Case-based reinforcement learning for cognitive spectrum assignment in cellular networks with dynamic topologies

Case-based reinforcement learning is a combination of reinforcement learning (RL) and case-based reasoning which has been successfully applied to a variety of artificial intelligence problems concerned with dynamic environments. This paper demonstrates how case-based RL can be applied to distributed dynamic spectrum assignment in cellular networks with dynamic topologies, and what performance improvements can be expected from using this approach in favour of a standard RL algorithm. Simulation results have shown that augmenting a stateless Q-learning algorithm with case-based reasoning functionality has significantly improved the temporal performance of a 9 base station network with dynamic topology. It has mitigated the performance degradation in terms of the probabilities of call blocking and dropping after transitions between different phases of the network topology, thus substantially increasing the usable range of traffic loads of the network.