Artificial Intelligence and Machine Learning for Future Terahertz Wireless Networks: Design, Development, and Conceptualization

Abstract

This thesis explores the integration of artificial intelligence (AI) and machine learning (ML) techniques into future terahertz (THz) wireless communication networks, focusing on system design, optimization, and deployment. We first address one of the major impediments to THz communications: the severe impact of phase noise (PN) compounded by additive white Gaussian noise (AWGN). By developing an in-house PN model based on empirical measurements, we synthetically generated a large dataset to train several regression models. Among these, the AdaBoost Regressor (ABR) demonstrated superior performance, achieving high predictive accuracy in estimating bit error rates (BER) across diverse system configurations. Building upon the predictive capabilities of the regression model, we integrated the Differential Evolution (DE) optimization algorithm to automate waveform parameter selection for minimized BER, revealing critical insights into the nonlinear relationship between PN, bandwidth, and AWGN. Our findings demonstrated that while increasing bandwidth exacerbates AWGN, it can simultaneously mitigate the detrimental effects of PN, a relationship that cannot be captured analytically and thus necessitates a data-driven approach. In parallel, we developed a large language model (LLM)-based expert system hosted on OpenAI’s platform to accelerate not only undergraduate research on the SUNY Poly ACES v testbed, but AI predictive capabilites for future THz networks. By combining the k-nearest neighbors regressor (kNN), which predicts real-world THz link performance, with natural language interfaces, students can now access predictive tools and receive experimental advice conversationally. This approach bridges the gap between complex system modeling and intuitive experimental operation, paving the way for a new era of AI-assisted wireless systems. While this work focuses on specific waveform optimization within a single testbed, it serves as an early building block toward AI-driven, self-optimizing THz networks that will form the backbone of ultra-high-speed, adaptive wireless systems in the coming decade. Overall, this work contributes a novel, modular framework for THz waveform optimization and educational facilitation, highlighting how AI-driven methods can address core challenges in high-frequency wireless communications while fostering human-in-the-loop innovation. Our methodology paves the way for the intelligent, adaptive frameworks that will shape the evolution of THz communications, setting the stage for the fully autonomous wireless networks of the future.