A Reliable Time-Domain Spectrum Hole Prediction for Cognitive Radio Networks Using Regularized Multi-Layer Perceptron

Abstract

A novel spectrum prediction technique based on multi-layer perceptron is proposed to effectively identify spectrum holes in time domain for cognitive radio networks (CRNs). This scheme adopts a more comprehensive input space (e.g. traffic parameters of primary network) to reduce the sampling bias resulted from simple binary input space (e.g. status of spectrum holes) which is commonly used in the conventional spectrum hole prediction schemes. Additionally, regularization is proposed to mitigate the impact of the noise introduced by the stochastic CRNs. The simulation results show that a more reliable spectrum hole predictor can be obtained if being trained using our proposed novel input space and regularization mechanism.