“Soft decision” spectrum prediction based on back-propagation neural networks

In the cognitive radio system, spectrum prediction attracts more and more attention, which predicts future spectrum holes to save energy of spectrum sensing and to improve the efficiency of spectrum access. The current research on spectrum prediction is similar to the hard decision in the communication system. However, the hard decision loses amount of channel information during the process of obtaining channel statuses, which decreases the predictive accuracy of spectrum prediction. Therefore, we propose a “soft decision” model for spectrum prediction based on back-propagation neural networks. In the proposed model, the power values of frequency sampling point instead of the channel status are used as the inputs of the spectrum prediction model. Our experimental results demonstrate that the predictive accuracy of the proposed “soft decision” spectrum prediction model is better than the performance of conventional “hard decision”.