Hong Xing
The Hong Kong University of Science and Technology (Guangzhou), China
Dr. Hong XING received the B.Eng. degree from Zhejiang University, China, and the Ph.D. degree in Wireless Communications from King’s College London, U.K. in 2016. Since Jan. 2022, she has been an Assistant Professor with the IoT Thrust, The Hong Kong University of Science and Technology (HKUST, Guangzhou), China, and an Affiliate Assistant Professor with the Dept. of ECE, HKUST, HK SAR. From Mar. 2019 to Dec. 2021, she was appointed as a Research Associate Professor with Shenzhen University, China. Prior to that, from Feb. 2016 to Jan. 2019, she was a Post-Doctoral Research Fellow at King’s College London and Queen Mary University of London, UK, respectively. Her research interests include federated learning, edge inference, AI for integrated sensing and communications, and neuromorphic computing. She was the co-recipient of the Best Paper Award of IEEE International Conference on Communications (ICC) 2025. She received the Best 50 Paper recognition at IEEE Global Communications Conference (GLOBECOM) 2014. She currently serves as Editor of IEEE Wireless Communications Letters and Co-Chair of IEEE ICASSP 2027 Special Session on Reliable and Sustainable AI for Future Wireless Intelligence.
With increasing utilization of large-scale antenna arrays and high-frequency spectrum such as millimeter-wave and sub-terahertz bands, future wireless systems are envisioned to support high-resolution source localization and subsequent highly directional wireless communications. However, wireless positioning with low measurement overhead and robustness is very demanding, due to, e.g., lack of spatial resolution in single-antenna downlink (DL) users, enormous data labelling for supervised learning based localization, coherent multi-path propagation of potential sources, and missed target detection during source enumeration etc.
In this talk, we address these challenges in two prototypical scenarios. First, in far-field MISO line-of-sight (LoS) DL transmissions, we present a unified unsupervised learning framework that adopts as the input of a common neural network available pilot’s information, which is compatible with both deterministic and stochastic maximum likelihood (ML) estimation. Upon a brief preliminary on split conformal prediction (SCP) and conformal risk control (CRC), next, we introduce our recent investigation in reliable source enumeration and positioning. In near-field multi-source transmissions with coherent multipaths, we develop a CRC based source enumeration scheme that calibrates the penalty coefficient of a penalized log-likelihood estimator to ensure the expected miss-detection ratio falling below a prescribed level. We further propose a two-stage MUSIC informed MUSIC-Net that enables direct source positioning without involved NLoS-path related parameter estimation, followed by a SCP-based calibration, which constructs predictive regions with guaranteed coverage for all sources.