Circuit models and AMP algorithms for future-generation wireless communication systems
No Thumbnail Available
Date
2024-10-12
Authors
Akrout, Mohamed
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Due to the significant increase in high data rate services and the demands of future wireless networks, researchers in the physical layer community are exploring new trends including i) integrating electromagnetic theory with communication theory, and ii) developing low-complexity digital signal processing (DSP) algorithms. This thesis aims to address gaps in the communication and DSP literatures.
The first part of the thesis incorporates physical limitations of antennas, such as size and mutual coupling, into circuit models for near- and far-field communications. Traditionally, constraints like antenna size and bandwidth are not included in information-theoretic performance analysis. A key finding is that mutual coupling can widen the operational bandwidth of large-scale antenna arrays, revealing a "bandwidth gain" in massive multi-input-multi-output (MIMO) technology.
The second part of the dissertation addresses recent developments in the approximate message passing (AMP) literature, where algorithms rely heavily on some assumptions (i.e., AWGN model, separable denoisers) which are not practical in many engineering applications. We extend the vector AMP approach, initially used for high-dimensional linear regression in compressive sensing, to handle arbitrary independent and identically distributed (i.i.d.) noise priors. Additionally, a bilinear generalized vector AMP algorithm is proposed, tracking the correlation matrices of the linear minimum mean square error (LMMSE) estimation. While this increases complexity, it allows the algorithm to outperform state-of-the-art solutions with discrete-valued priors. The thesis also introduces a non-separable denoiser for estimating permutation matrices, addressing the unlabeled sensing problem. Despite the computational intractability of estimating permutation matrices for even small problem sizes, the proposed unlabeled compressed sensing (UCS) approach approximates the intractable permutation denoiser using two connected assignment denoisers through a belief propagation procedure. Theoretical performance guarantees are provided through state evolution (SE) equations predicting empirical mean square error (MSE) in large systems. Simulations demonstrate the algorithm's effectiveness and superiority over existing methods.
At the intersection of DSP, antenna, and communication theories, this thesis highlights the need to revisit information theoretic concepts from an electromagnetic perspective. It emphasizes the importance of circuit-based models for their ability to define and optimize the physical characteristics and constraints of communication components in an era dominated by data-driven approaches.
Description
Keywords
signal processing, wireless communication, circuit theory, message passing