A spectrum prediction approach based on neural networks optimized by genetic algorithm in cognitive radio networks

Cognitive radio enabled dynamic spectrum access is a promising solution to alleviate the spectrum low-utilization problem. Secondary users need to sense the bands before transmitting on them to avoid collision with the licensed users. To reduce delay and energy consumption of spectrum sensing, spectrum prediction is incorporated to predict the future usage of channels before spectrum sensing. In this paper, a three-step ahead spectrum prediction framework is designed based on neutral network, which is optimized by generic algorithm to avoid the local optimization problem. As it is difficult to obtain the statistics of channel usage in CR networks, the neural network based on genetic algorithm model does not require a priori knowledge of the underlying distributions of the observed process. Simulation results show that the proposed scheme can predict spectrum usage effectively and significantly improve the prediction accuracy compared to traditional neural network based prediction algorithms.