On the Use of Deep Learning Models for Interference Detection and Mitigation

Interference and noise are two of the primary factors that degrade the reliability of wireless communication systems. With the rapid expansion of satellite communication networks, the radio frequency spectrum has become increasingly congested, creating severe challenges not only for general wireless communication but also for highly sensitive applications such as radio astronomy. Weak astronomical signals, often several orders of magnitude below the thermal noise floor, are particularly vulnerable to disruption from anthropomorphic signals that overlap in frequency and time. This thesis addresses the problem of detecting and suppressing interference in modern wireless environments under a wide range of noise and interference conditions. Building on recent advancements in deep learning, which have shown strong capabilities in learning patterns directly from data without detailed prior models, we propose a framework composed of two major modules: interference detection and interference mitigation. Detection is performed using advanced deep learning architectures such as using a hybrid Convolutional Neural Network(CNN) models knows as InceptionTimePlus and MiniRocketPlus, with results compared to classical methods including matched filtering, energy detection, and FFT-based thresholding. Interference suppression is achieved through a hybrid approach that combines a two-stage convolutional autoencoder pipeline with adaptive successive interference cancellation, each optimized for different bandwidth interference. To further enhance adaptability, a recommender system is introduced that leverages the parameter estimates to dynamically select the most effective mitigation strategy for the given interference scenario. Experimental evaluations demonstrate that the proposed framework significantly reduces bit error rates across diverse interference conditions, providing a flexible and data-driven solution to interference management. For the radio astronomy use case, the methods achieve improvements in interference mitigation for the scenarios involving non-AWGN or frequency offsets. Additionally, CNN-based classifiers were also developed for classification of noise and interference in the mixed signal which can be then used for selecting accurate mitigation models. While motivated by the need to protect the integrity of radio astronomy observations in the presence of satellite-based interference, the techniques developed are broadly applicable to wireless communication systems operating in congested spectrum environments.

Name thesis:degree_name

Master of Science

Level thesis:degree_level

masters

Discipline thesis:degree_discipline

Computer Engineering

Department dc:contributor.department

Electrical and Computer Engineering

Grantor dc:publisher

Virginia Tech

Year dc:date.issued

2025

Author dc:creator

• Kothari, Hiten Prakash

Chair dc:contributor.committeechair

• Buehrer, Richard M.

Committee members dc:contributor.committeemember

• Jones, Creed Farris
• Dhillon, Harpreet Singh

• Radio Frequency Interference
• Interference Mitigation
• Autoencoders
• Signal Detection
• U-Net

Statement dc:rights

• In Copyright

Licence dc:rights.uri

• http://rightsstatements.org/vocab/InC/1.0/

Language dc:language.iso

en

Dc Identifier Other

vt_gsexam:44919

OAI identifier oai:identifier

oai:vtechworks.lib.vt.edu:10919/139855