Classification-Based Predictive Channel Selection for Cognitive Radios

The proposed method classifies traffic patterns of primary channels in cognitive radio systems and applies different prediction rules to different types of traffic. This allows a more accurate prediction of the idle times of primary channels. An intelligent channel selection scheme then uses the prediction results to find the channels with the longest idle times for secondary use. We tested the method with Pareto and exponentially distributed stochastic traffic and with deterministic traffic. The predictive method using past information improves the throughput of the system compared to a system based on instantaneous idle time information. The classification-based predictive method improves the performance compared to pure prediction when the channels of interest include both stochastic and deterministic traffic. The amount of collisions with a primary user can drop 60 % within a given interval compared to a predictive system operating without classification.