Machine Learning for Signal Demodulation: Evaluating the QAM-16, QAM-64, and QAM-256 Constellations as Hardware Alternatives*
Session Number
1
Advisor(s)
Dr. Carol Davids, Illinois Institute of Technology
Location
A121
Discipline
Computer Science
Start Date
15-4-2026 10:15 AM
End Date
15-4-2026 11:00 AM
Abstract
This project aims to discover the effectiveness of the QAM-16, QAM-64, and QAM-256 constellations in Quadrature Amplitude Modulation (QAM) systems, with the goal of improving machine learning based classification of radio signal transmissions. Traditional communication systems rely on expensive and inflexible hardware for signal decoding. However, this research, conducted in Dr. Davids’ lab and sponsored by INdigital, investigates a software-based alternative using an AI model trained on data from constellations of various magnitudes to reduce hardware dependency and improve efficiency. Software defined radio allows for the dynamic changing of constellations in response to variable channel conditions. This unique aspect of software-defined allows us to drastically improve information transmission efficiency and speed. We’ve trained decision trees of depth 8, 12, 16, and no maximum depth on set intervals of Impulse and Additive White Gaussian Noise (AWGN) for each constellation. We’ve found that decision trees of depth 2^n where n is the number of bits transmitted perform best on each constellation. Prediction time remains reasonable, which means that these models can potentially be used in real transmissions. Ultimately, this research aims to inform the development of faster, cost-effective, and ML-driven communication systems, advancing the practicality of constellation-based signal decoding in real-world environments.
Machine Learning for Signal Demodulation: Evaluating the QAM-16, QAM-64, and QAM-256 Constellations as Hardware Alternatives*
A121
This project aims to discover the effectiveness of the QAM-16, QAM-64, and QAM-256 constellations in Quadrature Amplitude Modulation (QAM) systems, with the goal of improving machine learning based classification of radio signal transmissions. Traditional communication systems rely on expensive and inflexible hardware for signal decoding. However, this research, conducted in Dr. Davids’ lab and sponsored by INdigital, investigates a software-based alternative using an AI model trained on data from constellations of various magnitudes to reduce hardware dependency and improve efficiency. Software defined radio allows for the dynamic changing of constellations in response to variable channel conditions. This unique aspect of software-defined allows us to drastically improve information transmission efficiency and speed. We’ve trained decision trees of depth 8, 12, 16, and no maximum depth on set intervals of Impulse and Additive White Gaussian Noise (AWGN) for each constellation. We’ve found that decision trees of depth 2^n where n is the number of bits transmitted perform best on each constellation. Prediction time remains reasonable, which means that these models can potentially be used in real transmissions. Ultimately, this research aims to inform the development of faster, cost-effective, and ML-driven communication systems, advancing the practicality of constellation-based signal decoding in real-world environments.