Self Organising Cognitive Radio Networks

This thesis investigates the application of learning and cognition to self organisation of ad hoc and green small cell networks in order to improve performance in terms of throughput, delay and the network energy consumption to achieve ‘green communication’. Initially, an attempt is made to improve the spatial re-use of the network by dividing it into disjoint sets of nodes through a clustering process. A novel distributed clustering algorithm is developed that exploit cognitive radio based principles in that they have the ability to learn from received signal strength indicator (RSSI) beacons, to form clusters which reduce the average distance between nodes, as well as reducing the level of overlap between clusters. By making nodes repeatedly learn about their environment through RSSI, nodes effectively compete to become a cluster head, with the winning nodes being those that are located in an area of locally high node density. It is demonstrated that the resulting cluster formation through repeated learning is better than with no learning and node degree. The benefit of applying a hierarchical architecture to ad hoc and green small cell networks via two-hop backhauling is examined with respect to its energy efficiency. Energy efficiency is investigated in terms of the energy consumption ratio (ECR) and the energy reduction gain (ERG). The results are compared to that of a traditional single hop architecture with no hierarchical formation. It is shown that under certain conditions, dual hop clustered networks can potentially be more energy efficient that single hop transmission, but care needs to be taken to ensure that the backhaul links within the network do not become bottlenecks at high offered traffic levels. The application of directional antennas at a Hub Base Station significantly helps to reduce the total energy consumption of the network as well the backhaul connectivity of a dual-hop clustered network. Introducing Reinforcement Learning to channel assignment on the first hop reduces end to end delay and thus minimises the amount of time and energy for the nodes in the network to be in transmission or reception mode. The reinforcement learning schemes can exploit the spectrum in which it perceives as a good option based upon individual channel historical information and thereby further improve the network spatial re-use.