A Heuristic-based Dynamic Spectrum Sharing Framework in Cognitive Radio Networks

Abstract

The inefficient use of the limited spectrum resources caused by the widely used static spectrum assignment policy has inspired the conception of the cognitive radio network (CRN). A CRN enables mobile users to optimally adapt their operating parameters according to the interactions with the surrounding radio environment. However, existing dynamic spectrum assignment approaches, a key component in CRN, suffer from their complexity, inflexibility, and lack of support for different classes of users.

In this thesis, we address the above issues by developing a heuristic-based dynamic spectrum sharing framework that can improve spectrum utilization efficiency and throughput in a CRN. Our framework consists of three parts. The first part addresses spectrum assignment that allocates new users to vacant spectrum bands without impeding existing users, for which a customized genetic algorithm is introduced. The second part focuses on the migration option where sometimes it might be beneficial for an existing user to move to a different spectrum band or disconnect to make room for a new user. This problem is solved by using a simulated annealing (SA) approach. In addition, we construct a migration cost model and study the impact of different traffic classes that are defined based on transmission rate requirements. The third part of our framework builds on the migration model and develops a round-based assignment approach that employs a prediction model to estimate when admitted users are likely to leave the network and uses the information to improve the migration decision. The prediction model relies on three newly constructed prediction methods to forecast user call duration, the fixation approach, the convergence approach, and the window-shopping approach, with each requiring less and less knowledge about user call preferences. Additionally, a price-based classification of user call characteristics is introduced.

Compared to other dynamic assignment approaches, our framework considers both overlay and underlay cases, offers a bounded response time and can work with different objective functions. It also considers user call characteristics and can accurately predict migration probability under various traffic conditions. Our simulation results show significant performance improvement in throughput and network utilization.