Learning-Aided Markov Chain Monte Carlo Scheme for Spectrum Sensing in Cognitive Radio

In this paper, a learning-aided stochastic strategy is studied for the non-cooperative spectrum sensing in cognitive radio (CR) networks. By sampling from the channel availability distribution in a Markov chain Monte Carlo (MCMC) way, the proposed learning-aided Metropolis-Hastings (LMH) algorithm generates the target channel sequence for fine sensing. The flexible proposal distribution in MH sampling is fully exploited, and a learning mechanism based on the multiple sampling stages within one Markov move is proposed, which tries to take advantages of the samples obtained by the previous stage. The reversibility of the Markov chain in the proposed LMH sampling is studied in detail while its faster convergence rate in the Markov mixing is demonstrated as well, which leads to better spectrum sensing performance and efficiency in cognitive radio.