University Events

Description: Topic: A Machine Learning-enabled Intelligent, Adaptive, and Autonomous Radio System

Location: Lester W. Cory Conference Room, Science & Engineering Building (SENG), Room 213A

Zoom Conference Link: https://umassd.zoom.us/j/99702327435
Meeting ID: 997 0232 7435 Passcode: 295971

Abstract:
Wireless communications continue to see advancements from machine learning and artificial intelligence. Typically, this is applied to individual components, attempting to add new capabilities or enhance old ones. This dissertation aims to further the research of these components, and to combine these into an intelligent radio system. This includes advancements to spectrum sensing, signal characterization, synchronization, and demodulation. The focus is on making these components autonomous, adaptive, and intelligent.

Faster Region-based Convolutional Neural Network (FRCNN) along with open world learning is leveraged in spectrum sensing and signal characterization. Multiple signals in cluttered RF environments are simultaneously localized and characterized using an object detection method. The results of the spectrum sensing algorithm are used to separate multiple signals in time domain. The separated signals are then classified by their modulation type, allowing signal characterization of multiple and cluttered RF signals. When signals of an unknown modulation type are received the network adapts to classify them by first recognizing them using novelty detection and then applying incremental learning to remember them. Next, the signal is demodulated to obtain the original information. Due to the asynchronous nature of wireless communications, and impairments from the channel, it is typical to synchronize signals prior to demodulation. However, deep learning has been shown to be able to skip this step. This dissertation evaluates the benefit of synchronization to deep learning by comparing the bit error rate of a demodulator with and without synchronization. Additionally, a deep learning-based demodulator is designed with object-detection principles to perform symbol-by-symbol demodulation. Finally, reinforcement learning is used to reconfigure the RF frontend by allocating a portion of the wireless spectrum to transmit over. This is done in a continuous space, to allow maximal spectral efficiency.

To demonstrate the ability of the system to work in real-world environments, the system is tested over-the-air using software defined radio.

Advisor(s): Dr. Ruolin Zhou, Associate Professor, Department of Electrical & Computer Engineering, UMASS Dartmouth

Committee Members: Dr. Dayalan Kasilingam, Professor and Chairperson, Department of Electrical & Computer Engineering, UMASS Dartmouth; Dr. Honggang Wang, Professor, Department of Electrical & Computer Engineering, UMASS Dartmouth; Dr. Vasu Chakravarthy, Principal Electronics Engineer, Air Force Research

NOTE: All ECE Graduate Students are ENCOURAGED to attend.
All interested parties are invited to attend. Open to the public.

*For further information, please contact Dr. Ruolin Zhou at 508.910.6922 or via email at rzhou1@umassd.edu.