Cognitive Radio with Reinforcement Learning Applied to Multicast Downlink Transmission and Distributed Occupancy Detection

This paper shows how channel assignment in multicast terrestrial communication systems with different user populations and distributed channel occupancy detection can be improved using intelligence based on reinforcement learning. The schemes greatly reduce the number of reassignments and improve the dropping probability, at the expense of increased blocking. It is found that compared to detection by single users, detection by multiple users reduces the 'hidden node' problem. Using different minimum quality of service threshold percentages can partly control and improve the performance, in place of the more traditional SINR threshold levels. At the same time, with reinforcement leaning, the ability of find an optimal channel for users is significantly improved, because the channel weighting can help the users avoid the interference.