Shenghui Song
Hong Kong University of Science and Technology, Hong Kong SAR, China
Dr. Shenghui Song is now an Associate Professor jointly appointed by the Division of Integrative Systems and Design (ISD) and the Department of Electronic and Computer Engineering (ECE) at the Hong Kong University of Science and Technology (HKUST). His research is primarily in the areas of Wireless Communications and Machine Learning with current focus on Information Theory, Integrated Sensing and Communication, Distributed Intelligence, and Semantic Communications. Dr. Song is an editorial board member of Entropy. He served as the Tutorial Program Co-Chairs of the 2022 IEEE MeditCom, and the Workshop Technical Program Chairs for the IEEE GlobeCom 2026 and ICMLCN 2026. He was named the Exemplary Reviewer for IEEE Communications Letter. Dr. Song is also interested in the research on Engineering Education and served as an Associate Editor for the IEEE Transactions on Education. He has won several teaching awards at HKUST, including the Michael G. Gale Medal for Distinguished Teaching in 2018, the Best Ten Lecturers in 2013, 2015, and 2017, the School of Engineering Distinguished Teaching Award in 2012, the Teachers I Like Award in 2013, 2015, 2016, and 2017, and the MSc (Telecom) Teaching Excellent Appreciation Award for 2020-21 and 2022-23. Dr. Song was one of the honorees of the Third Faculty Recognition at HKUST in 2021.
Holographic multiple-input multiple-output (HMIMO) is a promising paradigm for future wireless systems, enabling unprecedented spatial resolution through densely integrated electromagnetic apertures. Realizing the potential of HMIMO, however, requires advances in channel modeling, signal processing, and learning algorithms. This talk presents our recent research on HMIMO communications from three complementary perspectives. First, we investigate the fundamental limits of non-centered and non-separable channels, developing analytical characterizations of mutual information and outage performance that capture key propagation features of HMIMO systems. Second, we address the practical challenge of mutual coupling in densely packed arrays through a vector-factorization framework that jointly estimates wireless channels and coupling effects. Third, we present a score-based unsupervised learning approach for near-field HMIMO channel estimation, which learns channel distributions directly from noisy pilot observations and achieves Bayes-optimal performance without requiring labeled training data or explicit channel priors. Together, these results provide a unified perspective on how to model, estimate, and learn HMIMO channels. We conclude with key system-level insights and discuss how the proposed methodologies extend beyond HMIMO to a broad range of large-scale MIMO systems.