Electrical & Computer Engineering, Department of

 

First Advisor

Hamid R. Sharif-Kashani

Date of this Version

Spring 5-19-2023

Document Type

Article

Citation

Ghasemzadeh, P. (2023). A Novel Graph Neural Network-based Framework for Automatic Modulation Classification in Mobile Environments (Doctoral dissertation). University of Nebraska - Lincoln.

Comments

A DISSERTATION Presented to the Faculty of The Graduate College at the University of Nebraska In Partial Fulfillment of Requirements For the Degree of Doctor of Philosophy, Major: Engineering (Computer Engineering - Electrical and Computer Engineering), Under the Supervision of Professor Hamid R. Sharif-Kashani. Lincoln, Nebraska: May, 2023

Copyright © 2023 Pejman Ghasemzadeh

Abstract

Automatic modulation classification (AMC) refers to a signal processing procedure through which the modulation type and order of an observed signal are identified without any prior information about the communications setup. AMC has been recognized as one of the essential measures in various communications research fields such as intelligent modem design, spectrum sensing and management, and threat detection. The research literature in AMC is limited to accounting only for the noise that affects the received signal, which makes their models applicable for stationary environments. However, a more practical and real-world application of AMC can be found in mobile environments where a higher number of distorting effects is present. Hence, in this dissertation, we have developed a solution in which the distorting effects of mobile environments, e.g., multipath, Doppler shift, frequency, phase and timing offset, do not influence the process of identifying the modulation type and order classification. This solution has two major parts: recording an emulated dataset in mobile environments with real-world parameters (MIMOSigRef-SD), and developing an efficient feature-based AMC classifier. The latter itself includes two modules: feature extraction and classification. The feature extraction module runs upon a dynamic spatio-temporal graph convolutional neural network architecture, which tackles the challenges of statistical pattern recognition of received samples and assignment of constellation points. After organizing the feature space in the classification module, a support vector machine is adopted to be trained and perform classification operation. The designed robust feature extraction modules enable the developed solution to outperform other state-of-the-art AMC platforms in terms of classification accuracy and efficiency, which is an important factor for real-world implementations. We validated the performance of our developed solution in a prototyping and field-testing process in environments similar to MIMOSigRef-SD. Therefore, taking all aspects into consideration, our developed solution is deemed to be more practical and feasible for implementation in the next generations of communication systems.

Advisor: Hamid R. Sharif-Kashani

Share

COinS