Spectrum behavior learning in Cognitive Radio based on artificial neural network

As an unlicensed wireless system, how to discover idle spectrum-bands efficiently and handover to minimize interferences to primary (licensed) users is the main focus for Cognitive Radio (CR). Therefore, the prerequisite for being “cognitive” lies in a deeper understanding of the characteristics of current spectrum behavior, such as a better model for spectrum behavior prediction, so as to design better and more efficient spectrum sensing and access schemes for secondary users in CR. A practical spectrum behavior learning method based on Multilayer Perception (MLP) artificial neural network (ANN) is introduced in this paper, by which, state of different channels in future timeslots (either idle or busy) can be forecasted through supervised learning such that CR nodes can create a handover channel list in advance without interruption to its ongoing transmission. Performances are evaluated with an existing 7-days spectrum data set from a previous measurement, which was conducted in a metro city located in south China. The empirical results show that the method proposed in this paper can well fit the future spectrum behavior with the mean Root Mean Square Error (RMSE) as low as 0.04. Besides, due to the generalization ability of ANN, the model generated from one radio service can be leveraged to predict spectrum behavior of another service with acceptable accuracy.