Channel classification based fast spectrum sensing

In this paper, we propose a novel fast spectrum sensing scheme to reduce the huge consumption of wide band sensing. All potential channels are clustered into highly related groups based on the correlation among them. In each group, only one channel needs to be detected, while the states of other channels are estimated according to historical states and their correlation with the detected channel. Through detecting several channels, all the channel states can be obtained, thus consumption and sensing time are reduced. The influence of historical states on current state is modeled via Markov chain while the dependence of estimated channels (EC) on detecting channel (DC) is formulated by maximum a posteriori (MAP) principle. As Markov model and channel correlation may provide conflicting results, minimum entropy principle is adopted to unify the results of the two methods. Tested with real-world measurement data, our scheme is proved to improve sensing efficiency considerably under minimum loss in accuracy.