A Q-Learning Based Spectrum Handoff Scheme with SINR-MOS over Cognitive Radio Networks

The spectrum handoff is an important technique for modern communication especially in the field of cognitive radio networks (CRNs). Reinforcement learning maximizes the rewards in interaction with the environment by learning strategies and can be used to realize the automatic spectrum handoff. A Q-learning based spectrum handoff scheme with signal to interference plus noise ratio-mean opinion score (SINR-MOS) is proposed in this paper. Firstly, two different MOSs based on SINR of each channel are designed. Secondly, the spectrum handoff scheme is realized by Q-learning, a tipical reinforcement learning method, with the SINR-MOSs as the reward functions. Finally, some simulation experiments are designed and the experimental results demonstrate that the proposed scheme has a good convergence performance.