Adaptive and efficient resource allocation for cognitive radio networks
Date
2026
Authors
Shaghluf, Nagwa
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The increasing demand for wireless connectivity, massive device access, and data-intensive applications has intensified the need for efficient and intelligent spectrum utilization in modern wireless networks. Although radio spectrum is a limited resource, many licensed bands remain underutilized due to static allocation policies. Cognitive Radio Networks (CRNs) address this challenge by enabling Secondary Users (SUs) to dynamically access underutilized spectrum while protecting Primary Users (PUs) from harmful interference. However, efficient resource allocation in CRNs remains challenging due to sensing uncertainty, dynamic channel conditions, interference coupling, limited Channel State Information (CSI), and the high computational complexity of conventional optimization methods.
This dissertation develops an intelligent and adaptive resource allocation framework for CRNs by improving spectrum awareness through predictive spectrum sensing and progressively integrating Multi-Agent Deep Reinforcement Learning (MADRL) and Non-Orthogonal Multiple Access (NOMA). The first contribution is a Predictive-Cooperative Spectrum Sensing (PCSS) framework that combines spectrum prediction with Cooperative Spectrum Sensing (CSS). Unlike conventional sensing schemes that rely only on instantaneous sensing decisions, the proposed PCSS approach uses historical cooperative sensing decisions to predict future PU channel availability before sensing is performed. This reduces unnecessary sensing of busy channels and improves spectrum awareness. The sensing time and fusion decision threshold are jointly optimized to characterize the trade-off between Spectrum Efficiency (SE) and Energy Efficiency (EE). Simulation results show that PCSS improves SE and achieves a better SE–EE trade-off compared with conventional sensing, local prediction, and cooperative spectrum prediction schemes.
The second contribution is a MADRL-based resource allocation framework for Cognitive Device-to-Device (C-D2D) communication underlaying cellular networks. The joint Resource Block (RB) assignment and power control problem is formulated as a Mixed-Integer Nonlinear Programming (MINLP) problem due to binary RB allocation, nonlinear SINR expressions, and interference-coupled constraints. To address this complexity, a Proximal Policy Optimization (PPO)-based MADRL approach is developed, where each C-D2D pair acts as an autonomous agent. Both Centralized Training with Decentralized Execution (CTDE) and Decentralized Training with Decentralized Execution (DTDE) architectures are investigated under perfect and imperfect CSI. Simulation results demonstrate that the proposed PPO-based MADRL framework improves C-D2D sum-rate, scalability, and robustness compared with random allocation, random search, and other DRL baselines.
The third contribution extends the framework to NOMA-enabled C-D2D networks. In this model, each C-D2D transmitter serves two receivers using power-domain NOMA while reusing cellular uplink RBs under CR protection constraints. This introduces additional challenges due to inter-group interference, intra-group NOMA interference, SIC decoding requirements, and fairness among C-D2D groups. A PPO-based MADRL framework is proposed for joint RB reuse and power allocation under constrained sum-rate and fairness-aware objectives. Different observation models, including local, partial, and full CSI, are systematically evaluated to study the impact of information availability on learning, coordination, and performance. The results show that structured partial CSI can approach full-CSI performance while reducing signaling overhead. Fairness-aware reward shaping further improves rate distribution, coordination stability, and EE when the fairness coefficient is properly selected.
Overall, this dissertation demonstrates that intelligent CRNs can be developed through a progressive framework that begins with spectrum awareness and extends toward decentralized learning-based resource allocation. The proposed approaches improve SE, EE, throughput, fairness, and constraint satisfaction compared with conventional optimization and learning-based baselines. The work provides a scalable foundation for future 5G, beyond-5G, and 6G networks requiring adaptive, distributed, and autonomous spectrum management under practical constraints.
Description
Keywords
multi-agent deep reinforcement learning, cognitive radio, wireless communication