Single-state Q-learning for self-organised radio resource management in dual-hop 5G high capacity density networks

In this paper, a dual-hop wireless backhaul and small cell access network has been exploited with effective spectrum sharing, to provide 1 Gb/s/km ultra high capacity density for 5G ultra-dense network deployments. We develop a Single-State Qlearning (SSQL) based radio resource management algorithm for dynamic spectrum access creating a self-organized network. It intelligently utilizes the instantaneous spectrum observation information from spectrum sensing or a database, to learn longterm optimized decisions based on historical information of the system. The conventional Q learning algorithm with state-action pairs has been simplified to a stateless format and applied in a fully distributed manner on individual data file transmissions, which reduces the complexity of the learning model and improves the applicability of Q-learning algorithms to self-organized wireless networks. The results show that not only does the proposed algorithm completely remove the requirement for frequency planning, but also it improves the convergence, QoS and system capacity substantially by achieving higher link capacity on both access and backhaul networks. Keywords5G, Ultra-dense Network, Radio Resource Management, Dynamic Spectrum Access, Self-organized Network, Qlearning