A Multilayer Perceptron Neural Network-Based Spectrum Prediction Approach with Gray Decision

摘要

In cognitive radio networks (CRNs), spectrum prediction for inferring spectrum availability can help unlicensed users to discover spectrum holes earlier and to improve spectrum utilization more efficiently. Multilayer perceptron (MLP) neural network-based spectrum prediction model can identify the traffic characteristics of the spectrum only using the history data of the spectrum status. We investigate the statistic characteristics of the MLP neural network's outputs, propose the gray decision to improve the performance of the MLP-base spectrum predictor. We prove that the performance of MLP-base predictor with gray decision will be improved significantly when the spectrum status change frequently.