Reinforcement Learning for Resource Allocation in Cognitive Radio Networks

This chapter discusses the use of machine learning to perform distributed resource allocation in cognitive radio (CR) networks. There are many reinforcement learning techniques; one of the most common is Q-learning. The chapter explains the use of Q-learning for cross-layer resource allocations and describes resource allocation based on the deep Q-learning technique. It shows how different CRs can cooperate during the learning process. The chapter illustrates the performance of the table-based Q-learning algorithm for cross-layer resource allocation and the performance impact when implementing cooperative learning. The figures compare the results from simulations of three different systems: a system performing joint cross-layer CR adaptation, called individual learning; a system called docitive that also performs joint cross-layer CR adaptation but considers a secondary user joining the network that learns through the cross-layer docitive approach; and a system identified as physical layer only.