Improving the performance of cognitive radios through classification, learning and predictive channel selection

Prediction of future idle times of different channels based on history information allows a cognitive radio (CR) to select the best channels for control and data transmission. In contrast to earlier work, the proposed method works not only with a specific type of traffic but learns and classifies the traffic type of each channel over time and can select the prediction method based on that. Different prediction rules apply to partially deterministic and stochastic ON-OFF patterns. New prediction methods for both traffic classes are developed in the paper. A CR predicts how long the channels are going to be idle. The channel with the longest predicted idle time is selected for secondary use. Simulations show that the proposed classification method works well and predictive channel selection method outperforms opportunistic random channel selection both with stochastic and deterministic ON-OFF patterns. Weibull, Pareto, and exponentially distributed traffic patterns are used in stochastic simulations to show general applicability of the proposed method. The classification-based method has even a higher gain when channels of interest include both stochastic and deterministic traffic. The collision rate with primary user over a given time interval can drop by more than 70% compared to the predictive system operating without classification.