An Improved Thompson Sampling Method for Dynamic Spectrum Access in Non-Stationary Environments

We consider the dynamic spectrum access problem in wireless sensor networks, where the unlicensed sensor nodes opportunistically access the licensed channels when they are not used by the licensed users. The sensor nodes try to access the channel with the highest probability of being idle to maximize their achievable data transmission. Due to the user mobility and the fluctuating traffic in the licensed network, the idle probabilities of the licensed channels vary with time. We propose a Thompson sampling based method assisted by a change detection technique to track the time-varying channel statistics. Numerical results show that our proposed method yields more channel access opportunities than the state-of-the-art ones.