Deep Compressed Sensing for THz UM-MIMO Channel Estimation

Access Restrictions
At the time of archiving, the student author of this thesis opted to temporarily restrict access to it. The full text of this thesis will become available to the public after the expiration of the embargo on 2025-01-25.

Abstract
Envisioned as a pivotal technology for sixth-generation (6G) and beyond, Terahertz (THz) band communications possess the potential to satisfy the escalating demand for ultra-high-speed wireless links. While ultra-massive multiple-input multiple-output (UM-MIMO) is tempting in counteracting the exceptionally high path loss at THz frequency, the channel estimation (CE) of this extensive antenna system introduces significant challenges. The success of deep learning (DL) in various fields makes it a promising candidate for THz-band CE. In this thesis, we propose a meta-learning-based channel estimator with generative adversarial network (GAN) architecture. Instead of training a GAN that learns the channel distribution only, our proposed model also learns how to infer the channel fast during the training. Our results show significant superiority over the baseline GAN estimator and traditional estimators. Our model can achieve a 3 dB lower normalized mean squared error (NMSE) than a GAN estimator with a 98% computation reduction in the online inference stage. Our designed training loss function provides a more meaningful measure of model performance during training than GAN training with adversarial loss. Moreover, the proposed model presents super-fast training convergence and a 67% reduction in pilot length compared to the baseline GAN estimator.