A Spectrum Prediction Technique Based on Convolutional Neural Networks

Secondary users in cognitive radio system use spectrum sensing technology to detect the primary users in the frequency band and use spectrum holes to communicate. Spectrum prediction technology is based on the existing spectrum sensing results to predict the future channel occupancy, so as to reduce the blocking rate, avoid malicious dynamic interference and other purposes. In this paper, a spectrum prediction method based on convolution neural network is proposed and some applications of this method in practical communication systems are given. This method can be trained in real time and has a certain adaptability to the dynamic environment. Using this method, the predicted results can be used to allocate resources reasonably, and the spectrum resource utilization rate is high. In addition, the time-consuming of broadband spectrum sensing can be shortened by combining the spectrum prediction method based on convolution neural network. At the end of this paper, the simulation results of spectrum prediction method based on convolution neural network are given and the efficiency of the algorithm is discussed.