Support Vector Regression-based Spectrum Prediction Under Quick-Changing Channel Occupancy

Spectrum prediction is a promising technique to infer future spectrum state from historical spectrum data by exploiting the correlation of spectrum states in time domain. To realize reliable prediction of spectrum state under quick-changing channel occupancy, this article proposes a support vector regression (SVR)-based spectrum prediction algorithm. By adopting three typical distribution models of spectrum state evolution, this article compares the prediction accuracy performance of the proposed algorithm and the existing back propagation (BP) neural network-based spectrum prediction algorithm. Simulation results show that the proposed algorithm has the capability of small sample learning and obtains better prediction accuracy with less training samples. On the basis of the previous study, research on the imperfect training spectrum sample which considering the probability of detection and the probability of false alarm are analyzed. The analysis shows that the prediction accuracy performance of the SVR-based spectrum prediction algorithm still achieves satisfactory result with the imperfect training samples.