Canonical Correlation Analysis for Next-Generation Cellular and Underlay Communication

Author: ORCID icon
Ibrahim, Mohamed Salaheldeen Youssef Rezk, Electrical Engineering - School of Engineering and Applied Science, University of Virginia
Sidiropoulos, Nikolaos, EN-Elec/Computer Engr Dept, University of Virginia

Recent years have witnessed a tremendous amount of research in exploiting machine learning tools for handling a wide variety of problems in wireless communications. While data-driven approaches, notably deep neural networks and deep reinforcement learning, have arguably gained center-stage prominence owing to their empirical success in numerous applications when a lot of training data is available, there are in fact several problems that can markedly benefit from classical machine learning tools and latent factor analysis techniques. This dissertation studies canonical correlation analysis (CCA) and its multi-view generalization (GCCA) in the context of modern wireless communications.

One of the main contributions of this dissertation is that it provides a new and broadly useful algebraic interpretation of (G)CCA as a method that can identify a common subspace between two or more matrices, even if the uncommon components are dominant. Beyond identifiability, it develops two performance analyses which show that the common subspace can be accurately estimated via (G)CCA even in the non-ideal case where there is background noise and strong interference of the individual components in the other matrix view(s). These theoretical findings are leveraged to solve the challenging problem of reliably detecting cell-edge (weak) users in cellular wireless systems. It is shown that cell-edge user signals can be reliably decoded via (G)CCA, at very low signal to noise plus interference ratio (SINR), without knowing their channels. The proposed (G)CCA approach, can tolerate strong interference, achieves superior detection performance compared to the state-of-the-art approaches, and is computationally tractable for practical implementation.

The second part of the dissertation introduces a novel framework that enables efficient spectrum utilization by allowing coexistence between two independently operated co-channel networks. Existing methods require some level of primary-secondary coordination, cross-channel state estimation and tracking, or activity detection -- which seriously complicate their practical use. This dissertation develops a simple and practical spectrum underlay solution which enables reliable secondary communication in the presence of the primary network, without primary-secondary coordination or channel state information, under potentially strong and time-varying interference from the primary system. It is shown that the proposed approach enjoys theoretical performance guarantees which are corroborated through laboratory experimentation using software-defined radios. The proposed approach works with digital or analog modulation, it is computationally cheap, and, as a side-benefit, it provides means for accurate synchronization of the secondary user even at very low SINR.

PHD (Doctor of Philosophy)
Statistical machine learning, Canonical correlation analysis, Wireless communications, Cellular networks, Signal processing
All rights reserved (no additional license for public reuse)
Issued Date: