报告人：Prof.Benoit Champagne

报告人单位：Dept of Electrical and Computer Engineering, McGill University

报告时间：12月9日（星期二）10:20

会议地点：崇德A楼604会议室

举办单位：计算机与信息工程学院（人工智能学院）

报告人简介:

Benoit Champagne received the B.Eng. in Engineering Physics from École Polytechnique de Montréal (1983), the M.Sc. in Physics from Université de Montréal (1985), and the Ph.D. in Electrical Engineering from University of Toronto (1990). From 1990 to 1999, he was Assistant and then Associate Professor at INRS-Telecommunications, Université du Quebec, Montreal. In 1999, he joined McGill University, Montreal, where he is now a Professor within the Dept. of Electrical and Computer Engineering; he served as Associate Chai of Graduate Studies now. His research focuses on the study of advanced methods and algorithms for the processing of information bearing signals by digital means. His interests span many areas of statistical signal processing and machine learning, including detection and estimation, sensor array processing, adaptive filtering, and applications thereof to broadband wireless communications and audio processing; he has co-authored more than 350 referred publications in these areas. He has been an Associate Editor for the IEEE Signal Processing Letters, the IEEE Trans. on Signal Processing and the EURASIP Journal on Applied Signal Processing. He has also served on the Technical Committees of several international conferences in the fields of communications and signal processing.

报告摘要:

Massive MIMO is central to meeting the reliability and data-rate requirements of next-generation wireless networks, yet symbol detection remains challenging due to computational complexity and performance loss under practical, non-ideal conditions. Many existing detectors rely on assumptions such as stationary channels, white Gaussian noise, or perfect CSI, which rarely hold in real deployments. This work introduces detection methods designed to operate reliably under such realistic impairments. We first present the Preconditioned Learned Conjugate Gradient Network (PrLcgNet), which accelerates training and improves symbol error rate performance in stationary M-MIMO systems by incorporating a tailored preconditioner. We then extend this approach to time-varying channels with the Dynamic Conjugate Gradient Network (DyCoGNet), which combines meta-learning and FEC-guided self-supervision to adapt rapidly to unforeseen dynamics without labeled data. Secondly, to address scenarios with unknown or non-analytic noise distributions, we propose the Zero-Forcing Latent Space Symbol Detector (ZF-LSSD). This method pairs zero-forcing initialization with score-based generative modeling in a latent space, enabling efficient approximate maximum-likelihood detection under complex noise conditions. Finally, we introduce Attention-Based Successive Interference Cancellation (ASIC), which integrates permutation-equivariant architectures with CSI-derived priors to enhance SIC robustness under imperfect CSI. Together, these methods improve adaptability and reliability in practical massive MIMO deployments.

审核:刘学军