Intelligent and Resilient Radio Resource Allocation for 5G and Beyond Wireless Communications

Ghasemi, Ahmad

Etd

Intelligent and Resilient Radio Resource Allocation for 5G and Beyond Wireless Communications

Public Deposited

This dissertation presents a comprehensive investigation of radio resource allocation for 5G and beyond wireless communication systems with an emphasis on resilience, intelligence, and security. It focuses on enhancing multi-input multi-output (MIMO) systems and intelligent reflective surface (IRS)-based mmWave communications by using machine learning, graph theory, graph neural networks, and low-rank approximation techniques. The study first proposes novel multi-streaming techniques for mmWave MIMO systems, emphasizing the optimum selection of subchannels and subantenna arrays via innovative beamforming methods. These methods, exploiting bipartite graph modeling and bi-clustering algorithms, significantly improve the system throughput while adhering to constant modulus and total power constraints. Second, the research considers a downlink multi-user-MIMO-millimeter wave communication system with imperfect channel state information (CSI) at the Base Station. It proposes a low-rank/Sparse Matrix Decomposition technique to mitigate sparsely distributed estimation errors, thus improving the sum-rate and establishing an effective measure for adversarial detection over the channel matrix. Further, the study explores adversarial attacks on a Graph Neural Network (GNN)-based radio resource management system in point-to-point (P2P) communications. It proposes four distinct adversarial attacks, each addressing different constraints, with the aim of manipulating the system behavior. Through an innovative eigenvalue analysis, this work also presents a mechanism to detect the presence of these attacks, highlighting significant changes in the system’s eigenvalue distribution under attack conditions. Lastly, the dissertation considers a multiuser system’s physical layer security with a Beyond Diagonal Reconfigurable Intelligent Surface (BD-RIS), in the presence of multiple eavesdroppers. It presents a hybrid approach combining Deep Reinforcement Learning (DRL) and Block Coordinate Descent (BCD) to optimize both the transmitter’s precoding and the BD-RIS beamforming matrix. The proposed approach enhances the sum-secrecy-rate and maintains a high level of service quality. Overall, this dissertation contributes to enhancing resource allocation performance measures such as spectral efficiency, energy efficiency, and fairness in next-generation wireless communication systems. It also paves the way for novel adversarial attack detection and prevention mechanisms, offering valuable insights into secure BD-RIS-based design.

Creator