial network’s (GAN) adversarial loss notion. It works in a similar way to the Variational Autoencoder (VAE), except instead of KL-divergence, it utilizes adversarial loss to regularize the latent code. Also, we have developed an AAE neural network that plays as a core component of the final solution, and we have depicted the results proving that this family of neural networks are more than capable to address the forementioned concerns. The PSD data are provided by Electrosense, a crowd-sourcing initiative to collect and analyse spectrum data.
A causa della complessità nell'utilizzo dello spettro elettromagnetico, il rilevamento di attività anomale nello spettro radio è un'impresa difficile. Le anomalie nello spettro radio possono assumere una varietà di forme, che vanno dalla presenza di un segnale indesiderato in una banda autorizzata all'assenza di un segnale previsto, rendendo difficile e inefficiente la loro identificazione con attività umana. Esistono diverse applicazioni per il monitoraggio dello spettro che rendono cruciale affrontare questo problema. tra cui possiamo citare militari, controllo delle frontiere, internet delle cose, organismi di regolamentazione e così via. Questo lavoro si concentra sulla raccolta dei dati, la preelaborazione e la visualizzazione necessarie per sviluppare di un algoritmo in grado di • monitorare lo spettro per il rilevamento di anomalie; • localizzare le anomalie; • estrarre le caratteristiche interpretabili. Inoltre, questo lavoro discute un algoritmo completo che in modo semi-supervisione può affrontare i problemi sopra menzionati con un alto livello di accuratezza, che è applicabile all'intero spettro. In questo lavoro viene discusso un rivelatore di anomalie dello spettro con caratteristiche interpretabili attraverso un rivelatore di adversarial auto encoder (AAE) basato su densità spettrale di potenza (PSD). Inoltre, si è sviluppata una rete neurale AAE che funge da componente centrale della soluzione finale e si illustrano i risultati che dimostrano che questa famiglia di reti neurali è adatta ad affrontare le specifiche menzionate. I dati PSD sono forniti da Electrosense, un'iniziativa di crowd-sourcing per raccogliere e analizzare i dati dello spettro. Si utilizzare piccoli sensori radio basati su hardware economico che offre informazioni sullo spettro aggregate su un'API aperta. Questi dati vengono recuperati, prestampati e imputati correttamente per l’AAE.
Unsupervised wireless spectrum anomaly detection with adversarial auto encoder (AAE)
BAHARIYE, ARASH
2020/2021
Abstract
Due to the sheer complexity of electromagnetic spectrum usage, detecting anomalous activity in wireless spectrum is a difficult undertaking. Anomalies in the wireless spectrum can take a variety of forms, ranging from the presence of an undesirable signal in a licensed band to the absence of an expected signal, making human tagging of anomalies difficult and inefficient. There are several applications for spectrum monitoring that makes addressing this issue crucial. among which we can mention military, border control, internet of things, regulatory bodies and so on. This work focuses to propose a solution to address mentioned problems, specifically it focuses on the data collection, preprocessing, and visualization needed for an algorithm to be developed to address the following concerns: • spectrum monitoring for anomaly detection; • anomaly localization; • interpretable feature extraction; Also, this work discusses a comprehensive algorithm that, in a semi-supervised manner, can address mentioned issues with high level of accuracy and that is applicable on the entire spectrum. Furthermore, this work includes implementation of the core component of the proposed algorithm that in essence is a variational auto encoder. In this work a spectrum anomaly detector with interpretable features is discussed through a power spectral density (PSD)-based adversarial autoencoder (AAE) anomaly detector for wireless spectrum anomaly detection. The adversarial autoencoder (AAE) is a smart concept that combines the autoencoder architecture with generative adversarial network’s (GAN) adversarial loss notion. It works in a similar way to the Variational Autoencoder (VAE), except instead of KL-divergence, it utilizes adversarial loss to regularize the latent code. Also, we have developed an AAE neural network that plays as a core component of the final solution, and we have depicted the results proving that this family of neural networks are more than capable to address the forementioned concerns. The PSD data are provided by Electrosense, a crowd-sourcing initiative to collect and analyse spectrum data.File | Dimensione | Formato | |
---|---|---|---|
thesis_arash_bahariye_10596378.docx
non accessibile
Dimensione
1.72 MB
Formato
Microsoft Word XML
|
1.72 MB | Microsoft Word XML | Visualizza/Apri |
thesis_arash_bahariye_10596378.pdf
non accessibile
Dimensione
2.39 MB
Formato
Adobe PDF
|
2.39 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/10589/179417