Abstract
The increasing demands for wireless spectrum and limited radio resources emphasise the need for more efficient spectrum sharing mechanisms. Dynamic spectrum sharing (DSS) has been cited as a promising mechanism for managing the radio spectrum. The objectives are to achieve flexible spectrum usage, improve spectrum efficiency, and combat spectrum scarcity problem, which constitute the main motivations setting out the scope of this thesis. This thesis, first, presents a comprehensive survey of the general principles of current and state of art spectrum sharing practices, and approaches to future ones. A concise cell-coupling-based approach has been introduced as a model for the investigation of the influence when several competing mobile networks simultaneously coexist and share a common pool of radio resources. A simple centralised medium-term DSS (CDSS) model has been employed in order to enhance spectrum utilisation and achieve an interference-free communication environment. The global view of the CDSS leads to a quick path in reaching a stable system and achieving a near optimum spectrum configuration. To align with the current trends, the thesis also investigates enabling technology for DSS in heterogeneous networks (HetNets) deployment, where factors such as ad hoc and distributed nature of network topology, heterogeneous network infrastructures, and several power profiles make the spectrum and interference management problem more intractable. The HetNet is modelled as a layer of closed access, randomly-located fem-tocells (FCs) overlaid upon a LTE radio access mobile network. In the context of dynamic learning games, this work proposes multi-objectives, fully distributed strategy based on heterogeneous reinforcement learning (RL) model (CODIPAS-HRL) for the femtocells opportunistic access. We present two different learning strategies; the modified Bush and Mosteller (MBM) and the modified Roth-Erev (MRE). The selforganisation capability enables FCs to autonomously identify available spectrum for opportunistic use, using HRL schemes, and tune their parameters accordingly in order to operate under restrictions of avoiding interference and satisfy QoS requirements. Finally, the proposed work takes the advantages of calculating the learning cost, the convergence behaviour for different learning rates and provides comparisons between different learning strategies. The simulation results show the convergence of the learning model to a solution concept based on satisfaction equilibrium, under the uncertainty of the HetNet environment and heterogeneous learning. Such a distributed intelligent scheme can provide a practical solution to the main challenges of spectral opportunity identification and interference management in future networks.